diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml index b482fddb..c1cd28bc 100644 --- a/.github/workflows/ci.yaml +++ b/.github/workflows/ci.yaml @@ -82,4 +82,4 @@ jobs: run: pip install uv && uv pip install --system ".[dev]" - name: Run tests - run: nbdev_test --timing --do_print --n_workers 0 --skip_file_re "computing_at_scale|distributed" + run: nbdev_test --timing --do_print --n_workers 0 --skip_file_re "computing_at_scale|distributed" \ No newline at end of file diff --git a/nbs/assets/arima_rst.csv b/nbs/assets/arima_rst.csv new file mode 100644 index 00000000..564a3405 --- /dev/null +++ b/nbs/assets/arima_rst.csv @@ -0,0 +1,197 @@ +,unique_id,ds,AutoARIMA +0,FOODS_1,2016-04-25,2567.6380220997385 +1,FOODS_1,2016-04-26,2640.0126507600617 +2,FOODS_1,2016-04-27,2654.020161270193 +3,FOODS_1,2016-04-28,2796.440947432241 +4,FOODS_1,2016-04-29,3208.3811781219893 +5,FOODS_1,2016-04-30,3773.861663861278 +6,FOODS_1,2016-05-01,3299.79743620078 +7,FOODS_1,2016-05-02,2640.917833365884 +8,FOODS_1,2016-05-03,2694.336636469822 +9,FOODS_1,2016-05-04,2694.1035608286916 +10,FOODS_1,2016-05-05,2835.3163066249176 +11,FOODS_1,2016-05-06,3236.293019356992 +12,FOODS_1,2016-05-07,3797.241666185409 +13,FOODS_1,2016-05-08,3321.3663551719346 +14,FOODS_1,2016-05-09,2657.5206296593224 +15,FOODS_1,2016-05-10,2707.524732902354 +16,FOODS_1,2016-05-11,2704.747898679939 +17,FOODS_1,2016-05-12,2843.9075334840127 +18,FOODS_1,2016-05-13,3243.2271452721875 +19,FOODS_1,2016-05-14,3802.8383176295374 +20,FOODS_1,2016-05-15,3325.8835081296597 +21,FOODS_1,2016-05-16,2661.1665009534677 +22,FOODS_1,2016-05-17,2710.4673778864567 +23,FOODS_1,2016-05-18,2707.1229576998144 +24,FOODS_1,2016-05-19,2845.8244841791743 +25,FOODS_1,2016-05-20,3244.7743489150994 +26,FOODS_1,2016-05-21,3804.0870921121996 +27,FOODS_1,2016-05-22,3326.8914153406417 +28,FOODS_2,2016-04-25,5247.139313835306 +29,FOODS_2,2016-04-26,4792.960953564678 +30,FOODS_2,2016-04-27,4590.964924105041 +31,FOODS_2,2016-04-28,4600.860960731239 +32,FOODS_2,2016-04-29,4942.967875785429 +33,FOODS_2,2016-04-30,6337.5344008647035 +34,FOODS_2,2016-05-01,6757.903656868389 +35,FOODS_2,2016-05-02,5607.448278542931 +36,FOODS_2,2016-05-03,5244.736379588703 +37,FOODS_2,2016-05-04,5004.324139775963 +38,FOODS_2,2016-05-05,4898.957821332703 +39,FOODS_2,2016-05-06,5310.50858073175 +40,FOODS_2,2016-05-07,6641.501751486881 +41,FOODS_2,2016-05-08,6994.60266694966 +42,FOODS_2,2016-05-09,5779.319357423172 +43,FOODS_2,2016-05-10,5407.162676474501 +44,FOODS_2,2016-05-11,5083.913842178118 +45,FOODS_2,2016-05-12,4944.398003219373 +46,FOODS_2,2016-05-13,5311.376375871703 +47,FOODS_2,2016-05-14,6593.071399563353 +48,FOODS_2,2016-05-15,6919.214380455565 +49,FOODS_2,2016-05-16,5687.59113569716 +50,FOODS_2,2016-05-17,5289.695188706264 +51,FOODS_2,2016-05-18,4961.060422560739 +52,FOODS_2,2016-05-19,4815.1162421391555 +53,FOODS_2,2016-05-20,5184.318070407874 +54,FOODS_2,2016-05-21,6467.664979048176 +55,FOODS_2,2016-05-22,6807.52592255971 +56,FOODS_3,2016-04-25,17666.731556718732 +57,FOODS_3,2016-04-26,16540.541376421595 +58,FOODS_3,2016-04-27,16160.633071055223 +59,FOODS_3,2016-04-28,16370.683658640599 +60,FOODS_3,2016-04-29,17889.869101221244 +61,FOODS_3,2016-04-30,22332.436097780206 +62,FOODS_3,2016-05-01,22713.044590532798 +63,FOODS_3,2016-05-02,17468.426315986457 +64,FOODS_3,2016-05-03,16293.254187261948 +65,FOODS_3,2016-05-04,15975.967829497 +66,FOODS_3,2016-05-05,16301.448695311468 +67,FOODS_3,2016-05-06,17742.192012594733 +68,FOODS_3,2016-05-07,22149.920466723906 +69,FOODS_3,2016-05-08,22672.51910973226 +70,FOODS_3,2016-05-09,17388.755964833897 +71,FOODS_3,2016-05-10,16229.554742510712 +72,FOODS_3,2016-05-11,15913.853788834986 +73,FOODS_3,2016-05-12,16229.441847880662 +74,FOODS_3,2016-05-13,17695.119173018684 +75,FOODS_3,2016-05-14,22118.38914081269 +76,FOODS_3,2016-05-15,22622.383902927766 +77,FOODS_3,2016-05-16,17352.915739873653 +78,FOODS_3,2016-05-17,16196.863301084244 +79,FOODS_3,2016-05-18,15886.258449760904 +80,FOODS_3,2016-05-19,16208.436163712715 +81,FOODS_3,2016-05-20,17673.61997937043 +82,FOODS_3,2016-05-21,22097.648256558292 +83,FOODS_3,2016-05-22,22608.340998500338 +84,HOBBIES_1,2016-04-25,3301.1184614493914 +85,HOBBIES_1,2016-04-26,3190.0305707575 +86,HOBBIES_1,2016-04-27,3199.935172884026 +87,HOBBIES_1,2016-04-28,3225.695752897986 +88,HOBBIES_1,2016-04-29,3567.265955094044 +89,HOBBIES_1,2016-04-30,4356.224282294094 +90,HOBBIES_1,2016-05-01,4089.925888720474 +91,HOBBIES_1,2016-05-02,3256.234765934616 +92,HOBBIES_1,2016-05-03,3167.254279226249 +93,HOBBIES_1,2016-05-04,3182.164242104323 +94,HOBBIES_1,2016-05-05,3197.8208347975133 +95,HOBBIES_1,2016-05-06,3562.1759714192 +96,HOBBIES_1,2016-05-07,4348.942474784928 +97,HOBBIES_1,2016-05-08,4062.088166835995 +98,HOBBIES_1,2016-05-09,3270.411197190324 +99,HOBBIES_1,2016-05-10,3153.9826259751076 +100,HOBBIES_1,2016-05-11,3185.763508441231 +101,HOBBIES_1,2016-05-12,3215.5124943559786 +102,HOBBIES_1,2016-05-13,3557.0053751276196 +103,HOBBIES_1,2016-05-14,4347.883194155448 +104,HOBBIES_1,2016-05-15,4072.703812696276 +105,HOBBIES_1,2016-05-16,3257.5290353634405 +106,HOBBIES_1,2016-05-17,3160.8541642155105 +107,HOBBIES_1,2016-05-18,3181.513304571057 +108,HOBBIES_1,2016-05-19,3202.9504121022114 +109,HOBBIES_1,2016-05-20,3559.2179367013387 +110,HOBBIES_1,2016-05-21,4348.099858357742 +111,HOBBIES_1,2016-05-22,4064.243078853568 +112,HOBBIES_2,2016-04-25,372.73314373276685 +113,HOBBIES_2,2016-04-26,394.9539049862272 +114,HOBBIES_2,2016-04-27,383.16517087090995 +115,HOBBIES_2,2016-04-28,383.62499302365416 +116,HOBBIES_2,2016-04-29,398.311355639119 +117,HOBBIES_2,2016-04-30,394.347605737723 +118,HOBBIES_2,2016-05-01,410.1164983512131 +119,HOBBIES_2,2016-05-02,381.5187849900045 +120,HOBBIES_2,2016-05-03,390.5415783971641 +121,HOBBIES_2,2016-05-04,384.30591664608176 +122,HOBBIES_2,2016-05-05,382.0656727697983 +123,HOBBIES_2,2016-05-06,391.0787416473706 +124,HOBBIES_2,2016-05-07,388.7836691251789 +125,HOBBIES_2,2016-05-08,393.876028907937 +126,HOBBIES_2,2016-05-09,390.2092112439395 +127,HOBBIES_2,2016-05-10,389.2354081364607 +128,HOBBIES_2,2016-05-11,388.6840382471152 +129,HOBBIES_2,2016-05-12,388.2380749588003 +130,HOBBIES_2,2016-05-13,387.8486951350242 +131,HOBBIES_2,2016-05-14,387.50441758710645 +132,HOBBIES_2,2016-05-15,387.19942014913687 +133,HOBBIES_2,2016-05-16,386.9291391176742 +134,HOBBIES_2,2016-05-17,386.6896116476871 +135,HOBBIES_2,2016-05-18,386.4773369473272 +136,HOBBIES_2,2016-05-19,386.28921406142393 +137,HOBBIES_2,2016-05-20,386.12249505879066 +138,HOBBIES_2,2016-05-21,385.9747446856811 +139,HOBBIES_2,2016-05-22,385.84380476581924 +140,HOUSEHOLD_1,2016-04-25,7401.891643264893 +141,HOUSEHOLD_1,2016-04-26,6713.059783961793 +142,HOUSEHOLD_1,2016-04-27,6568.574826414228 +143,HOUSEHOLD_1,2016-04-28,6780.766120981868 +144,HOUSEHOLD_1,2016-04-29,7572.471090810987 +145,HOUSEHOLD_1,2016-04-30,9760.255572651784 +146,HOUSEHOLD_1,2016-05-01,9626.516172187357 +147,HOUSEHOLD_1,2016-05-02,7339.549091656914 +148,HOUSEHOLD_1,2016-05-03,6715.501042329869 +149,HOUSEHOLD_1,2016-05-04,6588.688826450386 +150,HOUSEHOLD_1,2016-05-05,6792.018975120889 +151,HOUSEHOLD_1,2016-05-06,7585.826196179538 +152,HOUSEHOLD_1,2016-05-07,9784.576440586516 +153,HOUSEHOLD_1,2016-05-08,9625.229656671552 +154,HOUSEHOLD_1,2016-05-09,7370.017398852801 +155,HOUSEHOLD_1,2016-05-10,6703.274498285182 +156,HOUSEHOLD_1,2016-05-11,6599.658566441017 +157,HOUSEHOLD_1,2016-05-12,6803.892952779315 +158,HOUSEHOLD_1,2016-05-13,7578.147413631544 +159,HOUSEHOLD_1,2016-05-14,9797.63840894531 +160,HOUSEHOLD_1,2016-05-15,9618.004359773748 +161,HOUSEHOLD_1,2016-05-16,7369.325267410535 +162,HOUSEHOLD_1,2016-05-17,6704.177617699504 +163,HOUSEHOLD_1,2016-05-18,6602.321156728928 +164,HOUSEHOLD_1,2016-05-19,6806.125635799577 +165,HOUSEHOLD_1,2016-05-20,7579.715540310798 +166,HOUSEHOLD_1,2016-05-21,9800.535564906138 +167,HOUSEHOLD_1,2016-05-22,9618.816928719863 +168,HOUSEHOLD_2,2016-04-25,1944.6365790975688 +169,HOUSEHOLD_2,2016-04-26,1782.1136284351692 +170,HOUSEHOLD_2,2016-04-27,1783.3160335243492 +171,HOUSEHOLD_2,2016-04-28,1843.7479525740052 +172,HOUSEHOLD_2,2016-04-29,2017.7306193648328 +173,HOUSEHOLD_2,2016-04-30,2573.4319970333827 +174,HOUSEHOLD_2,2016-05-01,2555.781560345023 +175,HOUSEHOLD_2,2016-05-02,1911.5887343314832 +176,HOUSEHOLD_2,2016-05-03,1773.6009268551322 +177,HOUSEHOLD_2,2016-05-04,1783.737933427243 +178,HOUSEHOLD_2,2016-05-05,1836.4433591054317 +179,HOUSEHOLD_2,2016-05-06,2026.2345730343163 +180,HOUSEHOLD_2,2016-05-07,2599.273470339867 +181,HOUSEHOLD_2,2016-05-08,2547.324959977864 +182,HOUSEHOLD_2,2016-05-09,1919.6860442557204 +183,HOUSEHOLD_2,2016-05-10,1772.1267493740079 +184,HOUSEHOLD_2,2016-05-11,1776.2903338158771 +185,HOUSEHOLD_2,2016-05-12,1836.0676051839982 +186,HOUSEHOLD_2,2016-05-13,2012.270429155826 +187,HOUSEHOLD_2,2016-05-14,2570.3606621455087 +188,HOUSEHOLD_2,2016-05-15,2548.3296441260773 +189,HOUSEHOLD_2,2016-05-16,1906.4143149755678 +190,HOUSEHOLD_2,2016-05-17,1767.280129446266 +191,HOUSEHOLD_2,2016-05-18,1776.7401745381599 +192,HOUSEHOLD_2,2016-05-19,1830.4754939798 +193,HOUSEHOLD_2,2016-05-20,2018.588737506113 +194,HOUSEHOLD_2,2016-05-21,2589.769743131061 +195,HOUSEHOLD_2,2016-05-22,2541.838195247957 diff --git a/nbs/assets/lgbm_rst.csv b/nbs/assets/lgbm_rst.csv new file mode 100644 index 00000000..144ca90c --- /dev/null +++ b/nbs/assets/lgbm_rst.csv @@ -0,0 +1,197 @@ +,unique_id,ds,AutoLightGBM +0,FOODS_1,2016-04-25,2694.6738511439944 +1,FOODS_1,2016-04-26,2805.230420439614 +2,FOODS_1,2016-04-27,2662.0380598102083 +3,FOODS_1,2016-04-28,2645.027003376193 +4,FOODS_1,2016-04-29,3236.588718297946 +5,FOODS_1,2016-04-30,3747.51371444864 +6,FOODS_1,2016-05-01,3172.307149809725 +7,FOODS_1,2016-05-02,2657.3513449500338 +8,FOODS_1,2016-05-03,2691.9787279391626 +9,FOODS_1,2016-05-04,2746.2863197867673 +10,FOODS_1,2016-05-05,2762.255340569623 +11,FOODS_1,2016-05-06,3257.194144588295 +12,FOODS_1,2016-05-07,3799.031472854631 +13,FOODS_1,2016-05-08,3252.8382373569984 +14,FOODS_1,2016-05-09,2786.334069555473 +15,FOODS_1,2016-05-10,2820.205574835855 +16,FOODS_1,2016-05-11,2762.5056712532564 +17,FOODS_1,2016-05-12,2786.88595237666 +18,FOODS_1,2016-05-13,3309.4905882795897 +19,FOODS_1,2016-05-14,3796.0201508972023 +20,FOODS_1,2016-05-15,3273.3752810082988 +21,FOODS_1,2016-05-16,2742.19182422212 +22,FOODS_1,2016-05-17,2823.468902301943 +23,FOODS_1,2016-05-18,2854.8557474617855 +24,FOODS_1,2016-05-19,2871.135690726362 +25,FOODS_1,2016-05-20,3331.1931287348802 +26,FOODS_1,2016-05-21,3781.1473608863216 +27,FOODS_1,2016-05-22,3405.544149448268 +28,FOODS_2,2016-04-25,5120.102523686289 +29,FOODS_2,2016-04-26,5185.27038248254 +30,FOODS_2,2016-04-27,5115.111348340718 +31,FOODS_2,2016-04-28,4619.394726238464 +32,FOODS_2,2016-04-29,5432.537671232106 +33,FOODS_2,2016-04-30,6025.41273018486 +34,FOODS_2,2016-05-01,6220.986523247984 +35,FOODS_2,2016-05-02,4964.5749171463885 +36,FOODS_2,2016-05-03,5103.670724855448 +37,FOODS_2,2016-05-04,5144.3979512523165 +38,FOODS_2,2016-05-05,5218.996763184711 +39,FOODS_2,2016-05-06,5439.347347215497 +40,FOODS_2,2016-05-07,5865.1687497733365 +41,FOODS_2,2016-05-08,6368.92258946046 +42,FOODS_2,2016-05-09,5186.848082254273 +43,FOODS_2,2016-05-10,4880.374441261076 +44,FOODS_2,2016-05-11,4909.098152984957 +45,FOODS_2,2016-05-12,5081.108905626409 +46,FOODS_2,2016-05-13,5407.15017000438 +47,FOODS_2,2016-05-14,5830.236827339688 +48,FOODS_2,2016-05-15,6618.276011160659 +49,FOODS_2,2016-05-16,5242.077035682606 +50,FOODS_2,2016-05-17,5003.983891141501 +51,FOODS_2,2016-05-18,4958.840167821236 +52,FOODS_2,2016-05-19,4762.949855746844 +53,FOODS_2,2016-05-20,5389.750344472496 +54,FOODS_2,2016-05-21,5942.651453698519 +55,FOODS_2,2016-05-22,6269.243619511553 +56,FOODS_3,2016-04-25,17711.385739545563 +57,FOODS_3,2016-04-26,16357.865618151673 +58,FOODS_3,2016-04-27,16557.86563355973 +59,FOODS_3,2016-04-28,16163.752858976526 +60,FOODS_3,2016-04-29,18881.115214451296 +61,FOODS_3,2016-04-30,23092.123636884276 +62,FOODS_3,2016-05-01,23294.58207237022 +63,FOODS_3,2016-05-02,17732.312728761386 +64,FOODS_3,2016-05-03,16215.758103512571 +65,FOODS_3,2016-05-04,16609.186712046285 +66,FOODS_3,2016-05-05,17115.71866090232 +67,FOODS_3,2016-05-06,18163.574602023327 +68,FOODS_3,2016-05-07,22698.015289239014 +69,FOODS_3,2016-05-08,22955.84590815469 +70,FOODS_3,2016-05-09,17924.510549950868 +71,FOODS_3,2016-05-10,16652.368043668524 +72,FOODS_3,2016-05-11,16706.713769910348 +73,FOODS_3,2016-05-12,17070.997700996042 +74,FOODS_3,2016-05-13,18438.390974654023 +75,FOODS_3,2016-05-14,22698.015289239014 +76,FOODS_3,2016-05-15,22955.84590815469 +77,FOODS_3,2016-05-16,18333.211272508892 +78,FOODS_3,2016-05-17,16812.741370928095 +79,FOODS_3,2016-05-18,16706.713769910348 +80,FOODS_3,2016-05-19,16479.679299243857 +81,FOODS_3,2016-05-20,18502.532285923797 +82,FOODS_3,2016-05-21,22698.015289239014 +83,FOODS_3,2016-05-22,22955.84590815469 +84,HOBBIES_1,2016-04-25,3226.027840439743 +85,HOBBIES_1,2016-04-26,3195.3648189304345 +86,HOBBIES_1,2016-04-27,3172.479599244583 +87,HOBBIES_1,2016-04-28,3215.297092249357 +88,HOBBIES_1,2016-04-29,3550.478555688749 +89,HOBBIES_1,2016-04-30,4315.818469198181 +90,HOBBIES_1,2016-05-01,4321.600390227421 +91,HOBBIES_1,2016-05-02,3157.8276654934016 +92,HOBBIES_1,2016-05-03,3208.8238007824716 +93,HOBBIES_1,2016-05-04,3171.402461945512 +94,HOBBIES_1,2016-05-05,3122.93302254743 +95,HOBBIES_1,2016-05-06,3534.892640584684 +96,HOBBIES_1,2016-05-07,4261.123354991392 +97,HOBBIES_1,2016-05-08,4296.069679132532 +98,HOBBIES_1,2016-05-09,3273.3972145241705 +99,HOBBIES_1,2016-05-10,3221.5792680507666 +100,HOBBIES_1,2016-05-11,3200.345723427872 +101,HOBBIES_1,2016-05-12,3245.4991334325614 +102,HOBBIES_1,2016-05-13,3568.3027303655 +103,HOBBIES_1,2016-05-14,4320.567307842713 +104,HOBBIES_1,2016-05-15,4343.116200412291 +105,HOBBIES_1,2016-05-16,3289.762766694878 +106,HOBBIES_1,2016-05-17,3267.984449690506 +107,HOBBIES_1,2016-05-18,3200.345723427872 +108,HOBBIES_1,2016-05-19,3208.1360951146366 +109,HOBBIES_1,2016-05-20,3615.44813501723 +110,HOBBIES_1,2016-05-21,4326.047999208027 +111,HOBBIES_1,2016-05-22,4344.900175442336 +112,HOBBIES_2,2016-04-25,339.77407251192693 +113,HOBBIES_2,2016-04-26,397.4010613540528 +114,HOBBIES_2,2016-04-27,382.61025056378617 +115,HOBBIES_2,2016-04-28,393.55441552694975 +116,HOBBIES_2,2016-04-29,412.81304016665115 +117,HOBBIES_2,2016-04-30,416.1985117603962 +118,HOBBIES_2,2016-05-01,467.8505902120062 +119,HOBBIES_2,2016-05-02,359.72499730411994 +120,HOBBIES_2,2016-05-03,400.9547760594411 +121,HOBBIES_2,2016-05-04,384.0621884160096 +122,HOBBIES_2,2016-05-05,384.0621884160096 +123,HOBBIES_2,2016-05-06,412.81304016665115 +124,HOBBIES_2,2016-05-07,405.2543467972324 +125,HOBBIES_2,2016-05-08,467.8505902120062 +126,HOBBIES_2,2016-05-09,380.6591071937099 +127,HOBBIES_2,2016-05-10,400.9547760594411 +128,HOBBIES_2,2016-05-11,400.9547760594411 +129,HOBBIES_2,2016-05-12,400.9547760594411 +130,HOBBIES_2,2016-05-13,412.81304016665115 +131,HOBBIES_2,2016-05-14,416.1985117603962 +132,HOBBIES_2,2016-05-15,467.8505902120062 +133,HOBBIES_2,2016-05-16,390.6243107525114 +134,HOBBIES_2,2016-05-17,411.89894102260484 +135,HOBBIES_2,2016-05-18,400.9547760594411 +136,HOBBIES_2,2016-05-19,400.9547760594411 +137,HOBBIES_2,2016-05-20,412.81304016665115 +138,HOBBIES_2,2016-05-21,416.1985117603962 +139,HOBBIES_2,2016-05-22,467.8505902120062 +140,HOUSEHOLD_1,2016-04-25,7160.832175743806 +141,HOUSEHOLD_1,2016-04-26,6971.744965808053 +142,HOUSEHOLD_1,2016-04-27,6676.608231213473 +143,HOUSEHOLD_1,2016-04-28,6640.7455196004885 +144,HOUSEHOLD_1,2016-04-29,7565.152447398786 +145,HOUSEHOLD_1,2016-04-30,9465.935601433835 +146,HOUSEHOLD_1,2016-05-01,9702.66861697589 +147,HOUSEHOLD_1,2016-05-02,7205.825513464553 +148,HOUSEHOLD_1,2016-05-03,7113.679030081171 +149,HOUSEHOLD_1,2016-05-04,6649.31399566832 +150,HOUSEHOLD_1,2016-05-05,6674.348708680196 +151,HOUSEHOLD_1,2016-05-06,7697.496157413757 +152,HOUSEHOLD_1,2016-05-07,9399.00971175823 +153,HOUSEHOLD_1,2016-05-08,9439.07736476184 +154,HOUSEHOLD_1,2016-05-09,7493.79860897777 +155,HOUSEHOLD_1,2016-05-10,7002.1844984992185 +156,HOUSEHOLD_1,2016-05-11,6661.775405589411 +157,HOUSEHOLD_1,2016-05-12,6764.480000367033 +158,HOUSEHOLD_1,2016-05-13,7598.947495707824 +159,HOUSEHOLD_1,2016-05-14,9399.00971175823 +160,HOUSEHOLD_1,2016-05-15,9439.07736476184 +161,HOUSEHOLD_1,2016-05-16,7534.459703980786 +162,HOUSEHOLD_1,2016-05-17,7229.353310975897 +163,HOUSEHOLD_1,2016-05-18,6699.882771137218 +164,HOUSEHOLD_1,2016-05-19,6764.480000367033 +165,HOUSEHOLD_1,2016-05-20,7697.496157413757 +166,HOUSEHOLD_1,2016-05-21,9399.00971175823 +167,HOUSEHOLD_1,2016-05-22,9314.773035693921 +168,HOUSEHOLD_2,2016-04-25,1891.769292418806 +169,HOUSEHOLD_2,2016-04-26,1805.813605315031 +170,HOUSEHOLD_2,2016-04-27,1758.0613497523343 +171,HOUSEHOLD_2,2016-04-28,1800.1115326528136 +172,HOUSEHOLD_2,2016-04-29,1959.4740469519804 +173,HOUSEHOLD_2,2016-04-30,2660.449142902388 +174,HOUSEHOLD_2,2016-05-01,2619.654508404448 +175,HOUSEHOLD_2,2016-05-02,1839.0080089551213 +176,HOUSEHOLD_2,2016-05-03,1829.4970368277027 +177,HOUSEHOLD_2,2016-05-04,1823.5007626046254 +178,HOUSEHOLD_2,2016-05-05,1811.520891959066 +179,HOUSEHOLD_2,2016-05-06,1977.7038269928182 +180,HOUSEHOLD_2,2016-05-07,2816.242000429492 +181,HOUSEHOLD_2,2016-05-08,2634.914005288675 +182,HOUSEHOLD_2,2016-05-09,1878.4366945999661 +183,HOUSEHOLD_2,2016-05-10,1815.8433855980848 +184,HOUSEHOLD_2,2016-05-11,1811.1689759986775 +185,HOUSEHOLD_2,2016-05-12,1811.520891959066 +186,HOUSEHOLD_2,2016-05-13,1977.7038269928182 +187,HOUSEHOLD_2,2016-05-14,2753.4834680201943 +188,HOUSEHOLD_2,2016-05-15,2586.938159190155 +189,HOUSEHOLD_2,2016-05-16,1857.353491565642 +190,HOUSEHOLD_2,2016-05-17,1829.4970368277027 +191,HOUSEHOLD_2,2016-05-18,1811.520891959066 +192,HOUSEHOLD_2,2016-05-19,1811.520891959066 +193,HOUSEHOLD_2,2016-05-20,1977.7038269928182 +194,HOUSEHOLD_2,2016-05-21,2861.6326484548927 +195,HOUSEHOLD_2,2016-05-22,2645.9665234893946 diff --git a/nbs/assets/nhits_rst.csv b/nbs/assets/nhits_rst.csv new file mode 100644 index 00000000..99e3f9d1 --- /dev/null +++ b/nbs/assets/nhits_rst.csv @@ -0,0 +1,197 @@ +unique_id,ds,NHITS +FOODS_1,2016-04-25,2504.762 +FOODS_1,2016-04-26,2467.719 +FOODS_1,2016-04-27,2430.538 +FOODS_1,2016-04-28,2545.7861 +FOODS_1,2016-04-29,3158.0547 +FOODS_1,2016-04-30,3675.847 +FOODS_1,2016-05-01,3220.2334 +FOODS_1,2016-05-02,2603.3481 +FOODS_1,2016-05-03,2549.2483 +FOODS_1,2016-05-04,2526.7446 +FOODS_1,2016-05-05,2656.253 +FOODS_1,2016-05-06,3305.013 +FOODS_1,2016-05-07,3765.003 +FOODS_1,2016-05-08,3275.3057 +FOODS_1,2016-05-09,2697.433 +FOODS_1,2016-05-10,2621.8018 +FOODS_1,2016-05-11,2528.0818 +FOODS_1,2016-05-12,2711.9429 +FOODS_1,2016-05-13,3265.7722 +FOODS_1,2016-05-14,3737.7068 +FOODS_1,2016-05-15,3250.004 +FOODS_1,2016-05-16,2663.8232 +FOODS_1,2016-05-17,2565.505 +FOODS_1,2016-05-18,2516.273 +FOODS_1,2016-05-19,2611.7502 +FOODS_1,2016-05-20,3218.6646 +FOODS_1,2016-05-21,3715.1863 +FOODS_1,2016-05-22,3209.3425 +FOODS_2,2016-04-25,5217.8647 +FOODS_2,2016-04-26,4537.3613 +FOODS_2,2016-04-27,4325.717 +FOODS_2,2016-04-28,4522.6455 +FOODS_2,2016-04-29,5086.342 +FOODS_2,2016-04-30,6492.3867 +FOODS_2,2016-05-01,7200.9097 +FOODS_2,2016-05-02,6160.0244 +FOODS_2,2016-05-03,5388.8823 +FOODS_2,2016-05-04,5440.1694 +FOODS_2,2016-05-05,5520.1226 +FOODS_2,2016-05-06,6055.701 +FOODS_2,2016-05-07,7401.1245 +FOODS_2,2016-05-08,7905.652 +FOODS_2,2016-05-09,6759.285 +FOODS_2,2016-05-10,5789.94 +FOODS_2,2016-05-11,5636.6777 +FOODS_2,2016-05-12,5507.3813 +FOODS_2,2016-05-13,5849.2007 +FOODS_2,2016-05-14,7064.666 +FOODS_2,2016-05-15,7434.761 +FOODS_2,2016-05-16,6070.2705 +FOODS_2,2016-05-17,5095.434 +FOODS_2,2016-05-18,4870.112 +FOODS_2,2016-05-19,4596.719 +FOODS_2,2016-05-20,4982.272 +FOODS_2,2016-05-21,6231.084 +FOODS_2,2016-05-22,6697.241 +FOODS_3,2016-04-25,17230.807 +FOODS_3,2016-04-26,15789.233 +FOODS_3,2016-04-27,15369.597 +FOODS_3,2016-04-28,15617.65 +FOODS_3,2016-04-29,18008.902 +FOODS_3,2016-04-30,23040.523 +FOODS_3,2016-05-01,23571.592 +FOODS_3,2016-05-02,18340.871 +FOODS_3,2016-05-03,16562.752 +FOODS_3,2016-05-04,16747.717 +FOODS_3,2016-05-05,17052.62 +FOODS_3,2016-05-06,19778.832 +FOODS_3,2016-05-07,24705.719 +FOODS_3,2016-05-08,24711.145 +FOODS_3,2016-05-09,19328.555 +FOODS_3,2016-05-10,17409.87 +FOODS_3,2016-05-11,17146.104 +FOODS_3,2016-05-12,16995.588 +FOODS_3,2016-05-13,19478.918 +FOODS_3,2016-05-14,24267.672 +FOODS_3,2016-05-15,24103.676 +FOODS_3,2016-05-16,18767.125 +FOODS_3,2016-05-17,16687.102 +FOODS_3,2016-05-18,16438.53 +FOODS_3,2016-05-19,15937.973 +FOODS_3,2016-05-20,18391.8 +FOODS_3,2016-05-21,23098.781 +FOODS_3,2016-05-22,23017.9 +HOBBIES_1,2016-04-25,3293.069 +HOBBIES_1,2016-04-26,3175.8167 +HOBBIES_1,2016-04-27,3104.7383 +HOBBIES_1,2016-04-28,3130.194 +HOBBIES_1,2016-04-29,3652.3372 +HOBBIES_1,2016-04-30,4440.7046 +HOBBIES_1,2016-05-01,4260.2666 +HOBBIES_1,2016-05-02,3375.5198 +HOBBIES_1,2016-05-03,3186.3364 +HOBBIES_1,2016-05-04,3148.969 +HOBBIES_1,2016-05-05,3190.6125 +HOBBIES_1,2016-05-06,3787.1 +HOBBIES_1,2016-05-07,4503.3115 +HOBBIES_1,2016-05-08,4216.017 +HOBBIES_1,2016-05-09,3309.7373 +HOBBIES_1,2016-05-10,3130.4082 +HOBBIES_1,2016-05-11,3052.7866 +HOBBIES_1,2016-05-12,3083.465 +HOBBIES_1,2016-05-13,3624.693 +HOBBIES_1,2016-05-14,4420.6533 +HOBBIES_1,2016-05-15,4138.2466 +HOBBIES_1,2016-05-16,3327.1152 +HOBBIES_1,2016-05-17,3115.9226 +HOBBIES_1,2016-05-18,3068.1096 +HOBBIES_1,2016-05-19,3066.814 +HOBBIES_1,2016-05-20,3670.4724 +HOBBIES_1,2016-05-21,4458.9883 +HOBBIES_1,2016-05-22,4195.486 +HOBBIES_2,2016-04-25,359.69662 +HOBBIES_2,2016-04-26,337.56714 +HOBBIES_2,2016-04-27,336.36066 +HOBBIES_2,2016-04-28,346.82523 +HOBBIES_2,2016-04-29,385.13193 +HOBBIES_2,2016-04-30,448.0414 +HOBBIES_2,2016-05-01,428.58112 +HOBBIES_2,2016-05-02,347.9329 +HOBBIES_2,2016-05-03,325.00256 +HOBBIES_2,2016-05-04,332.32742 +HOBBIES_2,2016-05-05,348.48566 +HOBBIES_2,2016-05-06,397.46094 +HOBBIES_2,2016-05-07,463.35196 +HOBBIES_2,2016-05-08,445.56622 +HOBBIES_2,2016-05-09,360.51288 +HOBBIES_2,2016-05-10,336.52014 +HOBBIES_2,2016-05-11,344.28247 +HOBBIES_2,2016-05-12,358.7079 +HOBBIES_2,2016-05-13,409.6697 +HOBBIES_2,2016-05-14,475.46484 +HOBBIES_2,2016-05-15,455.52716 +HOBBIES_2,2016-05-16,381.3134 +HOBBIES_2,2016-05-17,355.26096 +HOBBIES_2,2016-05-18,351.80453 +HOBBIES_2,2016-05-19,363.197 +HOBBIES_2,2016-05-20,404.1736 +HOBBIES_2,2016-05-21,471.72308 +HOBBIES_2,2016-05-22,452.07477 +HOUSEHOLD_1,2016-04-25,7322.5625 +HOUSEHOLD_1,2016-04-26,6919.1865 +HOUSEHOLD_1,2016-04-27,6807.973 +HOUSEHOLD_1,2016-04-28,6930.8086 +HOUSEHOLD_1,2016-04-29,8116.5 +HOUSEHOLD_1,2016-04-30,10300.233 +HOUSEHOLD_1,2016-05-01,10267.285 +HOUSEHOLD_1,2016-05-02,7921.134 +HOUSEHOLD_1,2016-05-03,7284.6235 +HOUSEHOLD_1,2016-05-04,7280.303 +HOUSEHOLD_1,2016-05-05,7267.668 +HOUSEHOLD_1,2016-05-06,8540.003 +HOUSEHOLD_1,2016-05-07,10456.705 +HOUSEHOLD_1,2016-05-08,10088.888 +HOUSEHOLD_1,2016-05-09,7604.306 +HOUSEHOLD_1,2016-05-10,6985.8867 +HOUSEHOLD_1,2016-05-11,6745.779 +HOUSEHOLD_1,2016-05-12,6693.4326 +HOUSEHOLD_1,2016-05-13,7825.8516 +HOUSEHOLD_1,2016-05-14,9895.505 +HOUSEHOLD_1,2016-05-15,9434.72 +HOUSEHOLD_1,2016-05-16,7175.374 +HOUSEHOLD_1,2016-05-17,6486.9434 +HOUSEHOLD_1,2016-05-18,6495.8496 +HOUSEHOLD_1,2016-05-19,6345.693 +HOUSEHOLD_1,2016-05-20,7672.4624 +HOUSEHOLD_1,2016-05-21,9825.944 +HOUSEHOLD_1,2016-05-22,9556.082 +HOUSEHOLD_2,2016-04-25,1931.113 +HOUSEHOLD_2,2016-04-26,1782.4025 +HOUSEHOLD_2,2016-04-27,1758.2009 +HOUSEHOLD_2,2016-04-28,1763.3115 +HOUSEHOLD_2,2016-04-29,2031.7482 +HOUSEHOLD_2,2016-04-30,2627.6375 +HOUSEHOLD_2,2016-05-01,2618.0635 +HOUSEHOLD_2,2016-05-02,1939.3312 +HOUSEHOLD_2,2016-05-03,1775.9213 +HOUSEHOLD_2,2016-05-04,1784.6147 +HOUSEHOLD_2,2016-05-05,1768.6473 +HOUSEHOLD_2,2016-05-06,2086.8005 +HOUSEHOLD_2,2016-05-07,2645.3496 +HOUSEHOLD_2,2016-05-08,2584.8975 +HOUSEHOLD_2,2016-05-09,1910.047 +HOUSEHOLD_2,2016-05-10,1752.4678 +HOUSEHOLD_2,2016-05-11,1736.5 +HOUSEHOLD_2,2016-05-12,1733.4746 +HOUSEHOLD_2,2016-05-13,2037.8102 +HOUSEHOLD_2,2016-05-14,2586.608 +HOUSEHOLD_2,2016-05-15,2516.9414 +HOUSEHOLD_2,2016-05-16,1902.0074 +HOUSEHOLD_2,2016-05-17,1732.1702 +HOUSEHOLD_2,2016-05-18,1742.648 +HOUSEHOLD_2,2016-05-19,1724.6858 +HOUSEHOLD_2,2016-05-20,2035.2493 +HOUSEHOLD_2,2016-05-21,2608.6196 +HOUSEHOLD_2,2016-05-22,2549.3857 diff --git a/nbs/docs/getting-started/7_why_timegpt.ipynb b/nbs/docs/getting-started/7_why_timegpt.ipynb new file mode 100644 index 00000000..9b3c7f08 --- /dev/null +++ b/nbs/docs/getting-started/7_why_timegpt.ipynb @@ -0,0 +1,888 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "!pip install -Uqq nixtla" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/Users/yibeihu/opt/anaconda3/envs/report/lib/python3.9/site-packages/dask/dataframe/__init__.py:42: FutureWarning: \n", + "Dask dataframe query planning is disabled because dask-expr is not installed.\n", + "\n", + "You can install it with `pip install dask[dataframe]` or `conda install dask`.\n", + "This will raise in a future version.\n", + "\n", + " warnings.warn(msg, FutureWarning)\n" + ] + } + ], + "source": [ + "#| hide \n", + "from nixtla.utils import in_colab" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide \n", + "IN_COLAB = in_colab()" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "if not IN_COLAB:\n", + " from nixtla.utils import colab_badge\n", + " from dotenv import load_dotenv" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# Why TimeGPT?" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "In this notebook, we compare the performance of TimeGPT against three forecasting models: the classical model (ARIMA), the machine learning model (LightGBM), and the deep learning model (N-HiTS), using a subset of data from the M5 Forecasting competition. We want to highlight three top-rated benefits our users love about TimeGPT:\n", + "\n", + "🎯 **Accuracy**: TimeGPT consistently outperforms traditional models by capturing complex patterns with precision.\n", + "\n", + "⚑ **Speed**: Generate forecasts faster without needing extensive training or tuning for each series.\n", + "\n", + "πŸš€ **Ease of Use**: Minimal setup and no complex preprocessing make TimeGPT accessible and ready to use right out of the box!\n", + "\n", + "Before diving into the notebook, please visit our [dashboard](https://dashboard.nixtla.io) to generate your TimeGPT `api_key` and give it a try yourself!" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# Table of Contents\n", + "\n", + "1. [Data Introduction](#data-introduction)\n", + "2. [Model Fitting](#model-fitting-timegpt-arima-lgbregressor-n-hits)\n", + " 1. [Fitting Time GPT](#timegpt)\n", + " 2. [Fitting ARIMA](#classical-models-arima)\n", + " 3. [Fitting Light GBM](#machine-learning-models-lgbmregressor)\n", + " 4. [Fitting NHITS](#n-hits)\n", + "3. [Results and Evaluation](#performance-comparison-and-results)\n", + "4. [Conclusion](#conclusion)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/markdown": [ + "[![](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/Nixtla/nixtla/blob/main/nbs/docs/getting-started/7_why_timegpt.ipynb)" + ], + "text/plain": [ + "" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "#| echo: false\n", + "if not IN_COLAB:\n", + " load_dotenv()\n", + " colab_badge('docs/getting-started/7_why_timegpt')" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "import numpy as np\n", + "import pandas as pd\n", + "import matplotlib.pyplot as plt\n", + "\n", + "from nixtla import NixtlaClient\n", + "from utilsforecast.plotting import plot_series\n", + "from utilsforecast.losses import mae, rmse, smape\n", + "from utilsforecast.evaluation import evaluate" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "nixtla_client = NixtlaClient(\n", + " # api_key = 'my_api_key_provided_by_nixtla'\n", + ")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## 1. Data introduction" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "In this notebook, we’re working with an aggregated dataset from the M5 Forecasting - Accuracy competition. This dataset includes **7 daily time series**, each with **1,941 data points**. The last **28 data points** of each series are set aside as the test set, allowing us to evaluate model performance on unseen data." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "df = pd.read_csv('https://datasets-nixtla.s3.amazonaws.com/demand_example.csv', parse_dates=['ds'])" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
dsy
minmaxcountminmeanmedianmax
unique_id
FOODS_12011-01-292016-05-2219410.02674.0855232665.05493.0
FOODS_22011-01-292016-05-2219410.04015.9840293894.09069.0
FOODS_32011-01-292016-05-22194110.016969.08912916548.028663.0
HOBBIES_12011-01-292016-05-2219410.02936.1226172908.05009.0
HOBBIES_22011-01-292016-05-2219410.0279.053065248.0871.0
HOUSEHOLD_12011-01-292016-05-2219410.06039.5945395984.011106.0
HOUSEHOLD_22011-01-292016-05-2219410.01566.8402891520.02926.0
\n", + "
" + ], + "text/plain": [ + " ds y \n", + " min max count min mean median max\n", + "unique_id \n", + "FOODS_1 2011-01-29 2016-05-22 1941 0.0 2674.085523 2665.0 5493.0\n", + "FOODS_2 2011-01-29 2016-05-22 1941 0.0 4015.984029 3894.0 9069.0\n", + "FOODS_3 2011-01-29 2016-05-22 1941 10.0 16969.089129 16548.0 28663.0\n", + "HOBBIES_1 2011-01-29 2016-05-22 1941 0.0 2936.122617 2908.0 5009.0\n", + "HOBBIES_2 2011-01-29 2016-05-22 1941 0.0 279.053065 248.0 871.0\n", + "HOUSEHOLD_1 2011-01-29 2016-05-22 1941 0.0 6039.594539 5984.0 11106.0\n", + "HOUSEHOLD_2 2011-01-29 2016-05-22 1941 0.0 1566.840289 1520.0 2926.0" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "df.groupby('unique_id').agg({\"ds\":[\"min\",\"max\",\"count\"],\\\n", + " \"y\":[\"min\",\"mean\",\"median\",\"max\"]})" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(13391, 3) (196, 3)\n" + ] + } + ], + "source": [ + "df_train = df.query('ds <= \"2016-04-24\"')\n", + "df_test = df.query('ds > \"2016-04-24\"')\n", + "\n", + "print(df_train.shape, df_test.shape)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## 2. Model Fitting (TimeGPT, ARIMA, LightGBM, N-HiTS)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### 2.1 TimeGPT\n", + "TimeGPT offers a powerful, streamlined solution for time series forecasting, delivering state-of-the-art results with minimal effort. With TimeGPT, there's no need for data preprocessing or feature engineering -- simply initiate the Nixtla client and call `nixtla_client.forecast` to produce accurate, high-performance forecasts tailored to your unique time series.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "INFO:nixtla.nixtla_client:Validating inputs...\n", + "INFO:nixtla.nixtla_client:Inferred freq: D\n", + "INFO:nixtla.nixtla_client:Querying model metadata...\n", + "INFO:nixtla.nixtla_client:Preprocessing dataframes...\n", + "INFO:nixtla.nixtla_client:Calling Forecast Endpoint...\n" + ] + } + ], + "source": [ + "# Forecast with TimeGPT\n", + "fcst_timegpt = nixtla_client.forecast(df = df_train,\n", + " target_col = 'y', \n", + " h=28, # Forecast horizon, predicts the next 28 time steps\n", + " model='timegpt-1-long-horizon', # Use the model for long-horizon forecasting\n", + " finetune_steps=10, # Number of finetuning steps\n", + " level = [90]) # Generate a 90% confidence interval" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "metric\n", + "rmse 592.607378\n", + "smape 0.049403\n", + "Name: TimeGPT, dtype: float64" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "# Evaluate performance and plot forecast\n", + "fcst_timegpt['ds'] = pd.to_datetime(fcst_timegpt['ds'])\n", + "test_df = pd.merge(df_test, fcst_timegpt, 'left', ['unique_id', 'ds'])\n", + "evaluation_timegpt = evaluate(test_df, metrics=[rmse, smape], models=[\"TimeGPT\"])\n", + "evaluation_timegpt.groupby(['metric'])['TimeGPT'].mean()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### 2.2 Classical Models (ARIMA):\n", + "Next, we applied ARIMA, a traditional statistical model, to the same forecasting task. Classical models use historical trends and seasonality to make predictions by relying on linear assumptions. However, they struggled to capture the complex, non-linear patterns within the data, leading to lower accuracy compared to other approaches. Additionally, ARIMA was slower due to its iterative parameter estimation process, which becomes computationally intensive for larger datasets." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "> πŸ“˜ Why Use TimeGPT over Classical Models?\n", + ">\n", + "> * **Complex Patterns**: TimeGPT captures non-linear trends classical models miss.\n", + ">\n", + "> * **Minimal Preprocessing**: TimeGPT requires little to no data preparation.\n", + ">\n", + "> * **Scalability**: TimeGPT can efficiently scales across multiple series without retraining." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "from statsforecast import StatsForecast\n", + "from statsforecast.models import AutoARIMA" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "#Initiate ARIMA model\n", + "sf = StatsForecast(\n", + " models=[AutoARIMA(season_length=7)],\n", + " freq='D'\n", + ")\n", + "# Fit and forecast\n", + "fcst_arima = sf.forecast(h=28, df=df_train) " + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "fcst_arima = pd.read_csv('../../assets/arima_rst.csv', parse_dates=['ds'])" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "metric\n", + "rmse 724.957364\n", + "smape 0.055018\n", + "Name: AutoARIMA, dtype: float64" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "fcst_arima.reset_index(inplace=True)\n", + "test_df = pd.merge(df_test, fcst_arima, 'left', ['unique_id', 'ds'])\n", + "evaluation_arima = evaluate(test_df, metrics=[rmse, smape], models=[\"AutoARIMA\"])\n", + "evaluation_arima.groupby(['metric'])['AutoARIMA'].mean()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### 2.3 Machine Learning Models (LightGBM)\n", + "\n", + "Thirdly, we used a machine learning model, LightGBM, for the same forecasting task, implemented through the automated pipeline provided by our mlforecast library.\n", + "While LightGBM can capture seasonality and patterns, achieving the best performance often requires detailed feature engineering, careful hyperparameter tuning, and domain knowledge. You can try our mlforecast library to simplify this process and get started quickly!" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "> πŸ“˜ Why Use TimeGPT over Machine Learning Models?\n", + ">\n", + "> * **Automatic Pattern Recognition**: Captures complex patterns from raw data, bypassing the need for feature engineering.\n", + ">\n", + "> * **Minimal Tuning**: Works well without extensive tuning.\n", + ">\n", + "> * **Scalability**: Forecasts across multiple series without retraining." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "import optuna\n", + "from mlforecast.auto import AutoMLForecast, AutoLightGBM\n", + "\n", + "# Suppress Optuna's logging output\n", + "optuna.logging.set_verbosity(optuna.logging.ERROR)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "# Initialize an automated forecasting pipeline using AutoMLForecast.\n", + "mlf = AutoMLForecast(\n", + " models=[AutoLightGBM()],\n", + " freq='D',\n", + " season_length=7, \n", + " fit_config=lambda trial: {'static_features': ['unique_id']}\n", + ")\n", + "\n", + "# Fit the model to the training dataset.\n", + "mlf.fit(\n", + " df=df_train.astype({'unique_id': 'category'}),\n", + " n_windows=1,\n", + " h=28,\n", + " num_samples=10,\n", + ")\n", + "fcst_lgbm = mlf.predict(28)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "fcst_lgbm = pd.read_csv('../../assets/lgbm_rst.csv', parse_dates=['ds'])" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "metric\n", + "rmse 687.773744\n", + "smape 0.051448\n", + "Name: AutoLightGBM, dtype: float64" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "test_df = pd.merge(df_test, fcst_lgbm, 'left', ['unique_id', 'ds'])\n", + "evaluation_lgbm = evaluate(test_df, metrics=[rmse, smape], models=[\"AutoLightGBM\"])\n", + "evaluation_lgbm.groupby(['metric'])['AutoLightGBM'].mean()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### 2.4 N-HiTS\n", + "\n", + "Lastly, we used N-HiTS, a state-of-the-art deep learning model designed for time series forecasting. The model produced accurate results, demonstrating its ability to capture complex, non-linear patterns within the data. However, setting up and tuning N-HiTS required significantly more time and computational resources compared to TimeGPT." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "> πŸ“˜ Why Use TimeGPT Over Deep Learning Models?\n", + ">\n", + "> * **Faster Setup**: Quick setup and forecasting, unlike the lengthy configuration and training times of neural networks.\n", + ">\n", + "> * **Less Tuning**: Performs well with minimal tuning and preprocessing, while neural networks often need extensive adjustments.\n", + ">\n", + "> * **Ease of Use**: Simple deployment with high accuracy, making it accessible without deep technical expertise." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "from neuralforecast.core import NeuralForecast\n", + "from neuralforecast.models import NHITS" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| eval: false\n", + "# Initialize the N-HiTS model.\n", + "models = [NHITS(h=28, \n", + " input_size=28, \n", + " max_steps=100)]\n", + "\n", + "# Fit the model using training data\n", + "nf = NeuralForecast(models=models, freq='D')\n", + "nf.fit(df=df_train)\n", + "fcst_nhits = nf.predict()" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "fcst_nhits = pd.read_csv('../../assets/nhits_rst.csv', parse_dates=['ds'])" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "metric\n", + "rmse 605.011948\n", + "smape 0.053446\n", + "Name: NHITS, dtype: float64" + ] + }, + "execution_count": null, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "test_df = pd.merge(df_test,fcst_nhits, 'left', ['unique_id', 'ds'])\n", + "evaluation_nhits = evaluate(test_df, metrics=[rmse, smape], models=[\"NHITS\"])\n", + "evaluation_nhits.groupby(['metric'])['NHITS'].mean()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## 3. Performance Comparison and Results:\n", + "The performance of each model is evaluated using RMSE (Root Mean Squared Error) and SMAPE (Symmetric Mean Absolute Percentage Error). While RMSE emphasizes the models’ ability to control significant errors, SMAPE provides a relative performance perspective by normalizing errors as percentages. Below, we present a snapshot of performance across all groups. The results demonstrate that TimeGPT outperforms other models on both metrics.\n", + "\n", + "🌟 For a deeper dive into benchmarking, check out our benchmark repository. The summarized results are displayed below:" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Overall Performance Metrics\n", + "\n", + "| **Model** | **RMSE** | **SMAPE** |\n", + "|------------------|----------|-----------|\n", + "| ARIMA | 724.9 | 5.50% |\n", + "| LightGBM | 687.8 | 5.14% |\n", + "| N-HiTS | 605.0 | 5.34% |\n", + "| **TimeGPT** | **592.6**| **4.94%** |\n", + " \n", + "\n", + "#### Breakdown for Each Time-series\n", + "Followed below are the metrics for each individual time series groups. TimeGPT consistently delivers accurate forecasts across all time series groups. In many cases, it performs as well as or better than data-specific models, showing its versatility and reliability across different datasets." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| echo: false\n", + "evaluation_df = evaluation_arima.merge(evaluation_lgbm, on = ['unique_id','metric'], how = 'left')\\\n", + " .merge(evaluation_nhits, on = ['unique_id','metric'], how = 'left')\\\n", + " .merge(evaluation_timegpt, on = ['unique_id','metric'], how = 'left')" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [ + { + "data": { + "image/png": "iVBORw0KGgoAAAANSUhEUgAABW4AAAJOCAYAAAAnP56mAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjkuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8hTgPZAAAACXBIWXMAAA9hAAAPYQGoP6dpAAClZklEQVR4nOzdd1xW9f//8efFBlkuQHKAI3FrmmaaI0lEU/Pr3iu1XJllRsNVScvStGypaImamWaO3CvFnFiaWRoqKmI5wIkC5/dHP66Pl4BCjOsSHvfb7dzqnPM+57zeh0t98bre531MhmEYAgAAAAAAAADYDDtrBwAAAAAAAAAAsEThFgAAAAAAAABsDIVbAAAAAAAAALAxFG4BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsDIVbAAAAAAAAALAxFG4BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgDuQwEBAerXr5+1w4AV9evXTwEBAf/p2GbNmqlZs2a5Gg8AAEBOREREyGQy6fjx49YOBVZy/PhxmUwmRUREZPvYzZs3y2QyafPmzbkeF2BNFG6BAi4tAUpbHBwc9MADD6hfv346ffp0uvbNmjWTyWRSpUqVMjzfunXrzOf69ttvLfb9+uuv6tSpk8qVKycXFxc98MADeuKJJzR9+nSLdgEBARYx3b60atUqS/2Kj4/Xiy++qKCgILm5ualIkSKqW7eu3nzzTV26dClrNwd5rkuXLjKZTBo7dqy1Q8kzaZ/dp59+OsP9r776qrnNP//8k8/RAQCQv06fPq02bdrI09NTVatW1Q8//JCuzXfffScfHx8lJCSk25eamqp58+apQYMGKlasmDw8PPTggw+qT58+2rlzp7ldWpHGZDLp66+/zjCWRo0ayWQyqXr16hnuT0lJkb+/v0wmk1avXp1hmwkTJljkqm5ubqpatapee+01JSYmmtvdmXPfudwe+90sXbpUoaGhKlGihJycnOTv768uXbpo48aNWToeee/w4cMymUxycXEpsL93pH3u7ezsFBsbm25/YmKiXF1dZTKZNHz4cCtECBQeDtYOAED+mDRpkgIDA3Xjxg3t3LlTERER+umnn3Tw4EG5uLhYtHVxcdHRo0e1a9cu1a9f32Lf/Pnz5eLiohs3blhs37Fjh5o3b66yZctq0KBB8vPzU2xsrHbu3Klp06ZpxIgRFu1r166tF154IV2c/v7+9+zL7t271bp1a125ckW9evVS3bp1JUl79uzR22+/ra1bt2rt2rVZui/3qyNHjsjOzra/e0tMTNQPP/yggIAALViwQG+//bZMJpO1w8oTLi4uWrJkiT755BM5OTlZ7FuwYEGGf2YAACiI+vbtq9OnT+udd97R9u3b1blzZ/3+++/mp0Ru3LihF198UW+++aa8vLzSHT9y5Eh9/PHHat++vXr27CkHBwcdOXJEq1evVvny5fXII49YtHdxcVFkZKR69eplsf348ePasWNHujz3dhs3blRcXJwCAgI0f/58hYaGZtp25syZcnd315UrV7R27Vq99dZb2rhxo7Zv326R36Tl3HeqWLFipueWJMMwNGDAAEVERKhOnToaPXq0/Pz8FBcXp6VLl6pFixbavn27Hn300bue537Wu3dvdevWTc7OztYO5a6+/vpr+fn56eLFi/r2228z/fK+IHB2dtaCBQv00ksvWWz/7rvvrBQRUPhQuAUKidDQUNWrV0+S9PTTT6tEiRJ65513tHz5cnXp0sWibYUKFZScnKwFCxZYFG5v3LihpUuXqk2bNlqyZInFMW+99Za8vLy0e/dueXt7W+w7d+5cungeeOCBdAl2Vly6dEkdOnSQvb299u/fr6CgoHRxfPHFF9k+7/3AMAzduHFDrq6uNp/QStKSJUuUkpKi2bNn6/HHH9fWrVvVtGnTXDn31atXVaRIkVw5V25o1aqVli9frtWrV6t9+/bm7Tt27FBMTIw6duyY7s8MAAAFzfXr17Vx40Zt3rxZTZo00TPPPKMdO3ZozZo1GjJkiCTp/fffl5eXV4bFrvj4eH3yyScaNGiQPv/8c4t9U6dO1d9//53umNatW2v58uX6559/VKJECfP2yMhI+fr6qlKlSrp48WKG8X799dd66KGH1LdvX73yyit3zS86depkPv8zzzyjjh076rvvvtPOnTvVsGFDc7vbc+7smDJliiIiIjRq1Ch98MEHFsXgV199VV999ZUcHArmr+9p993e3l729vbWDueuDMNQZGSkevTooZiYGM2fPz/XCrepqam6efPmXb9syG+tW7fOsHAbGRmZ4e+EAHKfbQ/XApBnHnvsMUnSsWPHMtzfvXt3LVq0SKmpqeZtP/zwg65du5au0Jt2nmrVqqUr2kqSj49P7gQt6bPPPtPp06f1wQcfpCvaSpKvr69ee+01i22ffPKJqlWrJmdnZ/n7+2vYsGHpHmtq1qyZqlevrl9++UVNmzaVm5ubKlasaJ4OYsuWLWrQoIFcXV1VuXJlrV+/3uL4tMeJfv/9d3Xp0kWenp4qXry4nnvuuXQjLefMmaPHH39cPj4+cnZ2VtWqVTVz5sx0fQkICNCTTz6pNWvWqF69enJ1ddVnn31m3nf7HLe3bt3SxIkTValSJbm4uKh48eJq3Lix1q1bZ3HOjRs36rHHHlORIkXk7e2t9u3b6/Dhwxn25ejRo+rXr5+8vb3l5eWl/v3769q1axn8VDI2f/58PfHEE2revLmqVKmi+fPnZ9gu7Z6VLFnSfH9fffXVdPH89ttv6tGjh4oWLarGjRtLkpKTk/XGG2+oQoUKcnZ2VkBAgF555RUlJSVZXGPPnj0KCQlRiRIl5OrqqsDAQA0YMMCizcKFC1W3bl15eHjI09NTNWrU0LRp07LU1wceeEBNmjRRZGRkuntQo0aNTB/RXLx4serWrStXV1eVKFFCvXr1ynAKk2XLlql69epycXFR9erVtXTp0gzPl5qaqqlTp6patWpycXGRr6+vhgwZkukvrLebPn26qlWrJjc3NxUtWlT16tVL1x8AAO7mxo0bMgxDRYsWlfTvdELe3t7m/OH06dN6++23NW3atAyfHIqJiZFhGGrUqFG6fSaTKcOcsn379nJ2dtbixYsttkdGRqpLly6ZFgKvX7+upUuXqlu3burSpYuuX7+u77//Pst9ffzxx80x59T169cVHh6uoKAgvf/++xk+odS7d2+LARV//fWXOnfurGLFisnNzU2PPPKIVq5caXFM2nQS33zzjSZOnKgHHnhAHh4e6tSpkxISEpSUlKRRo0bJx8dH7u7u6t+/f7ocKu1R+Pnz56ty5cpycXFR3bp1tXXrVot2J06c0NChQ1W5cmW5urqqePHi6ty5c7r5atOmlNiyZYuGDh0qHx8flS5d2mLf7cdkJYe7evWqXnjhBZUpU0bOzs6qXLmy3n//fRmGkWFf0vIqZ2dnVatWTT/++OPdf0C32b59u44fP65u3bqpW7du2rp1q06dOpWuXWpqqqZNm6YaNWrIxcVFJUuWVKtWrbRnz54M723a7yppsezfv1+hoaHy9PSUu7u7WrRokW66jazk/2fPnlX//v1VunRpOTs7q1SpUmrfvn2W5xHu0aOHoqOj9fvvv1ucc+PGjerRo0eGx5w7d04DBw6Ur6+vXFxcVKtWLc2dOzddu0uXLqlfv37y8vKSt7e3+vbtm+nUE7///rs6deqkYsWKycXFRfXq1dPy5cvvGf+ff/6pjh07ys/PTy4uLipdurS6deuW4TQtgK0qmF/ZAbintH+s0xLrO/Xo0UMTJkzQ5s2bzYlpZGSkWrRokWHSXK5cOUVFRengwYOZFqlud+vWrQzn+yxSpIhcXV0zPW758uVydXVVp06d7nkN6d+i38SJExUcHKxnn31WR44c0cyZM7V7925t375djo6O5rYXL17Uk08+qW7duqlz586aOXOmunXrpvnz52vUqFF65pln1KNHD7333nvq1KmTYmNj5eHhYXG9Ll26KCAgQOHh4dq5c6c++ugjXbx4UfPmzTO3mTlzpqpVq6Z27drJwcFBP/zwg4YOHarU1FQNGzbM4nxHjhxR9+7dNWTIEA0aNEiVK1fOtJ/h4eF6+umnVb9+fSUmJmrPnj3at2+fnnjiCUnS+vXrFRoaqvLly2vChAm6fv26pk+frkaNGmnfvn3pXnTVpUsXBQYGKjw8XPv27dOXX34pHx8fvfPOO/e872fOnNGmTZvMSVr37t314YcfasaMGRZTCfzyyy967LHH5OjoqMGDBysgIEDHjh3TDz/8oLfeesvinJ07d1alSpU0efJkcyL+9NNPa+7cuerUqZNeeOEF/fzzzwoPD9fhw4fNxc1z586pZcuWKlmypF5++WV5e3vr+PHjFo94rVu3Tt27d1eLFi3M/Tt8+LC2b9+u55577p79lf79M/Pcc8/pypUrcnd3V3JyshYvXqzRo0dnOE1CRESE+vfvr4cffljh4eGKj4/XtGnTtH37du3fv9/8JcjatWvVsWNHVa1aVeHh4Tp//rw5Ab/TkCFDzOcdOXKkYmJiNGPGDO3fvz/d5/12X3zxhUaOHKlOnTqZv2z45Zdf9PPPP2ealAMAcKeiRYuqQoUKmjx5siZPnqwdO3YoOjra/L6Dl156SaGhoWrSpEmGx5crV07Sv19sdu7cWW5ubve8ppubm9q3b68FCxbo2WeflSQdOHBAhw4d0pdffqlffvklw+OWL1+uK1euqFu3bvLz81OzZs00f/78LP+7lzb4oXjx4hbbExIS0uW4JpMpXbvb/fTTT7pw4YJGjRqVpRGn8fHxevTRR3Xt2jWNHDlSxYsX19y5c9WuXTt9++236tChg0X78PBwubq66uWXX9bRo0c1ffp0OTo6ys7OThcvXtSECRPMU6kFBgZq3LhxFsdv2bJFixYt0siRI+Xs7KxPPvlErVq10q5du8x5/+7du7Vjxw5169ZNpUuX1vHjxzVz5kw1a9ZMv/32W7qf5dChQ1WyZEmNGzdOV69ezbCfWcnhDMNQu3bttGnTJg0cOFC1a9fWmjVrNGbMGJ0+fVoffvhhunv93XffaejQofLw8NBHH32kjh076uTJk3f9GaWZP3++KlSooIcffljVq1eXm5ubFixYoDFjxli0GzhwoCIiIhQaGqqnn35aycnJ2rZtm3bu3GkxInvjxo365ptvNHz4cJUoUUIBAQE6dOiQHnvsMXl6euqll16So6OjPvvsMzVr1sw8mETKWv7fsWNHHTp0SCNGjFBAQIDOnTundevW6eTJk1l6yW2TJk1UunRpRUZGatKkSZKkRYsWyd3dXW3atEnX/vr162rWrJmOHj2q4cOHKzAwUIsXL1a/fv106dIlc15tGIbat2+vn376Sc8884yqVKmipUuXqm/fvunOeejQITVq1EgPPPCAXn75ZRUpUkTffPONnnrqKS1ZsiTd5z3NzZs3FRISoqSkJI0YMUJ+fn46ffq0VqxYoUuXLmU4VQtgkwwABdqcOXMMScb69euNv//+24iNjTW+/fZbo2TJkoazs7MRGxtr0b5p06ZGtWrVDMMwjHr16hkDBw40DMMwLl68aDg5ORlz5841Nm3aZEgyFi9ebD5u7dq1hr29vWFvb280bNjQeOmll4w1a9YYN2/eTBdTuXLlDEkZLuHh4XftT9GiRY1atWplqe/nzp0znJycjJYtWxopKSnm7TNmzDAkGbNnz7botyQjMjLSvO333383JBl2dnbGzp07zdvXrFljSDLmzJlj3jZ+/HhDktGuXTuLGIYOHWpIMg4cOGDedu3atXSxhoSEGOXLl7fYlnaffvzxx3Tty5UrZ/Tt29e8XqtWLaNNmzZ3uRuGUbt2bcPHx8c4f/68eduBAwcMOzs7o0+fPun6MmDAAIvjO3ToYBQvXvyu10jz/vvvG66urkZiYqJhGIbxxx9/GJKMpUuXWrRr0qSJ4eHhYZw4ccJie2pqarp4unfvbtEmOjrakGQ8/fTTFttffPFFQ5KxceNGwzAMY+nSpYYkY/fu3ZnG+9xzzxmenp5GcnJylvp3O0nGsGHDjAsXLhhOTk7GV199ZRiGYaxcudIwmUzG8ePHzX34+++/DcMwjJs3bxo+Pj5G9erVjevXr5vPtWLFCkOSMW7cOPO22rVrG6VKlTIuXbpk3rZ27VpDklGuXDnztm3bthmSjPnz51vE9+OPP6bb3rRpU6Np06bm9fbt25v/3AMAkBMbNmwwihYtas7tRo0aZRiGYWzfvt1wdXU1jh8/ftfj+/TpY0gyihYtanTo0MF4//33jcOHD6drd3s+umLFCsNkMhknT540DMMwxowZY86rbs9tb/fkk08ajRo1Mq9//vnnhoODg3Hu3DmLdmn/hh85csT4+++/jZiYGOOzzz4znJ2dDV9fX+Pq1auGYfwv585ocXZ2vmufp02blmGelJlRo0YZkoxt27aZt12+fNkIDAw0AgICzHlv2j2qXr26RU7evXt3w2QyGaGhoRbnbdiwoUVuYRiGuQ979uwxbztx4oTh4uJidOjQwbwto/w2KirKkGTMmzfPvC3tPjVu3Dhd3pW2LyYmxjCMrOVwy5YtMyQZb775psX2Tp06GSaTyTh69KhFX5ycnCy2HThwwJBkTJ8+PdNrpLl586ZRvHhx49VXXzVv69GjR7rfTTZu3GhIMkaOHJnuHLfnuGm/Zxw6dMiizVNPPWU4OTkZx44dM287c+aM4eHhYTRp0sS87V75/8WLFw1JxnvvvXfPvt3p9tz1xRdfNCpWrGje9/DDDxv9+/c392HYsGHmfVOnTjUkGV9//bV5282bN42GDRsa7u7u5t8N0n5u7777rrldcnKy8dhjj6X7PatFixZGjRo1jBs3bpi3paamGo8++qhRqVIl87a0z/umTZsMwzCM/fv3p/udFbgfMVUCUEgEBwerZMmSKlOmjDp16qQiRYpo+fLlGY7aS9OjRw999913unnzpr799lvZ29tn+o3mE088oaioKLVr104HDhzQu+++q5CQED3wwAMZPsbSoEEDrVu3Lt3SvXv3u/YjMTEx3SjXzKxfv143b97UqFGjLB7HGzRokDw9PdM9Tubu7q5u3bqZ1ytXrixvb29VqVLF/M12WuzSv4+o3enOEbNpL2VbtWqVedvtI4rTRmU0bdpUf/31V7rHdgIDAxUSEnLPvnp7e+vQoUP6888/M9wfFxen6Oho9evXT8WKFTNvr1mzpp544gmL+NI888wzFuuPPfaYzp8/b/EG5czMnz9fbdq0Mf+sKlWqpLp161pMl/D3339r69atGjBggMqWLWtxfEaPCN4ZT1rMo0ePttie9tK7tJ9v2sjVFStW6NatWxnG6+3tratXr6abWiI7ihYtqlatWmnBggWS/h2h/uijj5pHD91uz549OnfunIYOHWoxj1mbNm0UFBRkjj3t59a3b1+LUQFPPPGEqlatanHOxYsXy8vLS0888YT++ecf81K3bl25u7tr06ZNmcbu7e2tU6dOaffu3f+5/wAASP9OIXDy5Ent3LlTJ0+e1IcffqjU1FSNHDlSL7zwgsqVK6eZM2cqKChIlStX1qeffmpx/Jw5czRjxgwFBgZq6dKlevHFF1WlShW1aNEiw+mEJKlly5YqVqyYFi5cKMMwtHDhwrvmlOfPn9eaNWss2nTs2NE8rUBGKleurJIlSyowMFBDhgxRxYoVtXLlynQjST/++ON0+e3q1avves/Scqus5rirVq1S/fr1zVNHSf/msYMHD9bx48f122+/WbTv06ePxVM3DRo0ML8M7XYNGjRQbGyskpOTLbY3bNjQ/DJgSSpbtqzat2+vNWvWKCUlRZJlfnvr1i2dP39eFStWlLe3t/bt25euD4MGDbrn6OKs5HCrVq2Svb29Ro4cabH9hRdekGEY6e59cHCwKlSoYF6vWbOmPD09M8zr77R69WqdP3/e4nPTvXt38wjvNEuWLJHJZNL48ePTnePOHLdp06YWOV1KSorWrl2rp556SuXLlzdvL1WqlHr06KGffvrJ/Hm5V/7v6uoqJycnbd68OUvTZmWmR48eOnr0qHbv3m3+b2Yj01etWiU/Pz+Le+To6KiRI0fqypUr2rJli7mdg4ODeZS8JNnb26d7ofWFCxe0ceNGdenSRZcvXzbnt+fPn1dISIj+/PPPTP9eSMud16xZk63p3gBbQ+EWKCTSkshvv/1WrVu31j///HPPF1ylzf+zevVqzZ8/X08++eRdE8qHH35Y3333nS5evKhdu3YpLCxMly9fVqdOndIlkCVKlFBwcHC6JaMi1+08PT11+fLlLPX5xIkTkpRuegEnJyeVL1/evD9N6dKl0yVTXl5eKlOmTLptkjJMgCpVqmSxXqFCBdnZ2VnMI7V9+3YFBweb55ktWbKkXnnlFUnKsHCbFZMmTdKlS5f04IMPqkaNGhozZozFo4GZ3QtJqlKliv755590j6ndWUxNm1bjXonf4cOHtX//fjVq1EhHjx41L82aNdOKFSvMyWZagpyVqTWk9PfixIkTsrOzS/eWZj8/P3l7e5v73LRpU3Xs2FETJ05UiRIl1L59e82ZM8diDrehQ4fqwQcfVGhoqEqXLq0BAwZka76zND169DA/frZs2bJMk9q7/TyCgoLM+9P+e+fnKqNj//zzTyUkJMjHx0clS5a0WK5cuZLhSwLTjB07Vu7u7qpfv74qVaqkYcOGafv27VnrNAAAd3B3d1eDBg3MOdScOXN09uxZvfzyy1q/fr3GjBmjt99+W++++65eeOEFiy8X7ezsNGzYMO3du1f//POPvv/+e4WGhmrjxo0WX7DfztHRUZ07d1ZkZKS2bt2q2NjYu055sGjRIt26dUt16tQx5ykXLlxQgwYNMp2Tf8mSJVq3bp02b96so0eP6uDBgxbFzDT169dPl982b978rvfL09NTkrKV42aW06Xtv92dOV1aLptRjpuampouH80oD3nwwQd17do18wvjrl+/rnHjxpnnmS1RooRKliypS5cuZTifaFZy3KzkcCdOnJC/v3+631Gyei+kf3PcrBQ2v/76awUGBsrZ2dn8ualQoYLc3NwsPjfHjh2Tv7+/xWCJzNx5H/7++29du3Yt059vamqqYmNjJd07/3d2dtY777yj1atXy9fXV02aNNG7776rs2fP3jOu29WpU0dBQUGKjIzU/Pnz5efnZ55K704nTpxQpUqV0s1hfefP48SJEypVqpTc3d0t2t3Z76NHj8owDL3++uvp8tu0wnhmOW5gYKBGjx6tL7/8UiVKlFBISIg+/vhj5rfFfYfCLVBIpCWRHTt21PLly1W9enX16NFDV65cyfSYUqVKqVmzZpoyZYq2bt2a5Tm/nJyc9PDDD2vy5MmaOXOmbt26le6FEf9VUFCQ/vjjD928eTNXzne7zL71z2y7cccLDzJyZyH42LFjatGihf755x998MEHWrlypdatW6fnn39ekixeBifprvP93q5JkyY6duyYZs+ererVq+vLL7/UQw89pC+//DJLx2fkv/b766+/liQ9//zzqlSpknmZMmWKbty48Z/fPpvZvchodO6d+7/99ltFRUVp+PDhOn36tAYMGKC6deuaP/8+Pj6Kjo7W8uXLzfOkhYaGZjjP1t20a9dOzs7O6tu3r5KSkjJ8kV9eSU1NlY+PT4Yj2detW2eelywjVapU0ZEjR7Rw4UI1btxYS5YsUePGjTMcKQIAQHYkJibq1Vdf1dtvv60iRYpowYIF6tSpk5566im1b99enTp1yrRYWrx4cbVr106rVq1S06ZN9dNPP6UrxKVJe4nShAkTVKtWrXRPptwu7XqNGjWyyFV++uknRUVFZTj6skmTJgoODlbTpk0tRmzmhrQX7v7666+5et40eZHj3mnEiBF666231KVLF33zzTdau3at1q1bp+LFi6fLb6Ws5bhZyeGy67/2OTExUT/88INiYmIsPjNVq1bVtWvXFBkZ+Z/uW1Zz/YxkJf8fNWqU/vjjD4WHh8vFxUWvv/66qlSpov3792frWj169NCiRYsUGRmprl27ZvhywbyQ9tl58cUXM81x7xzEcbspU6bol19+0SuvvKLr169r5MiRqlatWoYvlANsFYVboBCyt7dXeHi4zpw5oxkzZty1bY8ePbRt2zZ5enqqdevW2b5W2uT7cXFx/ynWO7Vt21bXr1/PUvEvbfTukSNHLLbfvHlTMTEx9xzd+1/c+ajS0aNHlZqaap78/4cfflBSUpKWL1+uIUOGqHXr1goODs5R0pamWLFi6t+/vxYsWKDY2FjVrFlTEyZMkJT5vZD+fUtriRIlVKRIkRzHYBiGIiMj1bx5cy1evDjdUrNmTfMvS2mPfx08ePA/XatcuXJKTU1Nd8/j4+N16dKldD/fRx55RG+99Zb27Nmj+fPn69ChQ1q4cKF5v5OTk9q2batPPvlEx44d05AhQzRv3jwdPXo0yzG5urrqqaee0ubNm/XEE0+oRIkSmcYuZfzzOHLkiHl/2n8zegTuzmMrVKig8+fPq1GjRhmOZq9Vq9ZdYy9SpIi6du2qOXPm6OTJk2rTpo3eeuutDF+sBgBAVk2aNEmBgYHq2bOnpH9fYOrv72/e7+/vn+mjzre7V07ZuHFjlS1bVps3b77rYIOYmBjt2LFDw4cPT5enLFq0SE5OToqMjMxOF3OscePGKlq0qBYsWGCeeuBuypUrl2lOl7Y/N2WUh/zxxx9yc3NTyZIlJUnffvut+vbtqylTpqhTp0564okn1LhxY126dCnH179bDleuXDmdOXMm3Wjl3L4X3333nW7cuKGZM2em+9y8+eabOnHihPlppQoVKujMmTO6cOFCtq9TsmRJubm5ZfrztbOzsxgpfbf8P02FChX0wgsvaO3atTp48KBu3rypKVOmZCuuHj16KC4uTn/88cdd/3yVK1dOf/75Z7pi/Z0/j3LlyikuLi5dAf7Ofqf9vuDo6JhhfhscHHzPKUZq1Kih1157TVu3btW2bdt0+vTpdFO0ALaMwi1QSDVr1kz169fX1KlT71qY6dSpk8aPH69PPvlETk5OmbbbtGlTht8yp81DmtHjPv/FM888o1KlSumFF17QH3/8kW7/uXPn9Oabb0r6dw4rJycnffTRRxaxzZo1SwkJCRm+CTWnPv74Y4v1tLcoh4aGSvrft/y3x5OQkKA5c+bk6Lrnz5+3WHd3d1fFihXNj5KVKlVKtWvX1ty5cy0S6IMHD2rt2rX/qSifke3bt+v48ePq37+/OnXqlG7p2rWrNm3apDNnzqhkyZJq0qSJZs+erZMnT1qcJysjFtJinjp1qsX2Dz74QJLMP9+LFy+mO1/t2rUlyXx/7rx/dnZ2qlmzpkWbrHrxxRc1fvx4vf7665m2qVevnnx8fPTpp59anH/16tU6fPiwOfbbf263P9a1bt26dNOPdOnSRSkpKXrjjTfSXS85Ofmuvzjd2X8nJydVrVpVhmFkOqccAAD38scff2jGjBmaNm2a+QkZX19fcxFH+neKJT8/P0nS2bNn0/37Jv37pfuGDRsynCIpjclk0kcffaTx48erd+/emcaU9gXySy+9lC5P6dKli5o2bZrpCOC84ubmprFjx+rw4cMaO3ZshnnQ119/rV27dkn6NwfatWuXoqKizPuvXr2qzz//XAEBAXcdbfxfREVFWcxTGxsbq++//14tW7Y057b29vbp4p4+fXqWCtGZyUoO17p1a6WkpKQbjPLhhx/KZDKZc/Cc+vrrr1W+fHk988wz6T43L774otzd3c2fm44dO8owDE2cODHdee6V49rb26tly5b6/vvvLaZai4+PV2RkpBo3bmyeWuNe+f+1a9fS/Z5XoUIFeXh4ZDu/rVChgqZOnarw8HDVr18/03atW7fW2bNntWjRIvO25ORkTZ8+Xe7u7mratKm5XXJysmbOnGlul5KSYv7dKY2Pj4+aNWumzz77LMMvbdKm6shIYmJiuvmaa9SoITs7u2z3H7AmB2sHAMB6xowZo86dOysiIiLdi5/SeHl5pfvWNiMjRozQtWvX1KFDBwUFBenmzZvasWOHFi1apICAAPXv39+i/enTp82P1N/O3d1dTz31VKbXKVq0qJYuXarWrVurdu3a6tWrl3l+sX379mnBggVq2LChpH+/sQ4LC9PEiRPVqlUrtWvXTkeOHNEnn3yihx9+WL169bpnv7IrJiZG7dq1U6tWrRQVFaWvv/5aPXr0MI92bNmypXlk55AhQ3TlyhV98cUX8vHxydGo5KpVq6pZs2aqW7euihUrpj179ujbb7/V8OHDzW3ee+89hYaGqmHDhho4cKCuX7+u6dOnZ/lnnBXz58+Xvb19pkXxdu3a6dVXX9XChQs1evRoffTRR2rcuLEeeughDR48WIGBgTp+/LhWrlyp6Ojou16rVq1a6tu3rz7//HNdunRJTZs21a5duzR37lw99dRT5vnk5s6dq08++UQdOnRQhQoVdPnyZX3xxRcWo8iffvppXbhwQY8//rhKly6tEydOaPr06apdu7Z5Tq6sqlWr1j1Htzo6Ouqdd95R//791bRpU3Xv3l3x8fGaNm2aAgICzFNnSFJ4eLjatGmjxo0ba8CAAbpw4YKmT5+uatWqWYxSaNq0qYYMGaLw8HBFR0erZcuWcnR01J9//qnFixdr2rRp6tSpU4bxtGzZUn5+fmrUqJF8fX11+PBhzZgxw+IFcwAAZNfzzz+vrl27WhR6OnXqpPbt25vn9//hhx+0YsUKSdKpU6dUv359Pf7442rRooX8/Px07tw5LViwQAcOHNCoUaMyfZpFktq3b6/27dvfNab58+erdu3a6eZ3TdOuXTuNGDFC+/bt00MPPZTdLmv16tUWhek0jz76qMXLpu40ZswYHTp0SFOmTNGmTZvUqVMn+fn56ezZs1q2bJl27dqlHTt2SJJefvllLViwQKGhoRo5cqSKFSumuXPnKiYmRkuWLMn1x9irV6+ukJAQjRw5Us7Ozvrkk08kyaIw+eSTT+qrr76Sl5eXqlatqqioKK1fv17Fixf/z9fNSg7Xtm1bNW/eXK+++qqOHz+uWrVqae3atfr+++81atSoXJnW4syZM9q0aVO6F6ClcXZ2VkhIiBYvXqyPPvpIzZs3V+/evfXRRx/pzz//VKtWrZSamqpt27apefPmFvl5Rt58802tW7dOjRs31tChQ+Xg4KDPPvtMSUlJevfdd83t7pX///HHH2rRooW6dOmiqlWrysHBQUuXLlV8fHym80XfzXPPPXfPNoMHD9Znn32mfv36ae/evQoICNC3336r7du3a+rUqea8sm3btmrUqJFefvllHT9+XFWrVtV3332X4fyzH3/8sRo3bqwaNWpo0KBBKl++vOLj4xUVFaVTp07pwIEDGcayceNGDR8+XJ07d9aDDz6o5ORkffXVV7K3t1fHjh2z3X/AagwABdqcOXMMScbu3bvT7UtJSTEqVKhgVKhQwUhOTjYMwzCaNm1qVKtW7a7n3LRpkyHJWLx4sXnb6tWrjQEDBhhBQUGGu7u74eTkZFSsWNEYMWKEER8fb3F8uXLlDEkZLuXKlctSv86cOWM8//zzxoMPPmi4uLgYbm5uRt26dY233nrLSEhIsGg7Y8YMIygoyHB0dDR8fX2NZ5991rh48aJFm8z6Xa5cOaNNmzbptksyhg0bZl4fP368Icn47bffjE6dOhkeHh5G0aJFjeHDhxvXr1+3OHb58uVGzZo1DRcXFyMgIMB45513jNmzZxuSjJiYmHteO21f3759zetvvvmmUb9+fcPb29twdXU1goKCjLfeesu4efOmxXHr1683GjVqZLi6uhqenp5G27Ztjd9++82iTVpf/v77b4vtaZ+l22O83c2bN43ixYsbjz32WIb70wQGBhp16tQxrx88eNDo0KGD4e3tbbi4uBiVK1c2Xn/99XvGYxiGcevWLWPixIlGYGCg4ejoaJQpU8YICwszbty4YW6zb98+o3v37kbZsmUNZ2dnw8fHx3jyySeNPXv2mNt8++23RsuWLQ0fHx/DycnJKFu2rDFkyBAjLi7urn0xjPSfhYxk1odFixYZderUMZydnY1ixYoZPXv2NE6dOpXu+CVLlhhVqlQxnJ2djapVqxrfffed0bdv3wz/vHz++edG3bp1DVdXV8PDw8OoUaOG8dJLLxlnzpwxt2natKnRtGlT8/pnn31mNGnSxChevLjh7OxsVKhQwRgzZky6P0sAAGTVypUrDXd3d4t/f9KEh4cb/v7+RqlSpYx33nnHvD0xMdGYNm2aERISYpQuXdpwdHQ0PDw8jIYNGxpffPGFkZqaam6bUT6akdtzvL179xqSLPKMOx0/ftyQZDz//POGYdw9D7ldWp6U2TJnzpy7Hp8mLScpVqyY4eDgYJQqVcro2rWrsXnzZot2x44dMzp16mTOn+rXr2+sWLHCok1m9yiz3w8y6mtanvP1118blSpVMpydnY06deoYmzZtsjj24sWLRv/+/Y0SJUoY7u7uRkhIiPH777+ny1nv9rvJnblmVnI4wzCMy5cvG88//7zh7+9vODo6GpUqVTLee+89i8/L7X25050x3mnKlCmGJGPDhg2ZtomIiDAkGd9//71hGIaRnJxsvPfee0ZQUJDh5ORklCxZ0ggNDTX27t17z3jS+h4SEmK4u7sbbm5uRvPmzY0dO3ZYtLlX/v/PP/8Yw4YNM4KCgowiRYoYXl5eRoMGDYxvvvkm036kyernPqM+xMfHmz8LTk5ORo0aNTL8/J8/f97o3bu34enpaXh5eRm9e/c29u/fn+Gfl2PHjhl9+vQx/Pz8DEdHR+OBBx4wnnzySePbb781t0n7vKd9Nv/66y9jwIABRoUKFQwXFxejWLFiRvPmzY3169ffs/+ALTEZxn+YQRsAYGHChAmaOHGi/v7777uOBAEAAADuFyaTScOGDbvnezEAAHmDOW4BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsDHPcAgAAAAAAAICNYcQtAAAAAAAAANgYCrcAAAAAAAAAYGMcrB3A/SA1NVVnzpyRh4eHTCaTtcMBAABAJgzD0OXLl+Xv7y87u8IzRoF8FQAA4P6QnXyVwm0WnDlzRmXKlLF2GAAAAMii2NhYlS5d2tph5BvyVQAAgPtLVvJVCrdZ4OHhIenfG+rp6WnlaAAAAJCZxMRElSlTxpy/FRbkqwAAAPeH7OSrFG6zIO1xM09PTxJhAACA+0Bhmy6AfBUAAOD+kpV8tfBM/AUAAAAAAAAA9wkKtwAAAAAAAABgYyjcAgAAAAAAAICNYY5bAACQb1JSUnTr1i1rh4H7nJOTk+zsGH8AAAByV2pqqm7evGntMHCfc3R0lL29fa6ci8ItAADIc4Zh6OzZs7p06ZK1Q0EBYGdnp8DAQDk5OVk7FAAAUEDcvHlTMTExSk1NtXYoKAC8vb3l5+eX4xfmUrgFAAB5Lq1o6+PjIzc3txwnMCi8UlNTdebMGcXFxals2bJ8lgAAQI4ZhqG4uDjZ29urTJkyPNmD/8wwDF27dk3nzp2TJJUqVSpH56NwCwAA8lRKSoq5aFu8eHFrh4MCoGTJkjpz5oySk5Pl6Oho7XAAAMB9Ljk5WdeuXZO/v7/c3NysHQ7uc66urpKkc+fOycfHJ0fTJvAVAgAAyFNpc9qSBCO3pE2RkJKSYuVIAABAQZCWUzANE3JL2u8+OX2/B4VbAACQL3ikHbmFzxIAAMgL5BjILbn1WaJwCwAAAAAAAAA2hsItAAAAAAAAANgYXk4GAACs4uDRE/l6veoVy/2n46KiotS4cWO1atVKK1euzNaxEyZM0LJlyxQdHf2frn39+nU98MADsrOz0+nTp+Xs7GyxPyAgQCdO/HsfXV1dVaFCBT333HN6+umnzW02b96s5s2b6+LFi/L29jave3t7Ky4uTi4uLua2u3fvVv369SX9+0bcOwUFBSkmJkYnTpyQn5/ff+oTAADA/YJ89d7IV/MWI24BAADuYtasWRoxYoS2bt2qM2fO5Ou1lyxZomrVqikoKEjLli3LsM2kSZMUFxengwcPqlevXho0aJBWr159z3N7eHho6dKlFttmzZqlsmXLZtj+p59+0vXr19WpUyfNnTs3230BAABA3iBf/VdBzFcp3AIAAGTiypUrWrRokZ599lm1adNGERER5n0RERHy9va2aL9s2TLziwgiIiI0ceJEHThwQCaTSSaTyXz8yZMn1b59e7m7u8vT01NdunRRfHx8uuvPmjVLvXr1Uq9evTRr1qwMY/Tw8JCfn5/Kly+vsWPHqlixYlq3bt09+9a3b1/Nnj3bvH79+nUtXLhQffv2zbD9rFmz1KNHD/Xu3dviOAAAAFgP+aplLAUtX6VwCwAAkIlvvvlGQUFBqly5snr16qXZs2dn+EhWRrp27aoXXnhB1apVU1xcnOLi4tS1a1elpqaqffv2unDhgrZs2aJ169bpr7/+UteuXS2OP3bsmKKiotSlSxd16dJF27ZtMz9mlpHU1FQtWbJEFy9elJOT0z3j6927t7Zt26aTJ09K+ne0REBAgB566KF0bS9fvqzFixerV69eeuKJJ5SQkKBt27Zl6T4AAAAg75Cv/qug5qsUbgEAADKRNoJAklq1aqWEhARt2bIlS8e6urrK3d1dDg4O8vPzk5+fn1xdXbVhwwb9+uuvioyMVN26ddWgQQPNmzdPW7Zs0e7du83Hz549W6GhoSpatKiKFSumkJAQzZkzJ911xo4dK3d3dzk7O6tTp04qWrSoxZxhmfHx8VFoaKh5VMXs2bM1YMCADNsuXLhQlSpVUrVq1WRvb69u3bplOqICAAAA+Yd89V8FNV+lcAsAAJCBI0eOaNeuXerevbskycHBQV27ds1xAnj48GGVKVNGZcqUMW+rWrWqvL29dfjwYUlSSkqK5s6da07CJalXr16KiIhQamqqxfnGjBmj6Ohobdy4UQ0aNNCHH36oihUrZimWAQMGKCIiQn/99ZeioqLUs2fPDNvNnj07XSyLFy/W5cuXs9xvAAAA5C7y1f8pqPmqg7UDAAAAsEWzZs1ScnKy/P39zdsMw5Czs7NmzJghOzu7dI+h3bp1K1euvWbNGp0+fTrd42gpKSnasGGDnnjiCfO2EiVKqGLFiqpYsaIWL16sGjVqqF69eqpateo9rxMaGqrBgwdr4MCBatu2rYoXL56uzW+//aadO3dq165dGjt2rEUsCxcu1KBBg3LQUwAAAPxX5Kv/Ksj5KiNuAQAA7pCcnKx58+ZpypQpio6ONi8HDhyQv7+/FixYoJIlS+ry5cu6evWq+bjo6GiL8zg5OSklJcViW5UqVRQbG6vY2Fjztt9++02XLl0yJ6+zZs1St27dLK4dHR19z0e+ypQpo65duyosLCxL/XRwcFCfPn20efPmTB87mzVrlpo0aaIDBw5YxDJ69OgC8fgZAADA/Yh89X8Kcr7KiFsA2XLwaOYTjWdV9YrlciESAMg7K1as0MWLFzVw4EB5eXlZ7OvYsaNmzZqlNWvWyM3NTa+88opGjhypn3/+2eItvpIUEBCgmJgYRUdHq3Tp0vLw8FBwcLBq1Kihnj17aurUqUpOTtbQoUPVtGlT1atXT3///bd++OEHLV++XNWrV7c4X58+fdShQwdduHBBxYoVyzD25557TtWrV9eePXtUr169e/b1jTfe0JgxYzIcvXDr1i199dVXmjRpUrpYnn76aX3wwQc6dOiQqlWrds/rAHllyar1OTq+Y+vgXIoEAID8Q776r4KerzLiFgAA4A6zZs1ScHBwuiRY+jcR3rNnj06dOqWvv/5aq1atUo0aNbRgwQJNmDAhXdtWrVqpefPmKlmypBYsWCCTyaTvv/9eRYsWVZMmTRQcHKzy5ctr0aJFkqR58+apSJEiatGiRbprt2jRQq6urvr6668zjb1q1apq2bKlxo0bl6W+Ojk5qUSJEjKZTOn2LV++XOfPn1eHDh3S7atSpYqqVKly349iAAAAuB+Rr/6roOerJuPOyS6QTmJiory8vJSQkCBPT09rhwNYFSNuAWTXjRs3FBMTo8DAQLm4uFg7HBQAd/tMFda8rbD2W2LELQAg58hXkdtyK19lxC0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAPeZZs2aadSoUdk6xmQyadmyZXkSDwAAAHA78tXc4WDtAAAAQOG0Yce+fL1ei0cf+k/HRUVFqXHjxmrVqpVWrlyZrWMnTJigZcuWKTo6OlvHRUREaNSoUbp06VKG+7/77js5Ojpm65z3snnzZjVv3lwXL16Ut7e3xb6zZ88qPDxcK1eu1KlTp+Tl5aWKFSuqV69e6tu3r9zc3CRJAQEBOnHihCTJzs5Ovr6+Cg0N1fvvv6+iRYtaXMfb21txcXFycXExX2f37t2qX7++JMkwjFztHwAAQHaRr2aOfDV/8lVG3AIAANzFrFmzNGLECG3dulVnzpyxdjiSpGLFisnDwyNfrvXXX3+pTp06Wrt2rSZPnqz9+/crKipKL730klasWKH169dbtJ80aZLi4uJ08uRJzZ8/X1u3btXIkSPTndfDw0NLly612DZr1iyVLVs2T/sDAABQ0JCvFtx8lcItAABAJq5cuaJFixbp2WefVZs2bRQREWHeFxERke6b/mXLlslkMpn3T5w4UQcOHJDJZJLJZDIff/LkSbVv317u7u7y9PRUly5dFB8fn+W47nz0LC4uTm3atJGrq6sCAwMVGRmpgIAATZ061eK4f/75Rx06dJCbm5sqVaqk5cuXS5KOHz+u5s2bS5KKFi0qk8mkfv36SZKGDh0qBwcH7dmzR126dFGVKlVUvnx5tW/fXitXrlTbtm0truHh4SE/Pz898MADat68ufr27at9+9KPVunbt69mz55tXr9+/boWLlyovn37Zvk+AAAAFHbkqwU7X6VwCwAAkIlvvvlGQUFBqly5snr16qXZs2dn+ZGorl276oUXXlC1atUUFxenuLg4de3aVampqWrfvr0uXLigLVu2aN26dfrrr7/UtWvX/xxnnz59dObMGW3evFlLlizR559/rnPnzqVrN3HiRHXp0kW//PKLWrdurZ49e+rChQsqU6aMlixZIkk6cuSI4uLiNG3aNJ0/f15r167VsGHDVKRIkQyvnZb4Z+T06dP64Ycf1KBBg3T7evfurW3btunkyZOSpCVLliggIEAPPfTfHhEEAAAojMhXC3a+SuEWAAAgE7NmzVKvXr0kSa1atVJCQoK2bNmSpWNdXV3l7u4uBwcH+fn5yc/PT66urtqwYYN+/fVXRUZGqm7dumrQoIHmzZunLVu2aPfu3dmO8ffff9f69ev1xRdfqEGDBnrooYf05Zdf6vr16+na9uvXT927d1fFihU1efJkXblyRbt27ZK9vb2KFSsmSfLx8ZGfn5+8vLx09OhRGYahypUrW5ynRIkScnd3l7u7u8aOHWuxb+zYsXJ3d5erq6tKly4tk8mkDz74IF0sPj4+Cg0NNY/qmD17tgYMGJDt/gMAABRm5KsFO1+lcAsAAJCBI0eOaNeuXerevbskycHBQV27dtWsWbNydN7Dhw+rTJkyKlOmjHlb1apV5e3trcOHD/+nOB0cHCy++a9YsaL55Qq3q1mzpvn/ixQpIk9PzwxHOtzLrl27FB0drWrVqikpKcli35gxYxQdHa1ffvlFGzZskCS1adNGKSkp6c4zYMAARURE6K+//lJUVJR69uyZ7VgAAAAKK/LVzBWUfNUhX68GAABwn5g1a5aSk5Pl7+9v3mYYhpydnTVjxgzZ2dmlewzt1q1b+R1mttz5Zl+TyaTU1NRM21esWFEmk0lHjhyx2F6+fHlJ/47SuFOJEiVUsWJFSVKlSpU0depUNWzYUJs2bVJwcLBF29DQUA0ePFgDBw5U27ZtVbx48f/ULwAAgMKIfLXg56uMuAUAALhDcnKy5s2bpylTpig6Otq8HDhwQP7+/lqwYIFKliypy5cv6+rVq+bjoqOjLc7j5OSU7pv7KlWqKDY2VrGxseZtv/32my5duqSqVatmO9bKlSsrOTlZ+/fvN287evSoLl68mK3zODk5SZJFvMWLF9cTTzyhGTNmWPQzO+zt7SUpw0fhHBwc1KdPH23evJlpEgAAALKBfPVfBT1fZcQtAADAHVasWKGLFy9q4MCB8vLystjXsWNHzZo1S2vWrJGbm5teeeUVjRw5Uj///LPFW3wlKSAgQDExMYqOjlbp0qXl4eGh4OBg1ahRQz179tTUqVOVnJysoUOHqmnTpqpXr5752JSUlHSJtbOzs6pUqWKxLSgoSMHBwRo8eLBmzpwpR0dHvfDCC3J1db3rixjuVK5cOZlMJq1YsUKtW7c2z3n2ySefqFGjRqpXr54mTJigmjVrys7OTrt379bvv/+uunXrWpzn8uXLOnv2rAzDUGxsrF566SWVLFlSjz76aIbXfeONNzRmzBhG2wIAAGQD+WrhyFcZcQsAAHCHWbNmKTg4OF0SLP2bCO/Zs0enTp3S119/rVWrVqlGjRpasGCBJkyYkK5tq1at1Lx5c5UsWVILFiyQyWTS999/r6JFi6pJkyYKDg5W+fLltWjRIotjr1y5ojp16lgsbdu2zTDeefPmydfXV02aNFGHDh00aNAgeXh4yMXFJct9fuCBBzRx4kS9/PLL8vX11fDhwyVJFSpU0P79+xUcHKywsDDVqlVL9erV0/Tp0/Xiiy/qjTfesDjPuHHjVKpUKfn7++vJJ59UkSJFtHbt2kwTXScnJ5UoUSJbSTsAAEBhR75aOPJVk3HnZBdIJzExUV5eXkpISJCnp6e1wwGs6uDREzk+R/WK5XIhEgD3ixs3bigmJkaBgYHZSszw3506dUplypTR+vXr1aJFC2uHk+vu9pkqrHlbYe23JC1ZtT5Hx3dsHXzvRgBQAOT070up4P6dSb6a/8hXs5a3MVUCAADAfW7jxo26cuWKatSoobi4OL300ksKCAhQkyZNrB0aAAAAQL76H1G4BQAAuM/dunVLr7zyiv766y95eHjo0Ucf1fz589O9lRcAAACwBvLV/4bCLQAAwH0uJCREISEh1g4DAAAAyBD56n/Dy8kAAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsDIVbAAAAAAAAALAxFG4BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsjIO1AwAAAIXTklXr8/V6HVsHZ/uYfv36ae7cuQoPD9fLL79s3r5s2TJ16NBBhmFo8+bNat68uS5evChvb2+L4wMCAjRq1CiNGjXKYr127dpq3rz5Xa+9adMmPfbYY3rvvfcUERGhEydOyNXVVZUqVdKgQYP09NNPZ7s/AAAAyDryVfJVa6NwCwAAcBcuLi565513NGTIEBUtWjRXzvnoo48qLi7OvP7cc88pMTFRc+bMMW8rVqyYJk6cqM8++0wzZsxQvXr1lJiYqD179ujixYu5EgcAAADuf+SrBReFWwAAgLsIDg7W0aNHFR4ernfffTdXzunk5CQ/Pz/zuqurq5KSkiy2SdLy5cs1dOhQde7c2bytVq1auRID8s/WrVv13nvvae/evYqLi9PSpUv11FNPmfcbhqHx48friy++0KVLl9SoUSPNnDlTlSpVsl7QAADgvkG+WnAxxy0AAMBd2Nvba/LkyZo+fbpOnTqVr9f28/PTxo0b9ffff+frdZG7rl69qlq1aunjjz/OcP+7776rjz76SJ9++ql+/vlnFSlSRCEhIbpx40Y+RwoAAO5H5KsFF4VbAACAe+jQoYNq166t8ePHZ9qmdOnScnd3t1hOnjyZo+t+8MEH+vvvv+Xn56eaNWvqmWee0erVq3N0TuS/0NBQvfnmm+rQoUO6fYZhaOrUqXrttdfUvn171axZU/PmzdOZM2e0bNmy/A8WAADcl8hXCyYKtwAAAFnwzjvvaO7cuTp8+HCG+7dt26bo6GiLxd/fP0fXrFq1qg4ePKidO3dqwIABOnfunNq2bcuLHgqQmJgYnT17VsHB/3sZiZeXlxo0aKCoqKhMj0tKSlJiYqLFAgAACjfy1YKHwi0AAEAWNGnSRCEhIQoLC8twf2BgoCpWrGixODjk/HUCdnZ2evjhhzVq1Ch99913ioiI0KxZsxQTE5Pjc8P6zp49K0ny9fW12O7r62vel5Hw8HB5eXmZlzJlyuRpnAAAwPaRrxY8vJwMAAAgi95++23Vrl1blStXtloMVatWlfTvvKkovMLCwjR69GjzemJiIsVbAABAvlrAULgFAADIoho1aqhnz5766KOP8uV6nTp1UqNGjfToo4/Kz89PMTExCgsL04MPPqigoKB8iQF5K+3NzPHx8SpVqpR5e3x8vGrXrp3pcc7OznJ2ds7r8AAAwH2GfLVgYaoEAACAbJg0aZJSU1Pz5VohISH64Ycf1LZtWz344IPq27evgoKCtHbt2lx5rA3WFxgYKD8/P23YsMG8LTExUT///LMaNmxoxcgAAMD9iny14DAZhmFYOwhbl5iYKC8vLyUkJMjT09Pa4QBWdfDoiRyfo3rFcrkQCYD7xY0bNxQTE6PAwEC5uLhYOxwUAHf7TNli3nblyhUdPXpUklSnTh198MEHat68uYoVK6ayZcvqnXfe0dtvv625c+cqMDBQr7/+un755Rf99ttvWf4zY4v9zi9LVq3P0fEdWwffuxEAFAA5/ftSKrh/Z5KvIrflVr5K6RsAAADIQ3v27FHz5s3N62lz0/bt21cRERF66aWXdPXqVQ0ePFiXLl1S48aN9eOPP/KLIwAAQCFH4RYAAADIQ82aNdPdHnIzmUyaNGmSJk2alI9RAQAAwNYxxy0AAAAAAAAA2BgKtwAAAAAAAABgY5gqAQAAAAAAIA/xkmcA/wUjbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAP6Dfv366amnnrJ2GAAAAECGyFfvfw7WDgAAABROt8Kn5uv1HMNGZbmtyWS66/7x48dr2rRpMgwjh1FlzdmzZxUeHq6VK1fq1KlT8vLyUsWKFdWrVy/17dtXbm5ukqSAgACdOPHvW6vd3NxUuXJlhYWFqXPnzhb7MtK3b19FRETkR3cAAADuC+SrWUe+mjco3AIAANwhLi7O/P+LFi3SuHHjdOTIEfM2d3d3ubu750ssf/31lxo1aiRvb29NnjxZNWrUkLOzs3799Vd9/vnneuCBB9SuXTtz+0mTJmnQoEFKTEzUlClT1LVrVz3wwAPavXu3UlJSJEk7duxQx44ddeTIEXl6ekqSXF1d86U/AAAAyDny1cKBqRIAAADu4OfnZ168vLxkMpkstrm7u6d79KxZs2YaMWKERo0apaJFi8rX11dffPGFrl69qv79+8vDw0MVK1bU6tWrLa518OBBhYaGyt3dXb6+vurdu7f++ecf8/6hQ4fKwcFBe/bsUZcuXVSlShWVL19e7du318qVK9W2bVuL83l4eMjPz08PPvigPv74Y7m6uuqHH35QyZIlzfEXK1ZMkuTj42PRTwAAANwfyFcLB6sWbsPDw/Xwww/Lw8NDPj4+euqppyy+HZCkGzduaNiwYSpevLjc3d3VsWNHxcfHW7Q5efKk2rRpIzc3N/n4+GjMmDFKTk62aLN582Y99NBDcnZ2VsWKFQvd0GoAAJD35s6dqxIlSmjXrl0aMWKEnn32WXXu3FmPPvqo9u3bp5YtW6p37966du2aJOnSpUt6/PHHVadOHe3Zs0c//vij4uPj1aVLF0nS+fPntXbtWg0bNkxFihTJ8Jp3e0zOwcFBjo6OunnzZu53FgAAAPcd8tX7i1ULt1u2bNGwYcO0c+dOrVu3Trdu3VLLli119epVc5vnn39eP/zwgxYvXqwtW7bozJkz+r//+z/z/pSUFLVp00Y3b97Ujh07NHfuXEVERGjcuHHmNjExMWrTpo2aN2+u6OhojRo1Sk8//bTWrFmTr/0FAAAFW61atfTaa6+pUqVKCgsLk4uLi0qUKKFBgwapUqVKGjdunM6fP69ffvlFkjRjxgzVqVNHkydPVlBQkOrUqaPZs2dr06ZN+uOPP3T06FEZhqHKlStbXKdEiRLmx9/Gjh2bYSw3b95UeHi4EhIS9Pjjj+d53wEAAGD7yFfvL1ad4/bHH3+0WI+IiJCPj4/27t2rJk2aKCEhQbNmzVJkZKT5BzhnzhxVqVJFO3fu1COPPKK1a9fqt99+0/r16+Xr66vatWvrjTfe0NixYzVhwgQ5OTnp008/VWBgoKZMmSJJqlKlin766Sd9+OGHCgkJyfd+AwCAgqlmzZrm/7e3t1fx4sVVo0YN8zZfX19J0rlz5yRJBw4c0KZNmzKcf+zYsWPmR8TutGvXLqWmpqpnz55KSkqy2Dd27Fi99tprunHjhtzd3fX222+rTZs2Oe4bAAAA7n/kq/cXm3o5WUJCgiSZf+h79+7VrVu3FBwcbG4TFBSksmXLKioqSo888oiioqJUo0YN8wdLkkJCQvTss8/q0KFDqlOnjqKioizOkdZm1KhRGcaRlJRk8aFKTEzMrS4CAIACzNHR0WLdZDJZbEt7TCw1NVWSdOXKFbVt21bvvPNOunOVKlVKN27ckMlkSjeVVPny5SVl/IKGMWPGqF+/fuY5yO71xmEAAAAUHuSr9xebKdympqZq1KhRatSokapXry5JOnv2rJycnOTt7W3R1tfXV2fPnjW3ub1om7Y/bd/d2iQmJur69evpPkTh4eGaOHFirvUNAAAgIw899JCWLFmigIAAOTikT8uKFCmiJ554QjNmzNCIESMynTfsdiVKlFDFihXzIlwAAAAUMuSr1mXVOW5vN2zYMB08eFALFy60digKCwtTQkKCeYmNjbV2SAAAoAAaNmyYLly4oO7du2v37t06duyY1qxZo/79+yslJUWS9Mknnyg5OVn16tXTokWLdPjwYR05ckRff/21fv/9d9nb21u5FwAAACioyFetyyZG3A4fPlwrVqzQ1q1bVbp0afN2Pz8/3bx5U5cuXbIYdRsfHy8/Pz9zm127dlmcLz4+3rwv7b9p225v4+npmeGQbWdnZzk7O+dK3wAAADLj7++v7du3a+zYsWrZsqWSkpJUrlw5tWrVSnZ2/36/XqFCBe3fv1+TJ09WWFiYTp06JWdnZ1WtWlUvvviihg4dauVeAAAAoKAiX7Uuk2EYhrUubhiGRowYoaVLl2rz5s2qVKmSxf6EhASVLFlSCxYsUMeOHSVJR44cUVBQkHmO29WrV+vJJ59UXFycfHx8JEmff/65xowZo3PnzsnZ2Vljx47VqlWr9Ouvv5rP3aNHD124cCHdC9IykpiYKC8vLyUkJMjT0zMX7wBw/zl49ESOz1G9YrlciATA/eLGjRuKiYlRYGCgXFxcrB0OCoC7faYKa95WWPstSUtWrc/R8R1bB9+7EQDkkC38HpXTvy+lgvt3Jvkqcltu5atWHXE7bNgwRUZG6vvvv5eHh4d5TlovLy+5urrKy8tLAwcO1OjRo1WsWDF5enpqxIgRatiwoR555BFJUsuWLVW1alX17t1b7777rs6ePavXXntNw4YNM4+afeaZZzRjxgy99NJLGjBggDZu3KhvvvlGK1eutFrfAQAAAAAAACAzVp3jdubMmUpISFCzZs1UqlQp87Jo0SJzmw8//FBPPvmkOnbsqCZNmsjPz0/fffedeb+9vb1WrFghe3t7NWzYUL169VKfPn00adIkc5vAwECtXLlS69atU61atTRlyhR9+eWXCgkJydf+AgAAAAAAAEBWWHXEbVZmaXBxcdHHH3+sjz/+ONM25cqV06pVq+56nmbNmmn//v3ZjhEAAABAerbw2C8AAEBBZtURtwAAAAAAAACA9Kw64hYAABQeVnwfKgoYPkvITbfCp+b4HI5ho3J8DgCA9ZFjILfk1meJEbcAACBPOTo6SpKuXbtm5UhQUNy8eVPSv+86AAAAyKm0nCItxwByKu13n7Tfhf4rRtwCAIA8ZW9vL29vb507d06S5ObmJpPJZOWocL9KTU3V33//LTc3Nzk4kMoCAICcc3BwkJubm/7++285OjrKzo5xjvhvDMPQtWvXdO7cOXl7e+d4oAHZLgAAyHN+fn6SZC7eAjlhZ2ensmXL8gUAAADIFSaTSaVKlVJMTIxOnMj5yzcBb29v8+9AOUHhFgAA5Lm0ZNjHx0e3bt2ydji4zzk5OTESBgAA5ConJydVqlSJ6RKQY46Ojrk2pReFWwD5bsOOfTk6vsWjD+VSJADym729PfOSAgAAwCbZ2dnJxcXF2mEAZgxVAAAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEO1g4AAAAAQOG0Ycc+a4cAAABgsxhxCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAACAlaWkpOj1119XYGCgXF1dVaFCBb3xxhsyDMPaoQEAAMBKHKwdAAAAAFDYvfPOO5o5c6bmzp2ratWqac+ePerfv7+8vLw0cuRIa4cHAAAAK6BwCwAAAFjZjh071L59e7Vp00aSFBAQoAULFmjXrl1WjgwAAADWwlQJAAAAgJU9+uij2rBhg/744w9J0oEDB/TTTz8pNDTUypEBAADAWhhxCwAAAFjZyy+/rMTERAUFBcne3l4pKSl666231LNnzwzbJyUlKSkpybyemJiYX6ECAAAgnzDiFgAAALCyb775RvPnz1dkZKT27dunuXPn6v3339fcuXMzbB8eHi4vLy/zUqZMmXyOGAAAAHmNwi0AAABgZWPGjNHLL7+sbt26qUaNGurdu7eef/55hYeHZ9g+LCxMCQkJ5iU2NjafIwYAAEBeY6oEAAAAwMquXbsmOzvLMRX29vZKTU3NsL2zs7OcnZ3zIzQAAABYCYVbAAAAwMratm2rt956S2XLllW1atW0f/9+ffDBBxowYIC1QwMAwOxW+NQcHe8YNipX4gAKCwq3AAAAgJVNnz5dr7/+uoYOHapz587J399fQ4YM0bhx46wdGgAAAKyEwi0AAABgZR4eHpo6daqmTp1q7VAAAABgI3g5GQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAAAAAgI1hjlsAAAAAAAAbt2HHPmuHACCfMeIWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEUbgEAAAAAAADAxlC4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMhVsAAAAAAAAAsDEO1g4AALJryar1OT5Hx9bBuRAJAAAAAABA3mDELQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAAAAAgI2hcAsAAAAAAAAANobCLQAAAAAAAADYGAq3AAAAAAAAAGBjrFq43bp1q9q2bSt/f3+ZTCYtW7bMYn+/fv1kMpksllatWlm0uXDhgnr27ClPT095e3tr4MCBunLlikWbX375RY899phcXFxUpkwZvfvuu3ndNQAAAAAAAAD4z6xauL169apq1aqljz/+ONM2rVq1UlxcnHlZsGCBxf6ePXvq0KFDWrdunVasWKGtW7dq8ODB5v2JiYlq2bKlypUrp7179+q9997ThAkT9Pnnn+dZvwAAAAAAAAAgJxysefHQ0FCFhobetY2zs7P8/Pwy3Hf48GH9+OOP2r17t+rVqydJmj59ulq3bq33339f/v7+mj9/vm7evKnZs2fLyclJ1apVU3R0tD744AOLAi8AAAAAAAAA2Aqbn+N28+bN8vHxUeXKlfXss8/q/Pnz5n1RUVHy9vY2F20lKTg4WHZ2dvr555/NbZo0aSInJydzm5CQEB05ckQXL17Mv44AAAAAAAAAQBZZdcTtvbRq1Ur/93//p8DAQB07dkyvvPKKQkNDFRUVJXt7e509e1Y+Pj4Wxzg4OKhYsWI6e/asJOns2bMKDAy0aOPr62veV7Ro0XTXTUpKUlJSknk9MTExt7sGAAAAAAAAAJmy6cJtt27dzP9fo0YN1axZUxUqVNDmzZvVokWLPLtueHi4Jk6cmGfnBwAAAAAAAIC7sfmpEm5Xvnx5lShRQkePHpUk+fn56dy5cxZtkpOTdeHCBfO8uH5+foqPj7dok7ae2dy5YWFhSkhIMC+xsbG53RUAAAAAAAAAyNR9Vbg9deqUzp8/r1KlSkmSGjZsqEuXLmnv3r3mNhs3blRqaqoaNGhgbrN161bdunXL3GbdunWqXLlyhtMkSP++EM3T09NiAQAAAAAAAID8YtXC7ZUrVxQdHa3o6GhJUkxMjKKjo3Xy5ElduXJFY8aM0c6dO3X8+HFt2LBB7du3V8WKFRUSEiJJqlKlilq1aqVBgwZp165d2r59u4YPH65u3brJ399fktSjRw85OTlp4MCBOnTokBYtWqRp06Zp9OjR1uo2AAAAAAAAANyVVQu3e/bsUZ06dVSnTh1J0ujRo1WnTh2NGzdO9vb2+uWXX9SuXTs9+OCDGjhwoOrWratt27bJ2dnZfI758+crKChILVq0UOvWrdW4cWN9/vnn5v1eXl5au3atYmJiVLduXb3wwgsaN26cBg8enO/9BQAAAAAAAICssOrLyZo1aybDMDLdv2bNmnueo1ixYoqMjLxrm5o1a2rbtm3Zjg8AAAAAAAAArOG+muMWAAAAAAAAAAoDCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYB2sHAAAAAADImQ079uX4HC0efSgXIgEAALmFwi2AQulW+NQcHe8YNipX4gAAAAAAAMgIUyUAAAAAAAAAgI2hcAsAAAAAAAAANobCLQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAAAAAgI2hcAsAAAAAAAAANobCLQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAAAAAgI2hcAsAAAAAAAAANobCLQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAAAAAgI2hcAsAAAAAAAAANobCLQAAAAAAAADYGAq3AAAAAAAAAGBjKNwCAAAAt/nmm2908+ZN8/qpU6eUmppqXr927Zreffdda4QGAACAQoTCLQAAAHCb7t2769KlS+b1qlWr6vjx4+b1y5cvKywsLP8DAwAAQKFC4RYAAAC4jWEYd10HAAAA8gOFWwAAAAAAAACwMRRuAQAAAAAAAMDGOGSn8blz5+Tj45Pp/uTkZO3bt0/169fPcWAAAACAtaxZs0ZeXl6SpNTUVG3YsEEHDx6UJIv5bwEAAIC8kq3CbalSpRQXF2cu3taoUUOrVq1SmTJlJEnnz59Xw4YNlZKSkvuRAgAAAPmkb9++FutDhgyxWDeZTPkZDgAAAAqhbBVu73wxw/Hjx3Xr1q27tgEAAADuJ6mpqdYOAQAAAMj9OW4ZfQAAAICCICkpSVevXrV2GAAAACikeDkZAAAAcJu///5boaGhcnd3l6enpx555BEdPXo0z697+vRp9erVS8WLF5erq6tq1KihPXv25Pl1AQAAYJuyNVWCyWTS5cuX5eLiIsMwZDKZdOXKFSUmJkqS+b8AAADA/Wrs2LGKjo7WpEmT5OLios8++0yDBg3Spk2b8uyaFy9eVKNGjdS8eXOtXr1aJUuW1J9//qmiRYvm2TUBAABg27I9x+2DDz5osV6nTh2LdaZKAAAAwP1s3bp1ioiIUEhIiCTpySefVJUqVZSUlCRnZ+c8ueY777yjMmXKaM6cOeZtgYGBeXItAAAA3B+yVbjNy1EGAAAAgC04c+aMatWqZV6vVKmSnJ2dFRcXp4CAgDy55vLlyxUSEqLOnTtry5YteuCBBzR06FANGjQow/ZJSUlKSkoyr/PkGwAAQMGTrcJt06ZN8yoOAAAAwGbY29unWzcMI8+u99dff2nmzJkaPXq0XnnlFe3evVsjR46Uk5OT+vbtm659eHi4Jk6cmGfxAAAAwPqyVbhNTk5WSkqKxSNi8fHx+vTTT3X16lW1a9dOjRs3zvUgAQAAgPySNj3Y7VOAXblyRXXq1JGd3f/e7XvhwoVcu2Zqaqrq1aunyZMnS5Lq1KmjgwcP6tNPP82wcBsWFqbRo0eb1xMTE1WmTJlciwcAAADWl63C7aBBg+Tk5KTPPvtMknT58mU9/PDDunHjhkqVKqUPP/xQ33//vVq3bp0nwQIAAAB57fZ5ZvNLqVKlVLVqVYttVapU0ZIlSzJs7+zsnGfz7QIAAMA2ZKtwu337ds2YMcO8Pm/ePKWkpOjPP/+Ul5eXxo4dq/fee4/CLQAAAO5bGY1wvVNKSkquXrNRo0Y6cuSIxbY//vhD5cqVy9XrAAAA4P5hd+8m/3P69GlVqlTJvL5hwwZ17NhRXl5ekv5Ncg8dOpS7EQIAAAA24o8//tDYsWNVunTpXD3v888/r507d2ry5Mk6evSoIiMj9fnnn2vYsGG5eh0AAADcP7JVuHVxcdH169fN6zt37lSDBg0s9l+5ciX3ogMAAACs7Nq1a5ozZ44ee+wxVa1aVVu2bLGYXzY3PPzww1q6dKkWLFig6tWr64033tDUqVPVs2fPXL0OAAAA7h/Zmiqhdu3a+uqrrxQeHq5t27YpPj5ejz/+uHn/sWPH5O/vn+tBAgAAAPlt586d+vLLL7V48WKVLVtWhw8f1qZNm/TYY4/lyfWefPJJPfnkk3lybgAAANx/sjXidty4cZo2bZoqVKigkJAQ9evXT6VKlTLvX7p0qRo1apTrQQIAAAD5ZcqUKapWrZo6deqkokWLauvWrfr1119lMplUvHhxa4cHAACAQiJbI26bNm2qvXv3au3atfLz81Pnzp0t9teuXVv169fP1QABAACA/DR27FiNHTtWkyZNkr29vbXDAQAAQCGVrcKtJFWpUkVVqlTJcN/gwYNzHBAAAABgTW+88YbmzJmjr776St27d1fv3r1VvXp1a4cFAACAQiZbhdutW7dmqV2TJk3+UzAAAACAtYWFhSksLExbtmzR7Nmz1aBBA1WsWFGGYejixYvWDg8AAACFRLYKt82aNZPJZJIkGYaRYRuTyaSUlJScRwYAAABYUdOmTdW0aVPNmDFDkZGRmj17tpo2bar69eurU6dOGj16tLVDBAAAQAGWrZeTFS1aVGXKlNHrr7+uP//8UxcvXky3XLhwIa9iBQAAAPKdh4eHhgwZop9//ln79+9X/fr19fbbb1s7LAAAABRw2SrcxsXF6Z133lFUVJRq1KihgQMHaseOHfL09JSXl5d5AQAAAAqiGjVqaOrUqTp9+rS1QwEAAEABl62pEpycnNS1a1d17dpVJ0+eVEREhIYPH66kpCT17dtXEydOlINDtt93BgAAANiMefPm3bONyWRS79698yEaAAAAFFb/ucpatmxZjRs3Tr1799bAgQP19ttv64UXXlCxYsVyMz4AAAAgX/Xr10/u7u5ycHC463sdKNwClm6FT83xORzDRuX4HAAAFBTZmiohTVJSkiIjIxUcHKzq1aurRIkSWrlyJUVbAAAA3PeqVKkiJycn9enTR1u2bOG9DgAAALCKbBVud+3apWeffVZ+fn5677331K5dO8XGxuqbb75Rq1at8ipGAAAAIN8cOnRIK1eu1PXr19WkSRPVq1dPM2fOVGJiorVDAwAAQCGSrakSHnnkEZUtW1YjR45U3bp1JUk//fRTunbt2rXLnegAAAAAK2jQoIEaNGigqVOnavHixZozZ45efPFFPfXUU5o9e7acnZ2tHSIAAAAKuGzPcXvy5Em98cYbme43mUxKSUnJUVAAAACALXB1dVWfPn0UEBCg8ePHa+HChZoxYwaFWwAAAOS5bE2VkJqaes/l8uXLeRUrAAAAkG9Onz6tyZMnq1KlSurWrZsefvhhHTp0SEWLFrV2aAAAACgE/tPLyTKSlJSkDz74QOXLl8+tUwIAAAD57ptvvlFoaKgqVaqk3bt3a8qUKYqNjdW7776roKAga4cHAACAQiJbUyUkJSVpwoQJWrdunZycnPTSSy+Z5/l67bXXZG9vr+effz6vYgUAAADyXLdu3VS2bFk9//zz8vX11fHjx/Xxxx+nazdy5EgrRAcAAIDCIluF23Hjxumzzz5TcHCwduzYoc6dO6t///7auXOnPvjgA3Xu3Fn29vZ5FSsAAACQ58qWLSuTyaTIyMhM25hMJgq3AAAAyFPZKtwuXrxY8+bNU7t27XTw4EHVrFlTycnJOnDggEwmU17FCAAAAOSb48ePWzsEFEIHj56wdggAAMDGZGuO21OnTqlu3bqSpOrVq8vZ2VnPP/88RVsAAAAUGFFRUVqxYoXFtnnz5ikwMFA+Pj4aPHiwkpKSrBQdAAAACotsFW5TUlLk5ORkXndwcJC7u3uuBwUAAABYy8SJE3Xo0CHz+q+//qqBAwcqODhYL7/8sn744QeFh4dbMUIAAAAUBtmaKsEwDPXr10/Ozs6SpBs3buiZZ55RkSJFLNp99913uRchAAAAkI8OHDigN99807y+cOFCNWjQQF988YUkqUyZMho/frwmTJhgpQgBAABQGGSrcNu3b1+L9V69euVqMAAAAIC1Xbx4Ub6+vub1LVu2KDQ01Lz+8MMPKzY21hqhAQAAoBDJVuF2zpw5eRUHAAAAYBN8fX0VExOjMmXK6ObNm9q3b58mTpxo3n/58mU5OjpaMUIAAAAUBtma4xYAAAAo6Fq3bq2XX35Z27ZtU1hYmNzc3PTYY4+Z9//yyy+qUKGCFSMEAABAYZCtEbcAAABAQffGG2/o//7v/9S0aVO5u7tr7ty5Fi/onT17tlq2bGnFCAEAAFAYULgFAAAAblOiRAlt3bpVCQkJcnd3l729vcX+xYsXy93d3UrRAQAAoLCgcAsAAABkwMvLK8PtxYoVy+dIAAAAUBgxxy0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGl5MBAAAAAAqMg0dP5Oj46hXL5VIkAADkDCNuAQAAAAAAAMDGULgFAAAAAAAAABtD4RYAAAAAAAAAbAyFWwAAAAAAAACwMRRuAQAAAAAAAMDGULgFAAAAAAAAABtD4RYAAAAAAAAAbIxVC7dbt25V27Zt5e/vL5PJpGXLllnsNwxD48aNU6lSpeTq6qrg4GD9+eefFm0uXLignj17ytPTU97e3ho4cKCuXLli0eaXX37RY489JhcXF5UpU0bvvvtuXncNAAAAAAAAAP4zB2te/OrVq6pVq5YGDBig//u//0u3/91339VHH32kuXPnKjAwUK+//rpCQkL022+/ycXFRZLUs2dPxcXFad26dbp165b69++vwYMHKzIyUpKUmJioli1bKjg4WJ9++ql+/fVXDRgwQN7e3ho8eHC+9hf3rw079uX4HC0efSgXIgEAAAAAAEBhYNXCbWhoqEJDQzPcZxiGpk6dqtdee03t27eXJM2bN0++vr5atmyZunXrpsOHD+vHH3/U7t27Va9ePUnS9OnT1bp1a73//vvy9/fX/PnzdfPmTc2ePVtOTk6qVq2aoqOj9cEHH1C4BQAAAAAAAGCTbHaO25iYGJ09e1bBwcHmbV5eXmrQoIGioqIkSVFRUfL29jYXbSUpODhYdnZ2+vnnn81tmjRpIicnJ3ObkJAQHTlyRBcvXszw2klJSUpMTLRYAAAAAAAAACC/2Gzh9uzZs5IkX19fi+2+vr7mfWfPnpWPj4/FfgcHBxUrVsyiTUbnuP0adwoPD5eXl5d5KVOmTM47BAAAAAAAAABZZLOFW2sKCwtTQkKCeYmNjbV2SAAAAAAAAAAKEZst3Pr5+UmS4uPjLbbHx8eb9/n5+encuXMW+5OTk3XhwgWLNhmd4/Zr3MnZ2Vmenp4WCwAAAAAAAADkF5st3AYGBsrPz08bNmwwb0tMTNTPP/+shg0bSpIaNmyoS5cuae/eveY2GzduVGpqqho0aGBus3XrVt26dcvcZt26dapcubKKFi2aT70BAAAAAAAAgKyzauH2ypUrio6OVnR0tKR/X0gWHR2tkydPymQyadSoUXrzzTe1fPly/frrr+rTp4/8/f311FNPSZKqVKmiVq1aadCgQdq1a5e2b9+u4cOHq1u3bvL395ck9ejRQ05OTho4cKAOHTqkRYsWadq0aRo9erSVeg0AAAAAAAAAd+dgzYvv2bNHzZs3N6+nFVP79u2riIgIvfTSS7p69aoGDx6sS5cuqXHjxvrxxx/l4uJiPmb+/PkaPny4WrRoITs7O3Xs2FEfffSReb+Xl5fWrl2rYcOGqW7duipRooTGjRunwYMH519HAQAAAAAAACAbrFq4bdasmQzDyHS/yWTSpEmTNGnSpEzbFCtWTJGRkXe9Ts2aNbVt27b/HCcAAAAAAAAA5CebneMWAAAAAAAAAAorCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAACADXn77bdlMpk0atQoa4cCAAAAK6JwCwAAANiI3bt367PPPlPNmjWtHQoAAACsjMItAAAAYAOuXLminj176osvvlDRokWtHQ4AAACsjMItAAAAYAOGDRumNm3aKDg42NqhAAAAwAY4WDsAAAAAoLBbuHCh9u3bp927d2epfVJSkpKSkszriYmJeRUaAAAArIQRtwAAAIAVxcbG6rnnntP8+fPl4uKSpWPCw8Pl5eVlXsqUKZPHUQIAACC/UbgFAAAArGjv3r06d+6cHnroITk4OMjBwUFbtmzRRx99JAcHB6WkpKQ7JiwsTAkJCeYlNjbWCpEDAAAgLzFVAgAAAGBFLVq00K+//mqxrX///goKCtLYsWNlb2+f7hhnZ2c5OzvnV4gAAACwAgq3AAAAgBV5eHioevXqFtuKFCmi4sWLp9sOAACAwoOpEgAAAAAAAADAxjDiFgAAALAxmzdvtnYIAAAAsDIKtwAAAAAA/H8bduzL8TlaPPpQLkQCACjsKNyiUDh49IS1QwAAAAAAAACyjDluAQAAAAAAAMDGULgFAAAAAAAAABvDVAnAfeRW+NQcHe8YNipX4gAAAAAAAMipJavW5/gcHVsH50IktokRtwAAAAAAAABgYyjcAgAAAAAAAICNYaoE3BVD1gEAAAAAAID8R+EWAAAAAAAAKGQOHj2R43NUr1guFyJBZpgqAQAAAAAAAABsDIVbAAAAAAAAALAxFG4BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsjIO1AwAAALjTklXrc3R8x9bBuRQJAAAAAFgHI24BAAAAAAAAwMZQuAUAAAAAAAAAG0PhFgAAAAAAAABsDIVbAAAAAAAAALAxFG4BAAAAAAAAwMY4WDsAFHy3wqfm6HjHsFG5EgcAoPDI6b89Ev/+AAAAALAuRtwCAAAAAAAAgI2hcAsAAAAAAAAANoapEmzYwaMncnR89YrlcikSAAAAAAXdklXrc3R8u1yKAwBsXU7/vpSkjq2DcyESFHQUbgEAgFlOvzSU+OIQAAAAAHIDhdsCbMOOfdYOAQBQCPHvDwAAAFA4kPvnLea4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMUyUAAAAAAAAAuC/dCp+a43M4ho3K8TnyAiNuAQAAAAAAAMDGMOIWAAAAAIBctGTV+hyfo2Pr4FyIBABwP2PELQAAAAAAAADYGEbcAgAAAAAAoNDYsGOftUMAsoQRtwAAAAAAAABgYyjcAgAAAAAAAICNYaoEAAAAAABQYN0Kn5qj4x3DRuVKHACQXYy4BQAAAAAAAAAbQ+EWAAAAAAAAAGwMUyUAAAAAAAAA+SinU3hITONRGDDiFgAAAAAAAABsDCNuAQAAAACwMbxQC8jYwaMnrB0CkG8o3AL5ZMmq9Tk+R7tciAMAAAAAAAC2j6kSAAAAAAAAAMDGULgFAAAAAAAAABtD4RYAAAAAAAAAbAyFWwAAAAAAAACwMRRuAQAAAAAAAMDGOFg7AAAAAAAAgIwsWbU+x+dolwtxAIA1MOIWAAAAAAAAAGwMhVsAAAAAAAAAsDFMlQAAAAAAANI5ePREjo6PP3c+lyIBgMKJEbcAAAAAAAAAYGMo3AIAAAAAAACAjbHpwu2ECRNkMpkslqCgIPP+GzduaNiwYSpevLjc3d3VsWNHxcfHW5zj5MmTatOmjdzc3OTj46MxY8YoOTk5v7sCAAAAAAAAAFlm83PcVqtWTevXrzevOzj8L+Tnn39eK1eu1OLFi+Xl5aXhw4fr//7v/7R9+3ZJUkpKitq0aSM/Pz/t2LFDcXFx6tOnjxwdHTV58uR87wsAAAAAAAAAZIXNF24dHBzk5+eXbntCQoJmzZqlyMhIPf7445KkOXPmqEqVKtq5c6ceeeQRrV27Vr/99pvWr18vX19f1a5dW2+88YbGjh2rCRMmyMnJKb+7AwAAAAAAAAD3ZNNTJUjSn3/+KX9/f5UvX149e/bUyZMnJUl79+7VrVu3FBwcbG4bFBSksmXLKioqSpIUFRWlGjVqyNfX19wmJCREiYmJOnToUP52BAAAAAAAAACyyKZH3DZo0EARERGqXLmy4uLiNHHiRD322GM6ePCgzp49KycnJ3l7e1sc4+vrq7Nnz0qSzp49a1G0Tdufti8zSUlJSkpKMq8nJibmUo8AAAAAAAAA4N5sunAbGhpq/v+aNWuqQYMGKleunL755hu5urrm2XXDw8M1ceLEPDs/AAAAAAAAANyNzU+VcDtvb289+OCDOnr0qPz8/HTz5k1dunTJok18fLx5Tlw/Pz/Fx8en25+2LzNhYWFKSEgwL7GxsbnbEQAAAAAAAAC4i/uqcHvlyhUdO3ZMpUqVUt26deXo6KgNGzaY9x85ckQnT55Uw4YNJUkNGzbUr7/+qnPnzpnbrFu3Tp6enqpatWqm13F2dpanp6fFAgAAAAAAAAD5xaanSnjxxRfVtm1blStXTmfOnNH48eNlb2+v7t27y8vLSwMHDtTo0aNVrFgxeXp6asSIEWrYsKEeeeQRSVLLli1VtWpV9e7dW++++67Onj2r1157TcOGDZOzs7OVewcAAAAAAAAAGbPpwu2pU6fUvXt3nT9/XiVLllTjxo21c+dOlSxZUpL04Ycfys7OTh07dlRSUpJCQkL0ySefmI+3t7fXihUr9Oyzz6phw4YqUqSI+vbtq0mTJlmrSwAAAAAAAABwTzZduF24cOFd97u4uOjjjz/Wxx9/nGmbcuXKadWqVbkdGgAAAAAAAADkmftqjlsAAAAAAAAAKAwo3AIAAAAAAACAjaFwCwAAAAAAAAA2hsItAAAAAAAAANgYCrcAAACAlYWHh+vhhx+Wh4eHfHx89NRTT+nIkSPWDgsAAABWROEWAAAAsLItW7Zo2LBh2rlzp9atW6dbt26pZcuWunr1qrVDAwAAgJU4WDsAAAAAoLD78ccfLdYjIiLk4+OjvXv3qkmTJlaKCgAAANZE4RYAAACwMQkJCZKkYsWKZbg/KSlJSUlJ5vXExMR8iQsAAAD5h6kSAAAAABuSmpqqUaNGqVGjRqpevXqGbcLDw+Xl5WVeypQpk89RAgAAIK9RuAUAAABsyLBhw3Tw4EEtXLgw0zZhYWFKSEgwL7GxsfkYIQAAAPIDUyUAAAAANmL48OFasWKFtm7dqtKlS2faztnZWc7OzvkYGQAAAPIbhVsAAADAygzD0IgRI7R06VJt3rxZgYGB1g4JAAAAVkbhFgAAALCyYcOGKTIyUt9//708PDx09uxZSZKXl5dcXV2tHB0AAACsgTluAQAAACubOXOmEhIS1KxZM5UqVcq8LFq0yNqhAQAAwEoYcQsAAABYmWEY1g4BAAAANoYRtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYxysHQAAFFYHj57I8TmqVyyXC5HkzK3wqTk+h2PYqByfAwAAAACAgoTCLQDcxzbs2Jej4y9dupDjGNrl+AwAAAAAAOBOTJUAAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADbGwdoBAAAAIO8sWbU+R8d3bB2cS5EAAAAAyA5G3AIAAAAAAACAjWHELQAAQB44ePREjs9RvWK5XIgEAAAAwP2IEbcAAAAAAAAAYGMo3AIAAAAAAACAjaFwCwAAAAAAAAA2hjluAQAAbNSGHfusHYJuhU/N8Tkcw0bl+BwAAABAYcOIWwAAAAAAAACwMRRuAQAAAAAAAMDGULgFAAAAAAAAABtD4RYAAAAAAAAAbAyFWwAAAAAAAACwMRRuAQAAAAAAAMDGULgFAAAAAAAAABtD4RYAAAAAAAAAbAyFWwAAAAAAAACwMQ7WDgAAgJw6ePREjs9RvWK5HB2/ZNX6HMfQsXVwjs8BAAAAACgYGHELAAAAAAAAADaGwi0AAAAAAAAA2BimSgAAQNKGHfusHYJuhU/N8Tkcw0bl+BwAAAAAAOtjxC0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNoXALAAAAAAAAADaGwi0AAAAAAAAA2BgKtwAAAAAAAABgYyjcAgAAAAAAAICNKVSF248//lgBAQFycXFRgwYNtGvXLmuHBAAAAJiRrwIAACBNoSncLlq0SKNHj9b48eO1b98+1apVSyEhITp37py1QwMAAADIVwEAAGCh0BRuP/jgAw0aNEj9+/dX1apV9emnn8rNzU2zZ8+2dmgAAAAA+SoAAAAsOFg7gPxw8+ZN7d27V2FhYeZtdnZ2Cg4OVlRUVLr2SUlJSkpKMq8nJCRIkhITE/M+2NtcuXw5R8dfvXolxzFcu3Y1x+dIvHEjR8c75sJ9517+yxbupZTz+8m9/J+CcC+lnN9P7uX/FIR7KeX8fnIv/6cg3MvsSMvXDMPIt2vmhvsxX+Uz/j+2kBdwL/+nINxLqWDkq9zL/7GFvzO5l/9TEO6lVDDy1YJyL7MjW/mqUQicPn3akGTs2LHDYvuYMWOM+vXrp2s/fvx4QxILCwsLCwsLC8t9usTGxuZXqpkryFdZWFhYWFhYWArXkpV8tVCMuM2usLAwjR492ryempqqCxcuqHjx4jKZTFaMLH8lJiaqTJkyio2Nlaenp7XDua9xL3MP9zL3cC9zD/cyd3E/c09hvJeGYejy5cvy9/e3dih5inz1X4XxM55XuJe5h3uZe7iXuYd7mXu4l7mnsN7L7OSrhaJwW6JECdnb2ys+Pt5ie3x8vPz8/NK1d3Z2lrOzs8U2b2/vvAzRpnl6ehaqP0B5iXuZe7iXuYd7mXu4l7mL+5l7Ctu99PLysnYI2Ua+mjOF7TOel7iXuYd7mXu4l7mHe5l7uJe5pzDey6zmq4Xi5WROTk76f+3deXxM994H8M+ZSUIixFLLtUVoSDVUbLVzbU0k5RFrH1qXR90WJVUPLSqWXlpqad1qEhXEvi+3dMPjXnpbtadqSYTUVhTZmpBkMt/njzRHRvQW/c2cmeTz/ivOnMnrl48zM5/5na1Zs2bYs2ePvsxqtWLPnj1o3bq1gSMjIiIiImJfJSIiIqKiSsQRtwAwbtw4DBkyBM2bN0fLli2xcOFCZGZmYujQoUYPjYiIiIiIfZWIiIiIbJSYidsBAwbg559/xtSpU3Ht2jU0adIEn3/+OapWrWr00JxWqVKlEBkZWeQ0PHp0zFIdZqkOs1SHWarFPNVhlq6FffXRcRtXh1mqwyzVYZbqMEt1mKU6zPL3aSIiRg+CiIiIiIiIiIiIiO4pEde4JSIiIiIiIiIiInIlnLglIiIiIiIiIiIicjKcuCUiIiIiIiIiIiJyMpy4JSIiIiIiIiIiInIynLglIiIiIiIiIiIicjKcuCUiIiIiIiIiIiJyMpy4JaIST0SMHgLRf8RtlIiIqORiDyBXwO2UyD7cjB4AFV8ZGRkoW7as0cMoFi5evIjU1FRUrFgRlSpVgqenp9FDcmnHjx9HcnIyPD09ERgYiBo1ahg9pGIjMzMTZcqUMXoYxcL58+dx584dmEwmPPXUU0YPp0TJy8uD2Ww2ehgu7eeff4bZbEZ2djb+9Kc/GT0cot/EvqoO+6o67Kr2xb6qDvuqcdhX/zhX6as84pbsYtOmTXj11Vdx4sQJo4fi8uLi4tCjRw+Ehoaibdu2mDNnDlJSUowelsuKjY1F7969MWnSJIwcORKRkZFITU01eljFwrp16zB27FicP3/e6KG4vJUrV+L5559Hz5490ahRI7zzzjsAeCSDPa1duxazZs0CAJjNZlitVoNH5LrWrFmDAQMGoEWLFujVqxfWrVtn9JCIHoh9VR32VXXYVe2LfVUd9lXHY19Vx6X6qhAptmPHDvHw8JDq1avLiBEjJD4+3ughuay1a9eKj4+PxMbGyvfffy+RkZFSr149OXbsmNFDc0mrVq2SsmXLytq1ayUlJUUWLVokvr6+cuPGDX0dq9Vq4Ahd144dO8TT01M0TZNBgwbJjz/+aPSQXNbatWvF29tbVq1aJfHx8bJixQrRNE327Nlj9NCKra1bt4qmaaJpmkRGRurL8/LyjBuUi4qLixMvLy+JioqS6Ohoef311yUgIEBOnjxp9NCIbLCvqsO+qg67qn2xr6rDvup47KvquFpf5cQtKXX9+nUJCwuTCRMmyMcffyxNmzaVYcOGsQw/hsTERGnbtq0sXLjQZnmzZs1k/PjxBo3KdZ09e1aCgoIkKipKX5aamirdu3eXJUuWyPr16/U3ahbiR3Pjxg0ZPHiwTJkyRfbu3Suenp4yYMAAluHHkJiYKO3atZPFixeLyL1tsXv37jJx4kSbZaRGYmKiBAcHy9ixY2XOnDlSvnx5mTJliv44y/DDO3bsmAQGBsqKFSv0ZSdOnJC6devKP/7xDwNHRmSLfVUd9lV12FXti31VHfZVx2NfVccV+yqvcUtKPfHEExg8eDAqV66Mzp07w9vbGwsWLMDChQsxduxYNG7c2GZ9EYGmaQaN1rmlp6ejVq1a6Ny5MwDAYrHAzc0NgYGByMrKMnh0rqdMmTKYMGEC2rZtqy976aWXcPjwYaSmpsLDwwNnz57Fzp070aJFCwNH6npKlSqFkJAQVK9eHZ06dcL+/fvRvn17AMCcOXNQu3Ztg0foOnJzc2GxWNCyZUsA0N8fq1SpgkuXLhk5tGLL29sbTz/9NAYOHIiAgAB4eHhg2rRpAICZM2fCZDLBarXCZOLVpX5Pamoq/Pz88Oyzz+rLGjduDF9fX5w9exZhYWG8Hhs5BfZVddhX1WFXtS/2VXXYVx2PfVUdl+yrRs8cU/GTm5tr8++4uDhp2rSpDB06VD+S4ebNm3LmzBkjhudSvv32W/1ni8UiIiJvv/22jBkzxma969evO3Rcrio9PV3/+e233xZfX185efKk5OTkyPnz56VTp04yevRosVgs3Ev8iDIzM23+ffjw4SJHMqSmpsqBAweMGJ5LOX36tP5zTk6OiIi88cYbMmLECJv1rl696tBxFWepqan6zzdv3pQFCxYUOZIhNTVVbt26ZcTwXMbly5cf+LnVsWNHeffdd40aFtEDsa+qw76qDruqfbGvqsO+6njsq2q4Yl/ldDwp5+aWfyB3Xl4eAODFF19EREQE4uPj8cEHH2Dfvn3o3bs3xowZY+QwnVrBRcYL9gKJiL7HJyUlBT/99JO+vH///vjggw+MGaiLKXzX6EmTJuHgwYN4+umn4e7uDj8/P7i7u8NqtcJsNvPImkfk5eUFIH+bzMvLQ7NmzfCvf/0LO3bswJtvvokjR46gZ8+eWLx4scEjdV7y640cAgICAEDfFgukpaXp6z3//PNYuXKl4wdZTPn4+ADIz7ZSpUoYMmQIpk2bho8++ghTp06F1WpFcHAwli5davBInVuNGjVsPrcK3kc9PDyQm5urL+/cuTN27Nhh2DiJAPZVFdhX1WNXtS/21T+OfdU47KtquGJf5aUSyG7MZrP+QnjxxRdhMpkwb948rF69Gv7+/tizZ4/RQ3Ra95/iULiYmUwmlC5dGgAQGhqK06dPY/Xq1Q4dn6vLy8tD6dKl9RwB6Hc+btiwoVHDKhY0TYPZbIbFYkHz5s1x4MABdOzYEZs2bULdunWxe/duo4fotO7/Alb4fSA3N1cvEmFhYfj++++xZcsWh46vJCj4P6hQoQJeeuklmEwmTJ8+HYsXL0bZsmURERFh7ABdiKZpep6enp769hwaGorExESEhIQYOTwiHfvq42NftR92VftiX3187KvGY19Vx1X6Ko+4JbvSNE3fK9e3b1+kpqaiSZMmOHr0KNzd3WGxWAweoesoOKqhatWq8Pb2Rq9evXDu3DkkJCQwy0dUeK+wiODWrVsYMmQI0tPT8de//tXAkRUfbm5uEBE0bdoUdevWxbPPPouTJ09yW31EBUeCeXt7o2zZsujXrx8SExORlJTELO2sQoUK6NatG9zd3REQEIDExERm/ogKPv9FBBaLBS+88AISExNx/vx5ZklOhX1VHfZVNdhVHYN9VQ32VeOwr/5xrtBXOXFLj61gA/89mqbhzp076NChAywWC/bv3w83Nzf95gX0cFkW7P1JS0tDVFQUfvzxR/zwww/6mwmzvOdht83s7GysWbMGL774Iq5evapvmwXlg4p62GwBICcnB+Hh4bhx4wb27t3L1/19HuV1n5eXh7i4OFy8eJGvewfJyMjAhAkT4O3tjX379nH7/QNycnIwbdo0nDlzBqdOneL2Sw7FvqoO+6o67Kr2xb6qDvuqc2NfVceZ+yonbumR3bx5E0DR0yT+E09PT0yfPh1JSUl8MynkcbL08/PDc889h8OHDzvVm4kzeNQ8S5UqBS8vL3Tq1AnffvutnqdT3UHSSfz888+wWCyPtK2aTCaMHj0aly5d4rZayKNkWbBO06ZN0bBhQ3z99dfM8hEV/sLxKF/k7ty5g6CgIJw6dYqfW7961CwLtt8WLVqgQ4cOOHToELdfchj2VXXYV9VhV7Uv9lV12Fcdi31VneLYVzV5lK2CSrzt27dj2bJlCA8Px0svvfRQzynYxApeEM70AjDS42RZwGq1wmQyMctC/kieBfLy8liEH2Dbtm3YvHkz2rRpgxEjRjxWRtxW8/2RLAuuwcgs/7jCNyJ4GLm5uXB3d7fjiFzXw2SZnJyM2rVrw2QyMUtyCPZVddhX1WFXtS/2VXXYV50D+6o6rt5XecQtPbTY2Fi8/PLLaNasGapUqfJIzy14keTk5PANHI+fZcHF3k0mE+7cucMsf/W4eRY+zSw7O5tF+AFiY2MxfPhwPPnkkwgKCnrojAqucVeA2+rjZ1lwXaWC91Fm+fCOHTuGTz75BP3798eIESPw2Wef4datWzbXs3yQ+7dfZypuRnncLAGgTp06+mmUzJLsjX1VHfZVddhV7Yt9VR32VcdjX1Wn2PZVIXoIW7ZskQoVKsjGjRvFarX+5nr3P1b43ytXrpTY2FjJzc212zhdgYosV61axSx/xTztZ9u2bVKuXDnZsGHDb2bzoMwLL1u8eLGsWLHCbmN0FczS8VauXCmBgYHSpUsX6dixo/j7+0uZMmVk8ODBcuHCBRH5/czXr18vO3fudNSQnRazJFfBvqoO+5U6zNK+2LHUYZaOx46lTnHOkhO39LtycnJk+PDh8vbbb9ssP3v2rKxfv17ef/99OXHihOTk5Ng8XvgFEB0dLZqmya5duxwyZmfFLNVinvZhtVolOztbhg4dKpMnT7Z5LCEhQVavXi0zZ86UU6dOPfC5BWJiYsRkMsnGjRvtPmZnxSyNER0dLV5eXhIbGys//fSTvnz8+PHi6+sr/fv3l6tXrxZ5XuHMP/74Y/Hx8ZEvv/zSIWN2VsySXAU7gTrMUh1maT/sWOowS2OwY6lT3LPkxC39ruzsbGnatKmMHDlSXzZ79mx57rnnxMfHRypUqCB169aVzz//XH+88AsgKipKfHx8ZPPmzQ4dtzNilmoxT/vq1q2bvPrqq/q/Z82aJSEhIeLt7S1169YVT09P2b9/v4jk53p/tuXKlZMtW7Y4fNzOiFk6TlxcnGiaZlO6Cuc5efJk8fHxkejoaJvHHvTeUNK/eDBLciXsBOowS3WYpf2xY6nDLB2HHUudkpAlJ27pgQpvzHl5eRIZGSlBQUHy1ltvSefOneXJJ5+UyMhIfa9b48aNpVevXkV+T8Eb+KZNmxw5fKfCLNVinvYVHx8v6enpkpubK6+88oq0bt1a3njjDenUqZP4+/vL9OnTJT4+XvLy8qRHjx7SqlUrsVgskpeXp/+O6OhoZivM0ghJSUlSuXJl6d69u9y+fdvmscK5tmvXTtq0afPA38HM8zFLcgXsBOowS3WYpf2xY6nDLB2PHUudkpIlJ27poRw/flxGjx4tzz77rPTo0UNOnjwpGRkZ+uNjx46V8PBwmxfHRx99JD4+Pk79AjACs1SLearz97//XcxmsyQkJIiIyM2bN2XQoEESHBwsPXr0kPj4eD3bvLw8GT16tISHh9v8jgULFki5cuVK/JEhzNI4c+bMkZYtW8rYsWP161kVyM7OFhGRefPmiZ+fn9y4ccNmb/vf//538fb2Zua/YpbkatgJ1GGW6jBLtdix1GGWxmHHUqckZMlb/VERx48fx8mTJ7F79254e3tj8ODBaNq0KRYtWoTc3Nwid9jLyspCfHw8WrVqBZPJBBFBVlYWNm/ejCVLlqBPnz4G/SXGY5ZqMU/7iY6OxhtvvIF169bB398fIoJKlSohNjYWJpOpyJ1hc3JykJCQgKCgIJtlhw8fxscff4zw8HBH/wlOg1kaw2q1wmQy4X//939hNpuxatUqAEBERATq1KkDEYGHhwcAICkpCYGBgahcubL+/HPnzmHLli1YunRpic+cWZIrYCdQh1mqwyztix1LHWZpDHYsdUpUlkbMFpPzWrVqlTRu3Fg6duwoTZs2lTp16oibm5u89tprkpiYKCL3DjnPzs6WS5cuSUhIiDRr1ky/82TB43fu3DHmj3ASzFIt5mk/MTExYjabZevWrTbLT5w4of9ckF1OTo6cP39eQkJCpEmTJnq2BXsuS/rdjpmlsQofrTRv3jwJCgoqsvf96tWr0q1bN5kzZ06R5168eNFRQ3V6zJKcGTuBOsxSHWZpX+xY6jBLY7FjqVNSsuTELemio6OlTJkyEhsbK1euXBGR/I05IiJCfHx85JVXXtHv0JeamiozZsyQP//5z9KuXTv9TqgWi8Ww8TsTZqkW87SftWvXiqZpsn37dpvlwcHBEhgYKHfv3tWX3b59W8aPHy/du3eX9u3bM9v7MEvHK8ircG4PKnBjxoyRy5cvi4hIaGiotGnTxuY5hU+ZKqmYJbkKdgJ1mKU6zNK+2LHUYZaOx46lTknNkpdKIADAihUr8Oqrr+LTTz9FSEgIRAQAYDKZsGDBAri5ueHDDz9E586d0a9fP1y6dAlubm4IDQ1FREQEzGYzLBZLkVMqSiJmqRbztK/r168DAG7fvq0v69u3L65du4YdO3agVKlS+nKz2YzSpUuje/fuzPYBmKVj7dmzB/Pnz8f69evh7e2NvLw8mM1mmEwm/dSpcePGAQBWr14Ns9mM7777Djdv3sT3338Ps9msP0fTNIP/GmMxS3IV7ATqMEt1mKX9sWOpwywdix1LnRKdpZGzxuQckpKSRNM06d27t82F8kVs92S0bt1a2rVrp+/RKLw3jnvd8jFLtZin/RQ+rem9994Tk8kky5Ytk/79+0tgYKAkJyeLyL29kXl5efLLL7/Y/A5mm49ZOp7VapVPPvlEGjVqJP369dPzLJxj4Z8XLlwoXl5e0rRpU/1oEZ7al49ZkqtgJ1CHWarDLO2LHUsdZul47FjqlPQsOXFLIiIye/Zs8fPzk8jISP2Q8gIFG/qUKVOkYcOGcuvWLSOG6DKYpVrMU71t27ZJWFiYpKSk6MtmzZolmqZJhQoVihQ3EZGQkBCZNWuWo4fq9JilcbKzs2XFihXSsmVL6d279wMLXMF6IiK7d+/WH3Pl4mYPzJJcBTuBOsxSHWZpH+xY6jBL47BjqVOSs+TEbQlVeE9agVmzZkmNGjVk2rRp+rWZCnv55ZclPDzcYWN0FcxSLeZpf99++62ULl1a+vfvL6mpqfryRYsWiaZpEhMTY1PcwsLCpHbt2vqXD7qHWRrr7t27snz5cmnRooVNgSt4/7h27Zp06dJFIiMj9efwaJEHY5bkjNgJ1GGW6jBLx2DHUodZGosdS52SmiUnbkuoW7duyY0bNyQxMVHS0tL05e+8845eOgrvMb5y5Yp07dpV5s+fb8RwnRqzVIt5OsahQ4ekUqVKEh4eblPgZs2aJSaTSWJiYkREpEePHlK/fv1icYqJvTBLx/jnP/8pixYtkgEDBsiECRPk008/FZH8MhYXF6cXuMzMTBHJf29o3769TeaUj1mSq2AnUIdZqsMsHYcdSx1m6RjsWOowy3s4cVsCrVu3Trp37y7Vq1cXd3d3adGihUydOlV/vKB0REZGyvXr10Uk/058bdu2LRZ7K1RilmoxT/u5ffu2zelRIiLffffdAwvc7NmzxcPDQ/70pz9Jw4YNWdzuwywdb8mSJVKtWjUJDg6W1q1bi5+fn2iaJiNHjpSrV6+KxWLRT53q06ePJCUlSdeuXZn5AzBLchXsBOowS3WYpX2xY6nDLB2PHUsdZmmLE7clzNKlS8XLy0sWLFggGzdulO3bt0u3bt3E09NT+vTpo6/3t7/9TWrVqiUzZsyQjh072uy1YOnIxyzVYp72s27dOunQoYM0adJEunXrZnNq328VuHfeeUdatmxZLD/4/ghm6XibN28Wb29v2bx5s2RlZYmIyPnz52Xu3Lni5uYmQ4cOldzcXMnOzpZly5ZJ69atRdM0CQgIYOb3YZbkKtgJ1GGW6jBL+2LHUodZOh47ljrMsihO3JYgBw8elNq1a8vGjRttll+9elUmT54sHh4eMnLkSH35nDlzRNM0eeaZZ4rtC+BxMUu1mKf9REVFiaenp7z77rvy/vvvS0BAgAwfPtxmnYMHD0rFihWlT58+NgWu4FpXzDYfs3Qsq9UqmZmZEh4eLjNmzBCRovl9+OGHommaLFmyRETyb0awePFiGTZsmL4uM2eW5FrYCdRhluowS/tix1KHWToWO5Y6zPK3ceK2BFm9erW0bdtWUlJS9I25YO/b9evXZciQIeLn5yfHjx/Xn7Nly5Zi/QJ4XMxSLeZpH7GxsWI2m/XrAYmIREREyMSJEyUpKUlSUlLk7t27IpJf4KpUqSKdOnWSjIwMff3CNyooyZilMbKyssTX11eioqJEpGiGt2/flq5du0rr1q31PfIWi4VfPB6AWZKrYCdQh1mqwyzthx1LHWZpDHYsdZjlg5lAJcaRI0dw5coVlC9fHm5ubgAAk8kEEUGVKlUwduxYJCcnIzk5WX9O79694ebmBovFoj+HmKVqzFO9f//73xg+fDheeeUVhIaG6suPHDmClStXolWrVqhXrx7mz5+PtLQ0tGzZElu2bIGnpye8vLz09TVNM2L4ToVZGsNiseDu3btIT09HZmYmgKIZVqhQAS1btsSVK1dgtVoBAGazGZqmQUT43vArZkmuhJ1AHWapDrO0D3YsdZilMdix1GGWv40TtyVI7dq1kZaWhmPHjkFE9OUFLwZfX19UqlRJf5EUVlxfAI+LWarFPNWrVasWunTpgnPnzmHjxo0AgAEDBuD69etYsWIF9u/fj759+2LGjBlITEwEALRt2xa7du2CyWTSPwiJWTraN998AyD/te3p6Yn27dtj1apVOHnypL6OiCAvLw8A4O7ujsDAQJQpU8bm9/CLB7Mk18ROoA6zVIdZ2gc7ljrM0rHYsdRhlr+PE7clSLdu3ZCdnY3o6Gjk5OToywteANeuXUPNmjXh5+dn1BBdBrNUi3mqV6tWLcTExMDd3R0xMTFo3rw5Tp8+jX379qFr165o0KAB5syZA7PZrH9YFmYy8eOhALN0nMuXL6NXr17o2bMnAKB06dLo1asXfvjhB8ybNw+nT58GkF/MzGYzLBYL9u/fj/r16xs5bKfELMlVsROowyzVYZb2wY6lDrN0HHYsdZjlQ3LslRnIUU6ePClfffWVzJ8/X/bu3SvJyckiIjJ37lwxm83y2muvyaVLl/T17969Kz169JDOnTvb3HWSmKVqzNN+kpOT5dixY5Kamqpf3+f8+fPSq1cvqVixoixYsMBm/XPnzklAQIBs27bNgNE6N2ZpjKysLFm9erXUrVtXwsPD9eWTJ08WTdOke/fusnHjRrl48aJ8/fXX0qNHD3n66af1/yNel+0eZkmugJ1AHWapDrO0L3YsdZilMdix1GGWD4cTt8XQsmXLJCAgQOrXry/ly5eXUqVKSZ06dWTfvn2SlZUl06dPF3d3d2nYsKEMGjRIRo4cKe3bt5dGjRrpdz1l6cjHLNVinvazYsUKadiwoVStWlXq1asnu3fv1h+7ePGihIWFSefOnSUuLk5fHhYWJu3atROLxWLEkJ0WszRGQfFKT0+X7du3i5+fnwwaNEh/fP78+RIUFCSapkmZMmXkmWeekdDQUP29gdnfwyzJFbATqMMs1WGW9sWOpQ6zNAY7ljrM8uFx4raYWbVqlZQuXVpWr14t58+fF5H8N/X27dtLqVKl5IsvvhARkb1798p//dd/SYsWLaR///4SGRnJu57eh1mqxTztJyoqSkqVKiVRUVHy/fffS8eOHaVly5Y265w7d05CQ0OlS5cusnr1agkNDRV/f/8S+cH3nzBLxyu4I6zVapXs7Gx9eatWrUTTNOnTp4++7Ny5c3L48GHZtWuXnDp1Sv9yzPeGfMySXAU7gTrMUh1maV/sWOowS8djx1KHWT46TtwWI5cvX5YWLVpITEyMiNgeNn7o0CHp0qWLlC9fXk6fPi0i+XuD73/D5ht4PmapFvO0n6VLl4qHh4fs3LlTX/bll19K79695dNPP5W9e/fqp/OdO3dOevbsKe7u7hIYGKgXt5L2wfdbmKXjffnllzJq1Ci5cOGCzfJ+/fpJYGCgfPTRR+Lr62tz6tT9eGRTPmZJroKdQB1mqQ6ztC92LHWYpeOxY6nDLB8PJ26LkYSEBKlZs6YcOnRIX1Z4o/7ss8+kYsWKMnv2bBFhufhPmKVazNM+Lly4INWrV5c2bdrYLP/zn/8s1apVk1q1akn16tWlefPmcu7cORHJv+7VW2+9xSND7sMsHavgC/HMmTOlQYMG8vrrr8uNGzdERCQ8PFyefvppuXr1qlgsFlm/fr34+vrKgAEDjByy02KW5GrYCdRhluowS/thx1KHWToWO5Y6zPKP4cRtMXDt2jUREfn6669F0zQ5ceKEzeOF9xi3adPmP+69KOmYpVrM075SU1MlJiZGatSoIaNGjRIRkf79+0tAQIAcO3ZM0tPTJTY2VmrUqGFzGl8BFrd7mKVjFX4vmD9/vjRt2lQmTpwowcHB8swzz+inqIrkn061YcMGKVWqlEyZMsWI4To1Zkmugp1AHWapDrO0P3YsdZilY7FjqcMs/xhO3Lq4hQsXSq1ateTmzZty9uxZqVChgkyZMkUyMzNt1ivYYxwSEqK/yZMtZqkW83SM1NRUWbZsmVSuXFmqVq0qQUFB+pcQkfx8GzVqxGwfArN0jKioKClbtqycPXtWXzZ37lypV6+ePPHEE3LgwAERyf+yXPCFOTMzU/bs2cMjnO7DLMlVsBOowyzVYZaOw46lDrN0DHYsdZjlH+cGclnR0dGYOHEi4uLiUKlSJVSqVAk9evTABx98gBYtWuC5555DqVKlAAAmkwkZGRnIyMhAQECAwSN3PsxSLeZpPxkZGUhPT4fVakWlSpXg4+ODgQMHAgCmT5+O+vXro2rVqgAAq9WKX375BeXKlUOdOnUMHLVzYpaOFxMTg1GjRmHTpk2oX7++vnz8+PFwd3dHbGwstm7dipo1a8LX1xeSv4MZXl5e6Ny5MwAgLy8PZrPZqD/BaTBLchXsBOowS3WYpX2xY6nDLB2PHUsdZqmIcXPG9EfExMSIh4eHbN261WZ5cnKyhISEyBNPPCFRUVFy5coVERG5dOmShIaGSlBQEPda3IdZqsU87Wf9+vUSHBwsVatWlcqVK0udOnVk2bJlcuvWLcnNzZVly5ZJlSpVZPjw4fpzQkJCJCgoiKdG3YdZOl5UVJSYzWbZvHmzzfKDBw/qP8+dO1eCgoLk9ddfl+TkZEcP0WUwS3IV7ATqMEt1mKV9sWOpwywdjx1LHWapDiduXdD//d//iaZpMn36dJvlL730ksycOVOSkpIkLCxMzGazlCtXTurXry9BQUHSunVr/U6SLB35mKVazNN+li5dKmXKlJFZs2bJ5s2b5ZNPPpGePXuKpmkyZswYuX79umRnZ8uyZcukWrVq8te//lV69+4t9evXZ7b3YZaOt3nzZtE0Tfbu3WuzvG/fvjJkyBDJyMjQl73//vvSvHlzGTZsmM2pf5SPWZKrYCdQh1mqwyztix1LHWbpeOxY6jBLtThx64ISEhKkffv20rNnT/3Op+Hh4dKgQQO5fPmyvt6mTZtk4cKF8u6778qOHTv0N27ufbuHWarFPO3j4MGDUqtWLVm7dm2RxyZMmCCapsl7770nIiIpKSmyfPlyKV26tDRo0EAvbsw2H7N0vIyMDBk1apRUqVJFli9fri/v06ePBAQE6HvXC3+5mDJlivzlL3+xuaM3MUtyLewE6jBLdZil/bBjqcMsHY8dSx1mqZ4mImL05Rro0SUmJmLMmDEwm81IS0tDVlYWNm/ejDp16vzHa4Dw+iBFMUu1mKd6S5YsQVxcHHbu3Ily5coByL+GlclkAgC8/PLL2LBhA86ePYtq1aohLS0NX3/9NZ577jmYzWZYLBa4ufGS5gCzNMrp06exdOlS7Ny5E2+++Sa++uorxMfHY9u2bahbty5EBJqm2fxfPGgZMUtyLewE6jBLdZilfbBjqcMsjcGOpQ6zVMyoGWP64xISEqRr167i4+MjGzZs0JcX3ImPHh6zVIt5qjVq1Chp3ry5iNhmWPDzv//9b/H29pZdu3YVeS5PkbLFLB3nm2++kdWrV+vZJiYmSkREhFStWlUqVqwo6enpImJ7REj79u1l7ty5+r/5npGPWZIrYydQh1mqwyzVY8dSh1k6DjuWOszSfjiN7cL8/f0RFRWFVq1aYdmyZThw4AAAQNM0CA+kfiTMUi3m+cd98cUXuH37NgCgdu3aOHXqFJKSkmwy1DQNQH7eubm5SE9PL/J7eGQIszTCypUr8Ze//AV79uzBoUOHAABPPvkkRo0ahRdeeAEVK1bE+vXrAQBubm7Iy8vD888/j59++gljxozRf0/B/0tJxizJ1bETqMMs1WGWarBjqcMsHY8dSx1maWdGzRiTOgkJCRIcHCzBwcFy4MABo4fj0pilWszz8aSnp0vDhg2ldu3akpKSIidPnpTq1avLwIED9Qu2F1y/SkTk2LFj0qJFCzl+/LhRQ3ZazNLxli9fLt7e3hIbGys//fRTkcdPnz4tY8eOlfr168uSJUtERCQ0NNTmZhq8Lls+ZknFCTuBOsxSHWb5+Nix1GGWjseOpQ6ztD9O3BYTCQkJEhoaKs2bN5cTJ04YPRyXxizVYp6P54cffpDmzZtL48aNJSUlRd566y0pX768DBs2zOYDMSsrS8LCwqRz5868mPtvYJaOc+bMGWnYsKHExcUVeezy5ct6Kbtw4YJERERIw4YNpXr16ixuD8AsqThiJ1CHWarDLB8fO5Y6zNJx2LHUYZaOwYnbYuTUqVMybtw4voErwCzVYp4PryCj3NxcSU5OlmbNmkmXLl0kNTVVxo0bJxUrVpSaNWvKhAkTZMSIEdKlSxdp1KiR/sHHjO9hlo63b98+qVu3rly4cEFftn79ehkyZIhUqFBBnnrqKdm4caOIiJw9e1aGDh0q3bp1Y3F7AGZJxRU7gTrMUh1m+WjYsdRhlo7HjqUOs3QMTYQX8SmOeCc+dZilWszzwW7duoVKlSoBAHJycuDh4QEACAkJwRdffIHWrVtj165dOHDgANauXYsTJ06gbt26aNKkCd5++224ubnxDrK/YpbG+eabbzBs2DCMHz8egwcPxujRoxEfH48aNWogNDQUX331Fb766iscP34ctWrVwuXLl1G9enWYTCZmfh9mSSUBO4E6zFIdZvnb2LHUYZbGYcdSh1k6iNEzx0REJd2//vUv6dSpk/zzn/+0Wd63b19p1KiR7N69W5o0aSJNmzaVlJQUERFJTU21WZd3kM3HLB1vw4YN+vWqbt26JQMGDJA6depIhQoVpF69erJhwwa5fv26iIhkZGSIt7e3rFq1yuZ38GiRfMySiIicFTuWOszS8dix1GGWjsfpbSIig1WpUgUigvfeew9lypRBs2bN0LdvX5w5cwafffYZatWqhTVr1mDQoEFo37499u3bp++hL8A7yOZjlo53/PhxzJ49G2azGUOHDsWiRYuQmJiIa9eu4fnnn4e7u7u+7vnz51GvXj3UqVPH5nfwyKZ8zJKIiJwVO5Y6zNLx2LHUYZaOx0slEBE5gcTERIwZMwZmsxlpaWnIzMzEli1bbD7kzpw5g+7du6NDhw5YtWqVcYN1cszS8WbMmIEZM2YgOjoa//M///PAdbKysvDCCy/g7t27+Oyzz1jYfgOzJCIiZ8WOpQ6zdDx2LHWYpYMZebgvERHdk5CQIF27dhUfHx/ZsGGDvrzwqSTJyck8NeohMEv7uz+7yMhIMZvNEhsba7M8LS1Ntm/fLsHBwdK4cWPeTOMBmCUREbkKdix1mKX9sWOpwyyNwyNuiYicSFJSEkaNGgWTyYRJkyahXbt2AIreKCMvL4+nSP0OZmkfp06dQsOGDQEUzW7atGn429/+hpUrV2LgwIHIycnBrFmzcPToUZQvXx6xsbG8mUYhzJKIiFwRO5Y6zNI+2LHUYZbG48QtEZGTKTh1CgCmTJmCtm3bGjwi18Us1frhhx8QFhaGQYMG4Z133gFQtMC9+eabeP/993H06FE0btwYN2/exNWrV9GoUSNomsYvHr9ilkRE5MrYsdRhlmqxY6nDLJ0DLzJBRORk/P398eGHH8JsNiMiIgLx8fFGD8llMUu1ypcvjxdeeAHbt2/HjBkzAOTfHCMvL09fZ+rUqejQoQNiY2NhsVjwxBNPoHHjxtA0DVarlcXtV8ySiIhcGTuWOsxSLXYsdZilc+CxykRETsjf3x9z587FJ598gsDAQKOH49KYpTo1atTAyJEj4eHhgTVr1sBqtWLatGkwm836KVBeXl7w9vaGiBQ5JYo3JbiHWRIRkatjx1KHWarDjqUOs3QOnLglInJSTz31FObNmweg6HWu6NEwy8eXkZGB9PR0WK1WVKxYETVr1sTIkSMBAOvWrYOmaYiMjNSL2i+//IKcnBybuyJTPmZJRETFDTuWOszy8bFjqcMsnQ8nbomIXACLmzrM8uFt2LABy5Ytw7Fjx2C1WlGmTBlMnToVffv2RUREBABg5cqV+PnnnxEZGYmLFy9i5syZuHHjBl577TVjB+9kmCURERV37FjqMMuHx46lDrN0Trw5GRERERURGxuLMWPGYPLkyWjQoAFSUlKwY8cO/OMf/8Do0aMxffp0aJqGDRs2YPr06cjKyoKfnx+qV6+OrVu3wt3dnTcj+BWzJCIiIlKPHUsdZum8OHFLRERENr777jv07dsXc+bMwcCBA20emzhxIubOnYvZs2dj4sSJyM3NRU5ODg4ePIhq1aohICAAJpNJv+5VSccsiYiIiNRjx1KHWTo3TtwSERGRjSVLliAuLg47d+5EuXLlANhea+3ll1/Ghg0bkJCQgKpVqxZ5Pq/Ldg+zJCIiIlKPHUsdZuncmCwRERHZOHHiBO7evYty5cqhYP+uyWTSfx42bBisViuOHj36wOezuN3DLImIiIjUY8dSh1k6N6ZLRERE+OKLL3D79m0AQO3atXHq1CkkJSVB0zS9tGmaBgDw9/dHbm4u0tPTDRuvM2OWREREROqxY6nDLF0HJ26JiIhKuIyMDIwbNw5BQUFITU1FaGgoypcvjylTpuD69evQNA25ubn6+pcvX0bjxo0REBBg4KidE7MkIiIiUo8dSx1m6Vo4cUtERFTClS1bFhs3bkSVKlXQsWNH1KhRA0OGDMHnn3+OSZMm4dq1a3B3dwcA3LlzB2+//TbKli2LRo0aGTxy58MsiYiIiNRjx1KHWboW3pyMiIioBCu4mYDFYsGVK1fQp08flC9fHps3b8aMGTOwfPlyeHl54b//+7+RmpqKpKQk3LhxA0eOHIG7uztvRlAIsyQiIiJSjx1LHWbpepg2ERFRCXTr1i0A+TcTyMnJgZubG3x9fVG5cmXs3bsXPXr0wNSpUxEXF4eOHTti165duHbtGtq2bYujR4/C3d0dFouFxQ3MkoiIiMge2LHUYZaui0fcEhERlTD79+/H1KlTMX36dHTo0EFf3q9fP5w9exYLFizA+PHjYTKZsGfPHpQvXx5paWnw8fHR183Ly4PZbDZi+E6FWRIRERGpx46lDrN0bZwqJyIiKmGqVKkCEcF7772HI0eOAAD69u2L06dPY+fOnejSpQvWrFkDEUH79u1x69Ytm+IGgMXtV8ySiIiISD12LHWYpWvjEbdEREQlUGJiIsaMGQOz2Yy0tDRkZmZiy5YtqFOnjr7OmTNn0L17d3To0AGrVq0ybrBOjlkSERERqceOpQ6zdF2cuCUiIiqhEhMTMXLkSBw6dAhLlixBv379AMDmpgM//vgjatasyb3sv4NZEhEREanHjqUOs3RNnLglIiIqwZKSkjBq1CiYTCZMmjQJ7dq1A4Aid4zlda1+H7MkIiIiUo8dSx1m6Xo4cUtERFTCFZw6BQBTpkxB27ZtDR6R62KWREREROqxY6nDLF0Lb05GRERUwvn7++PDDz+E2WxGREQE4uPjjR6Sy2KWREREROqxY6nDLF0LJ26JiIgI/v7+mDt3Ljp06IDAwECjh+PSmCURERGReuxY6jBL18FLJRAREVER91/nih4fsyQiIiJSjx1LHWbpvDhxS0RERERERERERORkOJ1ORERERERERERE5GQ4cUtERERERERERETkZDhxS0RERERERERERORkOHFLRERERERERERE5GQ4cUtERERERERERETkZDhxS0RERERERERERORkOHFLRERERERERERE5GQ4cUtERERERERERETkZDhxS0RERERERERERORkOHFLRERERERERERE5GT+HzoBJvoAFyV5AAAAAElFTkSuQmCC", + "text/plain": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "# | echo: false\n", + "colors = [\n", + " (\"#A9B9C3\", 0.5), # Grey-bluish color 1\n", + " (\"#7A8D9D\", 0.5), # Grey-bluish color 2\n", + " (\"#5B6D79\", 0.5), # Grey-bluish color 3\n", + " ('#F95D6A', 0.75) # Green color for the last\n", + "]\n", + "\n", + "\n", + "# Filter evaluation data by metric and set unique_id as index\n", + "rmse_df = evaluation_df[evaluation_df['metric'] == 'rmse'].set_index('unique_id')\n", + "smape_df = evaluation_df[evaluation_df['metric'] == 'smape'].set_index('unique_id')\n", + "\n", + "# Plot function with custom colors and opacity\n", + "def plot_metric(ax, df, title, ylabel):\n", + " x = np.arange(len(df))\n", + " bar_width = 0.2\n", + " for i, (col, (color, alpha)) in enumerate(zip(df.columns[1:], colors)):\n", + " ax.bar(x + i * bar_width, df[col], width=bar_width, label=col, color=color, alpha=alpha)\n", + " ax.set(title=title, ylabel=ylabel, xticks=x + bar_width * (len(df.columns[1:]) - 1) / 2, xticklabels=df.index)\n", + " ax.tick_params(axis='x', rotation=45)\n", + " ax.legend()\n", + "\n", + "# Generate side-by-side plots for RMSE and SMAPE\n", + "fig, axes = plt.subplots(1, 2, figsize=(14, 6))\n", + "plot_metric(axes[0], rmse_df, \"RMSE Comparison Across Models\", \"RMSE\")\n", + "plot_metric(axes[1], smape_df*100, \"%SMAPE Comparison Across Models\", \"SMAPE\")\n", + "\n", + "plt.tight_layout()\n", + "plt.show()\n", + " " + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Benchmark Results\n", + "For a more comprehensive dive into model accuracy and performance, explore our [Time Series Model Arena](https://github.com/Nixtla/nixtla/tree/main/experiments/foundation-time-series-arena)! TimeGPT continues to lead the pack with exceptional performance across benchmarks! 🌟\n", + "\n", + "![image](https://github.com/Nixtla/nixtla/assets/10517170/1c042591-0585-4a5b-a548-2017a28f2d4f)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## 4. Conclusion\n", + "At the end of this notebook, we’ve put together a handy table to show you exactly where TimeGPT shines brightest compared to other forecasting models. β˜€οΈ Think of it as your quick guide to choosing the best model for your unique project needs. We’re confident that TimeGPT will be a valuable tool in your forecasting journey. Don’t forget to visit our [dashboard](https://dashboard.nixtla.io) to generate your TimeGPT `api_key` and get started today! Happy forecasting, and enjoy the insights ahead! " + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ScenarioTimeGPTClassical Models (e.g., ARIMA)Machine Learning Models (e.g., XGB, LGBM)Deep Learning Models (e.g., N-HITS)
Seasonal Patternsβœ… Performs well with minimal setupβœ… Handles seasonality with adjustments (e.g., SARIMA)βœ… Performs well with feature engineeringβœ… Captures seasonal patterns effectively
Non-Linear Patternsβœ… Excels, especially with complex non-linear patterns❌ Limited performance❌ Struggles without extensive feature engineeringβœ… Performs well with non-linear relationships
Large Datasetβœ… Highly scalable across many series❌ Slow and resource-intensiveβœ… Scalable with optimized implementations❌ Requires significant resources for large datasets
Small Datasetβœ… Performs well; requires only one data point to startβœ… Performs well; may struggle with very sparse dataβœ… Performs adequately if enough features are extracted❌ May need a minimum data size to learn effectively
Preprocessing Requiredβœ… Minimal preprocessing needed❌ Requires scaling, log-transform, etc., to meet model assumptions.❌ Requires extensive feature engineering for complex patterns❌ Needs data normalization and preprocessing
Accuracy Requirementβœ… Achieves high accuracy with minimal tuning❌ May struggle with complex accuracy requirementsβœ… Can achieve good accuracy with tuningβœ… High accuracy possible but with significant resource use
Scalabilityβœ… Highly scalable with minimal task-specific configuration❌ Not easily scalableβœ… Moderate scalability, with feature engineering and tuning per task❌ Limited scalability due to resource demands
Computational Resourcesβœ… Highly efficient, operates seamlessly on CPU, no GPU neededβœ… Light to moderate, scales poorly with large datasets❌ Moderate, depends on feature complexity❌ High resource consumption, often requires GPU
Memory Requirementβœ… Efficient memory usage for large datasetsβœ… Moderate memory requirements❌ High memory usage for larger datasets or many series cases❌ High memory consumption for larger datasets and multiple series
Technical Requirements & Domain Knowledgeβœ… Low; minimal technical setup and no domain expertise neededβœ… Low to moderate; needs understanding of stationarity❌ Moderate to high; requires feature engineering and tuning❌ High; complex architecture and tuning
\n" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "python3", + "language": "python", + "name": "python3" + } + }, + "nbformat": 4, + "nbformat_minor": 2 +} diff --git a/nbs/src/nixtla_client.ipynb b/nbs/src/nixtla_client.ipynb index 64e145bb..bc61d157 100644 --- a/nbs/src/nixtla_client.ipynb +++ b/nbs/src/nixtla_client.ipynb @@ -145,6 +145,7 @@ "from dotenv import load_dotenv\n", "from fastcore.test import test_eq, test_fail\n", "from utilsforecast.data import generate_series\n", + "from utilsforecast.feature_engineering import fourier\n", "\n", "from nixtla.date_features import SpecialDates" ] @@ -432,7 +433,8 @@ " target_col: str,\n", " hist_exog: Optional[list[str]],\n", ") -> tuple[DFType, Optional[DFType]]:\n", - " exogs = [c for c in df.columns if c not in (id_col, time_col, target_col)]\n", + " base_cols = {id_col, time_col, target_col}\n", + " exogs = [c for c in df.columns if c not in base_cols]\n", " if hist_exog is None:\n", " hist_exog = []\n", " if X_df is None:\n", @@ -449,7 +451,7 @@ " return df, None\n", "\n", " # exogs in df that weren't declared as historic nor future\n", - " futr_exog = [c for c in X_df.columns if c not in (id_col, time_col)]\n", + " futr_exog = [c for c in X_df.columns if c not in base_cols]\n", " declared_exogs = {*hist_exog, *futr_exog}\n", " ignored_exogs = [c for c in exogs if c not in declared_exogs]\n", " if ignored_exogs:\n", @@ -467,6 +469,15 @@ " f\"but not in `df`: {missing_futr}.\"\n", " )\n", "\n", + " # features are provided through X_df but declared as historic\n", + " futr_and_hist = set(futr_exog) & set(hist_exog)\n", + " if futr_and_hist:\n", + " warnings.warn(\n", + " \"The following features were declared as historic but found in `X_df`: \"\n", + " f\"{futr_and_hist}, they will be considered as historic.\"\n", + " )\n", + " futr_exog = [f for f in futr_exog if f not in hist_exog]\n", + "\n", " # Make sure df and X_df are in right order\n", " df = df[[id_col, time_col, target_col, *futr_exog, *hist_exog]]\n", " X_df = X_df[[id_col, time_col, *futr_exog]]\n", @@ -549,7 +560,7 @@ " processed = ufp.process_df(\n", " df=df, id_col=id_col, time_col=time_col, target_col=target_col\n", " )\n", - " if X_df is not None:\n", + " if X_df is not None and X_df.shape[1] > 2:\n", " X_df = ensure_time_dtype(X_df, time_col=time_col)\n", " processed_X = ufp.process_df(\n", " df=X_df, id_col=id_col, time_col=time_col, target_col=None,\n", @@ -733,7 +744,8 @@ " if isinstance(v, np.ndarray):\n", " if np.issubdtype(v.dtype, np.floating):\n", " v_cont = np.ascontiguousarray(v, dtype=np.float32)\n", - " d[k] = np.nan_to_num(v_cont, \n", + " d[k] = np.nan_to_num(\n", + " v_cont, \n", " nan=np.nan, \n", " posinf=np.finfo(np.float32).max, \n", " neginf=np.finfo(np.float32).min,\n", @@ -984,6 +996,7 @@ " finetune_depth: _Finetune_Depth,\n", " finetune_loss: _Loss,\n", " clean_ex_first: bool,\n", + " hist_exog_list: Optional[list[str]],\n", " validate_api_key: bool,\n", " add_history: bool,\n", " date_features: Union[bool, list[Union[str, Callable]]],\n", @@ -1040,6 +1053,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " validate_api_key=validate_api_key,\n", " add_history=add_history,\n", " date_features=date_features,\n", @@ -1175,6 +1189,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " validate_api_key=validate_api_key,\n", " add_history=add_history,\n", " date_features=date_features,\n", @@ -1243,7 +1258,7 @@ " X = processed.data[:, 1:].T\n", " if futr_cols is not None:\n", " logger.info(f'Using future exogenous features: {futr_cols}')\n", - " if hist_exog_list is not None:\n", + " if hist_exog_list:\n", " logger.info(f'Using historical exogenous features: {hist_exog_list}')\n", " else:\n", " X = None\n", @@ -1827,6 +1842,7 @@ " finetune_depth: _Finetune_Depth,\n", " finetune_loss: _Loss,\n", " clean_ex_first: bool,\n", + " hist_exog_list: Optional[list[str]],\n", " date_features: Union[bool, Sequence[Union[str, Callable]]],\n", " date_features_to_one_hot: Union[bool, list[str]],\n", " model: _Model,\n", @@ -1836,7 +1852,7 @@ " \n", " schema, partition_config = _distributed_setup(\n", " df=df,\n", - " method='forecast',\n", + " method='cross_validation',\n", " id_col=id_col,\n", " time_col=time_col,\n", " target_col=target_col,\n", @@ -1864,6 +1880,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " date_features=date_features,\n", " date_features_to_one_hot=date_features_to_one_hot,\n", " model=model,\n", @@ -1891,6 +1908,7 @@ " finetune_depth: _Finetune_Depth = 1,\n", " finetune_loss: _Loss = 'default',\n", " clean_ex_first: bool = True,\n", + " hist_exog_list: Optional[list[str]] = None,\n", " date_features: Union[bool, list[str]] = False,\n", " date_features_to_one_hot: Union[bool, list[str]] = False,\n", " model: _Model = 'timegpt-1',\n", @@ -1947,8 +1965,9 @@ " finetune_loss : str (default='default')\n", " Loss function to use for finetuning. Options are: `default`, `mae`, `mse`, `rmse`, `mape`, and `smape`.\n", " clean_ex_first : bool (default=True)\n", - " Clean exogenous signal before making forecasts\n", - " using TimeGPT.\n", + " Clean exogenous signal before making forecasts using TimeGPT.\n", + " hist_exog_list : list of str, optional (default=None)\n", + " Column names of the historical exogenous features.\n", " date_features : bool or list of str or callable, optional (default=False)\n", " Features computed from the dates.\n", " Can be pandas date attributes or functions that will take the dates as input.\n", @@ -1990,6 +2009,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " date_features=date_features,\n", " date_features_to_one_hot=date_features_to_one_hot,\n", " model=model,\n", @@ -2053,9 +2073,29 @@ " targets = _array_tails(targets, orig_indptr, np.diff(processed.indptr))\n", " if processed.data.shape[1] > 1:\n", " X = processed.data[:, 1:].T\n", - " logger.info(f'Using the following exogenous features: {x_cols}')\n", + " if hist_exog_list is None:\n", + " hist_exog = None\n", + " futr_exog = x_cols\n", + " else:\n", + " missing_hist = set(hist_exog_list) - set(x_cols)\n", + " if missing_hist:\n", + " raise ValueError(\n", + " \"The following exogenous features were declared as historic \"\n", + " f\"but were not found in `df`: {missing_hist}.\"\n", + " )\n", + " futr_exog = [c for c in x_cols if c not in hist_exog_list]\n", + " # match the forecast method order [future, historic]\n", + " fcst_features_order = futr_exog + hist_exog_list\n", + " x_idxs = [x_cols.index(c) for c in fcst_features_order]\n", + " X = X[x_idxs]\n", + " hist_exog = [fcst_features_order.index(c) for c in hist_exog_list]\n", + " if futr_exog:\n", + " logger.info(f'Using future exogenous features: {futr_exog}')\n", + " if hist_exog_list:\n", + " logger.info(f'Using historical exogenous features: {hist_exog_list}')\n", " else:\n", " X = None\n", + " hist_exog = None\n", "\n", " logger.info('Calling Cross Validation Endpoint...')\n", " payload = {\n", @@ -2070,6 +2110,7 @@ " 'step_size': step_size,\n", " 'freq': standard_freq,\n", " 'clean_ex_first': clean_ex_first,\n", + " 'hist_exog': hist_exog,\n", " 'level': level,\n", " 'finetune_steps': finetune_steps,\n", " 'finetune_depth': finetune_depth,\n", @@ -2239,6 +2280,7 @@ " finetune_depth: _Finetune_Depth,\n", " finetune_loss: _Loss,\n", " clean_ex_first: bool,\n", + " hist_exog_list: Optional[list[str]],\n", " validate_api_key: bool,\n", " add_history: bool,\n", " date_features: Union[bool, list[Union[str, Callable]]],\n", @@ -2267,6 +2309,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " validate_api_key=validate_api_key,\n", " add_history=add_history,\n", " date_features=date_features,\n", @@ -2367,6 +2410,7 @@ " finetune_depth: _Finetune_Depth,\n", " finetune_loss: _Loss,\n", " clean_ex_first: bool,\n", + " hist_exog_list: Optional[list[str]],\n", " date_features: Union[bool, list[str]],\n", " date_features_to_one_hot: Union[bool, list[str]],\n", " model: _Model,\n", @@ -2388,6 +2432,7 @@ " finetune_depth=finetune_depth,\n", " finetune_loss=finetune_loss,\n", " clean_ex_first=clean_ex_first,\n", + " hist_exog_list=hist_exog_list,\n", " date_features=date_features,\n", " date_features_to_one_hot=date_features_to_one_hot,\n", " model=model,\n", @@ -2549,6 +2594,69 @@ ")" ] }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "# historic exog in cv\n", + "freq = 'D'\n", + "h = 5\n", + "series = generate_series(2, freq=freq)\n", + "series_with_features, _ = fourier(series, freq=freq, season_length=7, k=2)\n", + "splits = ufp.backtest_splits(\n", + " df=series_with_features,\n", + " n_windows=1,\n", + " h=h,\n", + " id_col='unique_id',\n", + " time_col='ds',\n", + " freq=freq,\n", + ")\n", + "_, train, valid = next(splits)\n", + "x_cols = train.columns.drop(['unique_id', 'ds', 'y']).tolist()\n", + "for hist_exog_list in [None, [], [x_cols[2], x_cols[1]], x_cols]:\n", + " cv_res = nixtla_client.cross_validation(\n", + " series_with_features,\n", + " n_windows=1,\n", + " h=h,\n", + " freq=freq,\n", + " hist_exog_list=hist_exog_list,\n", + " )\n", + " fcst_res = nixtla_client.forecast(\n", + " train,\n", + " h=h,\n", + " freq=freq,\n", + " hist_exog_list=hist_exog_list,\n", + " X_df=valid,\n", + " )\n", + " np.testing.assert_allclose(\n", + " cv_res['TimeGPT'], fcst_res['TimeGPT'], atol=1e-4, rtol=1e-3\n", + " )" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": {}, + "outputs": [], + "source": [ + "#| hide\n", + "# different hist exog, different results\n", + "for X_df in (None, valid):\n", + " res1 = nixtla_client.forecast(train, h=h, X_df=X_df, freq=freq, hist_exog_list=x_cols[:2])\n", + " res2 = nixtla_client.forecast(train, h=h, X_df=X_df, freq=freq, hist_exog_list=x_cols[2:])\n", + " np.testing.assert_raises(\n", + " AssertionError,\n", + " np.testing.assert_allclose,\n", + " res1['TimeGPT'],\n", + " res2['TimeGPT'],\n", + " atol=1e-4,\n", + " rtol=1e-3,\n", + " )" + ] + }, { "cell_type": "code", "execution_count": null, diff --git a/nixtla/nixtla_client.py b/nixtla/nixtla_client.py index fc4d7f28..0ae246c0 100644 --- a/nixtla/nixtla_client.py +++ b/nixtla/nixtla_client.py @@ -351,7 +351,8 @@ def _validate_exog( target_col: str, hist_exog: Optional[list[str]], ) -> tuple[DFType, Optional[DFType]]: - exogs = [c for c in df.columns if c not in (id_col, time_col, target_col)] + base_cols = {id_col, time_col, target_col} + exogs = [c for c in df.columns if c not in base_cols] if hist_exog is None: hist_exog = [] if X_df is None: @@ -368,7 +369,7 @@ def _validate_exog( return df, None # exogs in df that weren't declared as historic nor future - futr_exog = [c for c in X_df.columns if c not in (id_col, time_col)] + futr_exog = [c for c in X_df.columns if c not in base_cols] declared_exogs = {*hist_exog, *futr_exog} ignored_exogs = [c for c in exogs if c not in declared_exogs] if ignored_exogs: @@ -386,6 +387,15 @@ def _validate_exog( f"but not in `df`: {missing_futr}." ) + # features are provided through X_df but declared as historic + futr_and_hist = set(futr_exog) & set(hist_exog) + if futr_and_hist: + warnings.warn( + "The following features were declared as historic but found in `X_df`: " + f"{futr_and_hist}, they will be considered as historic." + ) + futr_exog = [f for f in futr_exog if f not in hist_exog] + # Make sure df and X_df are in right order df = df[[id_col, time_col, target_col, *futr_exog, *hist_exog]] X_df = X_df[[id_col, time_col, *futr_exog]] @@ -470,7 +480,7 @@ def _preprocess( processed = ufp.process_df( df=df, id_col=id_col, time_col=time_col, target_col=target_col ) - if X_df is not None: + if X_df is not None and X_df.shape[1] > 2: X_df = ensure_time_dtype(X_df, time_col=time_col) processed_X = ufp.process_df( df=X_df, @@ -904,6 +914,7 @@ def _distributed_forecast( finetune_depth: _Finetune_Depth, finetune_loss: _Loss, clean_ex_first: bool, + hist_exog_list: Optional[list[str]], validate_api_key: bool, add_history: bool, date_features: Union[bool, list[Union[str, Callable]]], @@ -961,6 +972,7 @@ def format_X_df( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, validate_api_key=validate_api_key, add_history=add_history, date_features=date_features, @@ -1096,6 +1108,7 @@ def forecast( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, validate_api_key=validate_api_key, add_history=add_history, date_features=date_features, @@ -1164,7 +1177,7 @@ def forecast( X = processed.data[:, 1:].T if futr_cols is not None: logger.info(f"Using future exogenous features: {futr_cols}") - if hist_exog_list is not None: + if hist_exog_list: logger.info(f"Using historical exogenous features: {hist_exog_list}") else: X = None @@ -1764,6 +1777,7 @@ def _distributed_cross_validation( finetune_depth: _Finetune_Depth, finetune_loss: _Loss, clean_ex_first: bool, + hist_exog_list: Optional[list[str]], date_features: Union[bool, Sequence[Union[str, Callable]]], date_features_to_one_hot: Union[bool, list[str]], model: _Model, @@ -1773,7 +1787,7 @@ def _distributed_cross_validation( schema, partition_config = _distributed_setup( df=df, - method="forecast", + method="cross_validation", id_col=id_col, time_col=time_col, target_col=target_col, @@ -1801,6 +1815,7 @@ def _distributed_cross_validation( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, date_features=date_features, date_features_to_one_hot=date_features_to_one_hot, model=model, @@ -1828,6 +1843,7 @@ def cross_validation( finetune_depth: _Finetune_Depth = 1, finetune_loss: _Loss = "default", clean_ex_first: bool = True, + hist_exog_list: Optional[list[str]] = None, date_features: Union[bool, list[str]] = False, date_features_to_one_hot: Union[bool, list[str]] = False, model: _Model = "timegpt-1", @@ -1884,8 +1900,9 @@ def cross_validation( finetune_loss : str (default='default') Loss function to use for finetuning. Options are: `default`, `mae`, `mse`, `rmse`, `mape`, and `smape`. clean_ex_first : bool (default=True) - Clean exogenous signal before making forecasts - using TimeGPT. + Clean exogenous signal before making forecasts using TimeGPT. + hist_exog_list : list of str, optional (default=None) + Column names of the historical exogenous features. date_features : bool or list of str or callable, optional (default=False) Features computed from the dates. Can be pandas date attributes or functions that will take the dates as input. @@ -1927,6 +1944,7 @@ def cross_validation( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, date_features=date_features, date_features_to_one_hot=date_features_to_one_hot, model=model, @@ -1990,9 +2008,29 @@ def cross_validation( targets = _array_tails(targets, orig_indptr, np.diff(processed.indptr)) if processed.data.shape[1] > 1: X = processed.data[:, 1:].T - logger.info(f"Using the following exogenous features: {x_cols}") + if hist_exog_list is None: + hist_exog = None + futr_exog = x_cols + else: + missing_hist = set(hist_exog_list) - set(x_cols) + if missing_hist: + raise ValueError( + "The following exogenous features were declared as historic " + f"but were not found in `df`: {missing_hist}." + ) + futr_exog = [c for c in x_cols if c not in hist_exog_list] + # match the forecast method order [future, historic] + fcst_features_order = futr_exog + hist_exog_list + x_idxs = [x_cols.index(c) for c in fcst_features_order] + X = X[x_idxs] + hist_exog = [fcst_features_order.index(c) for c in hist_exog_list] + if futr_exog: + logger.info(f"Using future exogenous features: {futr_exog}") + if hist_exog_list: + logger.info(f"Using historical exogenous features: {hist_exog_list}") else: X = None + hist_exog = None logger.info("Calling Cross Validation Endpoint...") payload = { @@ -2007,6 +2045,7 @@ def cross_validation( "step_size": step_size, "freq": standard_freq, "clean_ex_first": clean_ex_first, + "hist_exog": hist_exog, "level": level, "finetune_steps": finetune_steps, "finetune_depth": finetune_depth, @@ -2173,6 +2212,7 @@ def _forecast_wrapper( finetune_depth: _Finetune_Depth, finetune_loss: _Loss, clean_ex_first: bool, + hist_exog_list: Optional[list[str]], validate_api_key: bool, add_history: bool, date_features: Union[bool, list[Union[str, Callable]]], @@ -2201,6 +2241,7 @@ def _forecast_wrapper( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, validate_api_key=validate_api_key, add_history=add_history, date_features=date_features, @@ -2304,6 +2345,7 @@ def _cross_validation_wrapper( finetune_depth: _Finetune_Depth, finetune_loss: _Loss, clean_ex_first: bool, + hist_exog_list: Optional[list[str]], date_features: Union[bool, list[str]], date_features_to_one_hot: Union[bool, list[str]], model: _Model, @@ -2325,6 +2367,7 @@ def _cross_validation_wrapper( finetune_depth=finetune_depth, finetune_loss=finetune_loss, clean_ex_first=clean_ex_first, + hist_exog_list=hist_exog_list, date_features=date_features, date_features_to_one_hot=date_features_to_one_hot, model=model,