v0.6.2

Deprecations and Updates

1. Renamed a few expressions. Mostly now the "query_" prefix is no longer used for many features, but is kept for some others. E.g. instead of "query_lstsq", now the expression is "pds.lin_reg".

What's Changed

• added matthew corr coef by @abstractqqq in #276
• general maintenance by @abstractqqq in #277
• added auto_corr and ar_coeffs by @abstractqqq in #278
• added lcs_seq by @abstractqqq in #279
• Update polars by @abstractqqq in #280
• added basic splitting by @abstractqqq in #281
• fixes #282 by @abstractqqq in #283
• Misc features + Isotonic Rgression by @abstractqqq in #284
• Add monotonic checks by @abstractqqq in #285

Full Changelog: v0.6.1...v0.6.2

v0.6.1

What's Changed

• fixed bug in pca by @abstractqqq in #268
• added seed option for perturb and jitter by @abstractqqq in #269
• added multi target lr as an expression by @abstractqqq in #270
• added weights in lstsq plot by @abstractqqq in #272
• Simple lstsq by @abstractqqq in #273
• Fix bug in online lr by @abstractqqq in #274

Full Changelog: v0.6.0...v0.6.1

v0.6.0

Breaking

Dropped Python 3.8 support.

What's Changed

• added elastic net by @abstractqqq in #252
• try notebook in CI by @abstractqqq in #253
• Code reorg by @abstractqqq in #254
• simplified fft convolve by @abstractqqq in #255
• less allocation in fft convolve by @abstractqqq in #256
• added the most basic fairness metrics by @abstractqqq in #257
• Misc readme by @abstractqqq in #258
• Pykdtree by @abstractqqq in #259
• added docs on spatial by @abstractqqq in #260
• added elastic net model by @abstractqqq in #261
• Use Altair instead of plotly as plot backend by @abstractqqq in #266

New Contributors

• @AtlasPilotPuppy made their first contribution in #263

Full Changelog: v0.5.3...v0.6.0

v0.5.3

Breaking

• Lstsq queries not doesn't take method keyword. Whether it is ridge/lasso is determined by the l1_reg and l2_reg parameters.

Deprecation Plans

• v0.5.3 will be the last v0.5 versions. Polars < v1 support will be dropped for v0.6 onwards.

What's Changed

• Lstsq ux by @abstractqqq in #235
• Mase by @abstractqqq in #236
• Wls by @abstractqqq in #238
• removed unncessary options from kdt by @abstractqqq in #239
• Higher dim perf by @abstractqqq in #240
• added poly features by @abstractqqq in #241
• Faster kdtree construction by @abstractqqq in #242
• better default behavior for encoders by @abstractqqq in #243
• Less pointers in kdt by @abstractqqq in #244
• slightly improved knn str by @abstractqqq in #245
• added solver methods, refactored linear regression py class by @abstractqqq in #246
• Query sim cnt by @abstractqqq in #247
• Online lr 2 by @abstractqqq in #248
• normal lstsq with rcond by @abstractqqq in #249
• fixed legacy bugs in examples by @abstractqqq in #250
• added discrete entropy as a shorthand by @abstractqqq in #251

Full Changelog: v0.5.2...v0.5.3

v0.5.2

What's Changed

• added rank hot encoder | fixed a bug in the initial values of recursive lstsq by @abstractqqq in #223
• run bench by @abstractqqq in #224
• recursive rolling l2 by @abstractqqq in #226
• added filters in pipeline by @abstractqqq in #228
• added scriptable steps by @abstractqqq in #229
• Skip null policy by @abstractqqq in #232
• KNN queries now work correctly in .over() context by @abstractqqq in #232
• added knn freq cnts by @abstractqqq in #233

Full Changelog: v0.5.1...v0.5.2

v0.5.1

What's Changed

• L1 regression by @abstractqqq in #203
• Fix ridge regression by @abstractqqq in #204
• new readme by @abstractqqq in #205
• added epsilon for knn by @abstractqqq in #206
• fixed some typos by @abstractqqq in #207
• faster l1 l2 by @abstractqqq in #208
• removed a version of kdt that is not performant by @abstractqqq in #209
• added linear model class by @abstractqqq in #212
• added shrink_dtype in pipeline by @abstractqqq in #213
• Recursive linear reg by @abstractqqq in #214
• reorg features by @abstractqqq in #215
• fixed a pathing issue by @abstractqqq in #216
• added max bound option for knn by @abstractqqq in #217
• fix_typing by @abstractqqq in #218
• Initial pytest-bench setup by @CangyuanLi in #219
• Null policy by @abstractqqq in #220
• Rolling lr by @abstractqqq in #221
• Knn regression by @abstractqqq in #222

Full Changelog: v0.5.0...v0.5.1

v0.5.0

What's Changed

• better extreme case handling for roc auc by @abstractqqq in #187
• Mann whitney stats by @abstractqqq in #189
• updated polars ver by @abstractqqq in #190
• added polars v1 support by @abstractqqq in #191
• Refactor longest streak by @abstractqqq in #192
• added impute_nan by @abstractqqq in #194
• added bicor (biweight midcorrelation) by @abstractqqq in #195
• added eager execution on series by @abstractqqq in #197
• Kdt refactor. All kd tree related queries should be at least 10-20% faster now. by @abstractqqq in #198
• removed modules that don't really belong by @abstractqqq in #199
• removed unneeded rust dependency by @abstractqqq in #200
• Added L2 (ridge) regression option by @abstractqqq in #201

v0.5 Near Term Goals

1. Pipeline feature parity with categorical_encoder and Scikit-learn, with some exceptions
2. More thorough testing with v1
3. Uncaught Bugs / simple feature requests
4. Interpolations

v0.5 + Longer Term Goals

1. Lasso regression. Rolling regression.
2. Stand alone models (linear regression and Kdtree)
3. K-means / K-medoids

Full Changelog: v0.4.6...v0.5.0

v0.4.6

Warning

Support for polars=1.0 alpha, beta is not fully tested!

What's Changed

• Expand target types in encoders by @abstractqqq in #166
• added custom transforms by @abstractqqq in #167
• Gaussian noise by @abstractqqq in #168
• simplied target in pipeline by @abstractqqq in #169
• features in, out checks in pipeline, c3 stats, CID CE by @abstractqqq in #170
• add query_confusion_matrix by @CangyuanLi in #171
• better binary confusion matrix calculation by @abstractqqq in #172
• added funding by @abstractqqq in #173
• added winsorizing by @abstractqqq in #175
• added discord by @abstractqqq in #177
• fixed a argument name inconsistency due to polars version by @abstractqqq in #179
• added json serialization for pipeline by @abstractqqq in #181
• added some time series related statistics by @abstractqqq in #182
• Debug linear impute by @abstractqqq in #184
• better profile by @abstractqqq in #185
• added docs by @abstractqqq in #186

Full Changelog: v0.4.5...v0.4.6

v0.4.5

Breaking Changes

Previously, if you want to compare Edit distance between one column and a single string, you would do

df.select(
    pds.str_leven(pl.col("c"), "word")
)

But now you have to do

df.select(
    pds.str_leven(pl.col("c"), pl.lit("word"))
)

The previous behavior will now look for a column named "word" instead of using the word "word."

What's Changed

• refactored psi and add benford law by @abstractqqq in #158
• Skip some rows in KNN by @DGolubets in #157
• Added data_mask in knn_ptwise by @abstractqqq in #161
• Pretty Code Snippets by @s1lvester in #162
• Tests for entrophies by @abstractqqq in #163
• Woe encoding by @abstractqqq in #164
• Added Median Absolute Deviation and renamed str2 to string by @abstractqqq in #165

New Contributors

• @DGolubets made their first contribution in #157
• @s1lvester made their first contribution in #162

Full Changelog: v0.4.4...v0.4.5

v0.4.4

What's Changed

• added par option in convolve by @abstractqqq in #146
• More dia plots by @abstractqqq in #147
• fixed corr method bug by @abstractqqq in #148
• Add string pre-preprocessing code by @CangyuanLi in #150
• String Cleaning by @CangyuanLi in #152
• Pipeline by @abstractqqq in #155

New Contributors

• @CangyuanLi made their first contribution in #150

Full Changelog: v0.4.3...v0.4.4