v0.16.3

• add robust scaler
• relatively faster scaling in preprocessing
• Regression-based classifiers (see https://www.researchgate.net/publication/377227280_Regression-based_machine_learning_classifiers)
• DeepMTS (multivariate time series forecasting with deep quasi-random layers): see https://thierrymoudiki.github.io/blog/2024/01/15/python/quasirandomizednn/forecasting/DeepMTS
• AutoML for MTS (multivariate time series forecasting): see https://thierrymoudiki.github.io/blog/2023/10/29/python/quasirandomizednn/MTS-LazyPredict
• AutoML for DeepMTS (multivariate time series forecasting): see https://github.com/Techtonique/nnetsauce/blob/master/nnetsauce/demo/thierrymoudiki_20240106_LazyDeepMTS.ipynb
• Subsample continuous and discrete responses

v0.16.2

• AutoML for MTS (multivariate time series forecasting)
• AutoML for DeepMTS (multivariate time series forecasting)
• Subsample continuous and discrete responses
• Regression-based classifiers

v0.15.0

Automated Machine learning (AutoML)

• lazy predict for classification and regression (see https://thierrymoudiki.github.io/blog/2023/10/22/python/quasirandomizednn/nnetsauce-lazy-predict-preview)
• lazy predict for multivariate time series (see https://thierrymoudiki.github.io/blog/2023/10/29/python/quasirandomizednn/MTS-LazyPredict)
• lazy predict for deep quasi-randomized classifiers and regressors nnetworks (see this example for classification: examples/lazy_custom_deep_classification.py; and this example for regression: examples/lazy_custom_deep_regression.py

v0.14.0

version 0.14.0 (R & Python)

• update and align as much as possible with R version (new plotting function for multivariate time series (MTS), plot.MTS, is not S3, but it’s complicated)

# 0 - install packages ----------------------------------------------------

#utils::install.packages("remotes")
remotes::install_github("Techtonique/nnetsauce/R-package", force = TRUE)

# 1 - ENET simulations ----------------------------------------------------

obj <- nnetsauce::sklearn$linear_model$ElasticNet()
obj2 <- nnetsauce::MTS(obj, 
                       start_input = start(fpp::vn), 
                       frequency_input = frequency(fpp::vn),
                       kernel = "gaussian", replications = 100L)
X <- data.frame(fpp::vn)
obj2$fit(X)
obj2$predict(h = 10L)
typeof(obj2)

par(mfrow=c(2, 2))
plot.MTS(obj2, selected_series = "Sydney")
plot.MTS(obj2, selected_series = "Melbourne")
plot.MTS(obj2, selected_series = "NSW")
plot.MTS(obj2, selected_series = "BrisbaneGC")

# 2 - Bayesian Ridge ----------------------------------------------------

obj <- nnetsauce::sklearn$linear_model$BayesianRidge()
obj2 <- nnetsauce::MTS(obj,
                       start_input = start(fpp::vn), 
                       frequency_input = frequency(fpp::vn))
X <- data.frame(fpp::vn)
obj2$fit(X)
obj2$predict(h = 10L, return_std = TRUE)

par(mfrow=c(2, 2))
plot.MTS(obj2, selected_series = "Sydney")
plot.MTS(obj2, selected_series = "Melbourne")
plot.MTS(obj2, selected_series = "NSW")
plot.MTS(obj2, selected_series = "BrisbaneGC")

• colored graphics for Python class MTS

# !pip install nnetsauce —upgrade

import nnetsauce as ns
import numpy as np
import pandas as pd
from sklearn.linear_model import Ridge, BayesianRidge
from sklearn.ensemble import RandomForestRegressor
from time import time

url = "https://raw.githubusercontent.com/thierrymoudiki/mts-data/master/heater-ice-cream/ice_cream_vs_heater.csv"

df = pd.read_csv(url)

# ice cream vs heater (I don't own the copyright)
df.set_index('Month', inplace=True)
df.index.rename('date')

df = df.pct_change().dropna()

idx_train = int(df.shape[0]*0.8)
idx_end = df.shape[0]
df_train = df.iloc[0:idx_train,]

regr3 = Ridge()
obj_MTS3 = ns.MTS(regr3, lags = 4, n_hidden_features=7, #IRL, must be tuned
                  replications=50, kernel='gaussian',
                  seed=24, verbose = 1)
start = time()
obj_MTS3.fit(df_train)
print(f"Elapsed {time()-start} s")

obj_MTS3.plot("heater")
obj_MTS3.plot("ice cream")

v0.13.0

• Fix error in nodes' simulation (base.py)

• Use residuals and KDE for predictive simulations

• plot method for MTS (univariate and multivariate time series forecasting) objects

• this notebook contains some examples based on nnetsauce v0.13.0: https://github.com/Techtonique/nnetsauce/blob/master/nnetsauce/demo/thierrymoudiki_250923_nnetsauce_mts_plots.ipynb

100% numpy/scipy

100% numpy/scipy, rm Cython

Fix RandomBag in R

Fix RandomBag in R (problem with MemoryViews)

Fix RandomBag

Fix RandomBag (use input ndarrays in Cython)

fix sdist

fix sdist

RandomBagRegressor and MTS

• RandomBagRegressor: Bagging of models having fit and predict methods
• MTS: can use pandas data frames (with date index yyyy-mm-dd)