Make JAX optional -- fix a few bugs

v0.51.2

reformat and bump v0.51.2

Make JAX optional

Aim of the release is to make JAX's loading optional.

In order to have JAX loaded, you'd need to type:

pip install nnetsauce[jax]

v0491

v0491:

Multioutput MTS

v0441

Update after hiatus

v0390

• Add quantile regression to MTS class : 2 ways
  See examples/mts_timeseries_quantile.py

v0370

New version CustomBackPropRegressor (see https://docs.techtonique.net/nnetsauce/nnetsauce.html#CustomBackPropRegressor) and ElasticNet2Regressor (see https://docs.techtonique.net/nnetsauce/nnetsauce.html#ElasticNet2Regressor)

CustomBackPropRegressor

import nnetsauce as ns 
import numpy as np 
from sklearn.metrics import root_mean_squared_error
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_diabetes, fetch_california_housing
from sklearn.linear_model import Ridge
from time import time 

load_datasets = [load_diabetes(), fetch_california_housing()]

datasets_names = ["diabetes", "housing"]

for i, data in enumerate(load_datasets):
    
    X = data.data
    y= data.target

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 13)

    regr = ns.CustomBackPropRegressor(base_model=Ridge(), 
                                      type_grad="finitediff")

    start = time()
    regr.fit(X_train, y_train)
    preds = regr.predict(X_test)
    print("Elapsed: ", time()-start)

    print(f"RMSE for {datasets_names[i]} : {root_mean_squared_error(preds, y_test)}")

    regr = ns.CustomBackPropRegressor(base_model=Ridge(), 
                                      type_grad="finitediff", 
                                      type_loss="quantile")

    start = time()
    regr.fit(X_train, y_train)
    preds = regr.predict(X_test)
    print("Elapsed: ", time()-start)

    print(f"RMSE for {datasets_names[i]} : {root_mean_squared_error(preds, y_test)}")

ElasticNet2Regressor

import nnetsauce as ns 
import numpy as np 
from sklearn.metrics import root_mean_squared_error
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_diabetes, fetch_california_housing
from time import time 

load_datasets = [load_diabetes(), fetch_california_housing()]

datasets_names = ["diabetes", "housing"]

for i, data in enumerate(load_datasets):
    
    X = data.data
    y= data.target

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .2, random_state = 13)

    regr = ns.ElasticNet2Regressor(solver="lbfgs")

    start = time()
    regr.fit(X_train, y_train)
    preds = regr.predict(X_test)
    print("Elapsed: ", time()-start)

    print(f"RMSE for {datasets_names[i]} : {root_mean_squared_error(preds, y_test)}")

    regr = ns.ElasticNet2Regressor(solver="cd")

    start = time()
    regr.fit(X_train, y_train)
    preds = regr.predict(X_test)
    print("Elapsed: ", time()-start)

    print(f"RMSE for {datasets_names[i]} : {root_mean_squared_error(preds, y_test)}")

v0360

Fix _estimator_type = "classifier"

v0353

v0.35.3

fix PredictionInterval -- fit first

v0352

• Machine Learning with ARCH effects for time series forecasting
• AIC for class CustomRegressor and PredictionInterval

v0352

• Machine Learning with ARCH effects for time series forecasting
• AIC, AICc, BIC is class CustomRegressor