Last week, I (re)introduced learningmachine, an R package for Machine Learning that includes uncertainty quantification for regression and classification, and explainability through sensitivity analysis. This week, I talk about learningmachine for Python. The Python version is a port of the R package, which means:

• It’s faster to install if R is already installed on your machine (otherwise, the Python package will attempt to install R and the package dependencies by itself)
• If R and the package dependencies are not already installed it (learningmachine Python) may take a long time to get started, but ONLY the first time it’s installed and run

Not everything is ultra-smooth yet (documentation coming in a few weeks), but you can already do some advanced stuff, as shown below.

The next algorithm I’ll include in learningmachine is the Bayesian one described in this document, that learns in a way that’s most intuitive to us (online instead of batch).

Install learningmachine from GitHub (tested on macOS, OK on Posit Cloud, KO on Google Colab)

!pip install git+https://github.com/Techtonique/learningmachine_python.git --verbose

Examples

import learningmachine as lm
import numpy as np
import pandas as pd 
from sklearn.datasets import load_diabetes, load_wine
from sklearn.datasets import load_wine, load_iris, load_breast_cancer
from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split, cross_val_score
from rpy2.robjects.vectors import FloatMatrix, FloatVector, StrVector
from time import time
from sklearn.metrics import mean_squared_error
from math import sqrt


# 1. Regression

diabetes = load_diabetes()
X = pd.DataFrame(diabetes.data[:150], columns=diabetes.feature_names)
y = diabetes.target[:150]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                    random_state=1213)

print("\n ----- fitting krr ----- \n")   

fit_obj2 = lm.Regressor(method="krr", pi_method="none")
start = time()
fit_obj2.fit(X_train, y_train, lambda_=0.05) # R's `lambda` is renamed as `lambda_` in Python as `lambda` is reserved 
print("Elapsed time: ", time() - start)
print(fit_obj2.summary(X=X_test, y=y_test))

# 2. Classification

datasets = [load_wine(), load_iris(), load_breast_cancer()]

print("\n ----- fitting Kernel Ridge Regression ----- \n")   

for dataset in datasets: 
    
    print(f"Description: {dataset.DESCR}")
    X = pd.DataFrame(dataset.data, columns=dataset.feature_names)
    y = dataset.target

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                        random_state=123)
    
    fit_obj = lm.Classifier(method = "krr", 
                            pi_method="none")    

    start = time()
    fit_obj.fit(X_train, y_train, reg_lambda = 0.05)
    print("Elapsed time: ", time() - start)

    ## Compute accuracy
    print(fit_obj.summary(X=X_test, y=y_test,                           
                          class_index=0))
    

print("\n ----- fitting xgboost ----- \n")   

for dataset in datasets: 
    
    print(f"Description: {dataset.DESCR}")
    X = pd.DataFrame(dataset.data, columns=dataset.feature_names)
    y = dataset.target

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                        random_state=123)
    
    fit_obj = lm.Classifier(method = "xgboost", 
                            pi_method="kdesplitconformal",
                            type_prediction_set = 'score',
                            B=100)   
                            
    print("nb_hidden = 0 -----") # no hidden layer
    start = time()
    fit_obj.fit(X_train, y_train, nrounds=100, eta=0.05, max__depth=4, verbose=0) # dot ('.') in R parameters is replaced by '__'
    print("Elapsed time: ", time() - start)
    print(fit_obj.predict(X_test))
    print(fit_obj.summary(X=X_test, y=y_test, 
                          class_index=1)) # specify the class whose probability is of interest
    
    fit_obj = lm.Classifier(method = "xgboost", 
                            pi_method="kdesplitconformal",
                            type_prediction_set = 'score',
                            nb_hidden = 5,
                            B=100) 
                            
    print("nb_hidden = 5 -----") # hidden layer with 5 nodes 
    start = time()
    fit_obj.fit(X_train, y_train, nrounds=100, eta=0.05, max__depth=4, verbose=0) # dot ('.') in R parameters is replaced by '__'
    print("Elapsed time: ", time() - start)
    print(fit_obj.predict(X_test))
    print(fit_obj.summary(X=X_test, y=y_test, 
                          class_index=1)) # specify the class whose probability is of interest