smazzanti

feature_names = [
   'linear',            # 1
   'nonlinear_square',  # 2
   'nonlinear_sin',     # 3
   'interaction_1',     # 4
   'interaction_2',     # 5
   'interaction_3',     # 6
   'noise_1',           # 7
   'noise_2',           # 8
   'noise_3',           # 9

smazzanti

def X2y(X, with_error = True):
    
    # functional form of the dependence between y and X
    y_star = X['linear'] + X['nonlinear_square'] ** 2 + np.sin(3 * X['nonlinear_sin']) + (X['interaction_1'] * X['interaction_2'] * X['interaction_3'])
    
    # add random error called epsilon (this will be used for creating y)
    if with_error:
        np.random.seed(0)
        epsilon = np.random.normal(0, .1, len(y_star))
        return y_star + epsilon

smazzanti

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split

# make X and y
np.random.seed(0)
X = pd.DataFrame(np.random.normal(size = (20_000, len(feature_names))), columns = feature_names)
y = X2y(X, with_error = True)

# make X_trn, X_tst, y_trn, y_tst

smazzanti

from sklearn.dummy import DummyRegressor
from sklearn.linear_model import LinearRegression
from sklearn.neighbors import KNeighborsRegressor
from sklearn.svm import SVR
from sklearn.ensemble import RandomForestRegressor
from xgboost import XGBRegressor
from lightgbm import LGBMRegressor

# dictionary of models that will be used for comparison
models = {

smazzanti

# define a sklearn compatible wrapper for our data generating function
class UnbeatableRegressor():
    
    def __init__(self):
        pass
    
    def fit(self, X, y):
        pass  
    
    def predict(self, X):

smazzanti

from sklearn.metrics import mean_absolute_error
from eli5.sklearn import PermutationImportance

mae = pd.DataFrame(columns = ['train', 'test'])
fi = pd.DataFrame(columns = feature_names)

for model_name in list(models.keys()):

    # fit model
    models[model_name].fit(X_trn, y_trn)

smazzanti

from boruta import BorutaPy

# instantiate random forest
forest = RandomForestRegressor(n_jobs = -1, max_depth = 5)

# fit boruta
boruta_selector = BorutaPy(forest, n_estimators = 'auto', random_state = 0)
boruta_selector.fit(np.array(X_trn), np.array(y_trn))

# store results

smazzanti

from confusion_viz import ConfusionViz

y_test = [0, 0, 1, 0, 1]
probas_test = [.1, .2, .5, .7, .9]

conf_viz = ConfusionViz()

conf_viz.fit(
    y_true = y_test, 
    probas_pred = probas_test

smazzanti

! pip install git+https://github.com/smazzanti/confusion_viz

from confusion_viz import ConfusionViz

conf_viz = ConfusionViz()
conf_viz.fit(y_true, probas_pred)
conf_viz.show()

smazzanti

!pip install git+https://github.com/smazzanti/confusion_viz