boosting_samples_distribution.py

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.tree import DecisionTreeClassifier

# Load data
data_train = pd.read_csv('data/Higgs_train.csv')
data_test = pd.read_csv('data/Higgs_test.csv')

# Split into NumPy arrays
X_train = data_train.iloc[:, data_train.columns != 'class'].values
y_train = data_train['class'].values
X_test = data_test.iloc[:, data_test.columns != 'class'].values
y_test = data_test['class'].values

# Single decision tree with depth 3
tree1 = DecisionTreeClassifier(max_depth=3).fit(X_train, y_train)
y_train_predicted_tree1 = tree1.predict(X_train)

# Split training data into wrongly and correctly predicted samples
y_train_predicted_tree1_bool = y_train_predicted_tree1 == y_train
X_train_correct = X_train[y_train_predicted_tree1_bool]
X_train_wrong = X_train[np.logical_not(y_train_predicted_tree1_bool)]

# plot distribution of wrongly and correctly predicted samples
fig, axs = plt.subplots(7, 4, figsize=(30, 27))
i = 0;
for ax in axs.ravel():
    sns.kdeplot(X_train_correct[:,i], ax=ax, shade=True, label='correct')    
    sns.kdeplot(X_train_wrong[:,i], ax=ax, shade=True, label='wrong')
    ax.set_title(data_train.columns[i])
    if i >= 24:        
        ax.set_xlabel('predictor value')
    if i%4 == 0:
        ax.set_ylabel('frequency')
    i += 1
fig.suptitle("Distribution of predictors' values per classification success", fontsize=20);