Module edawesome.sklearn_help

Function for visual evaluation of sklearn models.

"""
Function for visual evaluation of sklearn models.
"""

import pandas as pd

import matplotlib.pyplot as plt
import seaborn as sns

from sklearn.model_selection import cross_val_score
from sklearn.metrics import confusion_matrix, roc_auc_score, roc_curve, f1_score, auc, RocCurveDisplay

def plot_roc_curve(fpr, tpr, title=None):

    plt.plot(fpr, tpr, linewidth=2)
    plt.plot([0,1], [0,1], 'k--')

    plt.axis([0,1,0,1])
    plt.xlabel('False Positive Rate')
    plt.ylabel('True Positive Rate')
    plt.title(title)

    plt.show()

def pretty_classification_report(estimator, X, y, class_names, how='cv'):
    
    y_pred = estimator.predict(X)

    if how == 'cv':
        f_score = cross_val_score(estimator, X, y, cv=5, scoring='f1').mean()
    elif how == 'test':
        f_score = f1_score(y, y_pred)
    else:
        raise ValueError('how must be either "cv" or "test"')   
    
    cm = confusion_matrix(y, y_pred)
    cm_df = pd.DataFrame(cm, index=class_names, columns=class_names)

    # heatmap of confusion matrix
    sns.heatmap(cm_df, annot=True, cbar=None, cmap="Blues", fmt='d')
    if how == 'cv':
        plt.title(f'Cross-validated f-score: {f_score:.3f}')
    elif how == 'test':
        plt.title(f'Test f-score: {f_score:.3f}')
    plt.ylabel('True Class'), plt.xlabel('Predicted Class')
    plt.tight_layout()
    plt.show()

    if how == 'cv':
        roc_auc = cross_val_score(estimator, X, y, cv=5, scoring='roc_auc').mean()
    elif how == 'test':
        roc_auc = roc_auc_score(y, y_pred)
    else:
        raise ValueError('how must be either "cv" or "test"')

    # roc graph
    fpr, tpr, thresholds = roc_curve(y, y_pred)
    roc_auc = auc(fpr, tpr)
    assert roc_auc == roc_auc_score(y, y_pred)

    RocCurveDisplay(fpr=fpr, tpr=tpr, roc_auc=roc_auc, estimator_name='').plot()
    plt.title(f'ROC curve: {roc_auc:.3f}')

    plt.show()




def plot_predicted_actual(Y_test, name, model=None, X_test=None, Y_pred=None):

    if (Y_pred is None) and (model is not None) and (X_test is not None):
        Y_pred = model.predict(X_test)
    
    if Y_pred is None:
        raise ValueError('Y_pred or model with X_test must be provided')

    fig, ax = plt.subplots(figsize=(7, 7))
    ax.scatter(Y_test, Y_pred, alpha=0.3)
    ax.plot([Y_test.min(), Y_test.max()], [Y_test.min(), Y_test.max()], 'r--', lw=2)

    ax.set_xlabel('Measured')
    ax.set_ylabel('Predicted')

    ax.set_title(f'{name}: predicted vs measured')

    plt.show()

Functions

def plot_predicted_actual(Y_test, name, model=None, X_test=None, Y_pred=None)

Expand source code

def plot_predicted_actual(Y_test, name, model=None, X_test=None, Y_pred=None):

    if (Y_pred is None) and (model is not None) and (X_test is not None):
        Y_pred = model.predict(X_test)
    
    if Y_pred is None:
        raise ValueError('Y_pred or model with X_test must be provided')

    fig, ax = plt.subplots(figsize=(7, 7))
    ax.scatter(Y_test, Y_pred, alpha=0.3)
    ax.plot([Y_test.min(), Y_test.max()], [Y_test.min(), Y_test.max()], 'r--', lw=2)

    ax.set_xlabel('Measured')
    ax.set_ylabel('Predicted')

    ax.set_title(f'{name}: predicted vs measured')

    plt.show()

def plot_roc_curve(fpr, tpr, title=None)

Expand source code

def plot_roc_curve(fpr, tpr, title=None):

    plt.plot(fpr, tpr, linewidth=2)
    plt.plot([0,1], [0,1], 'k--')

    plt.axis([0,1,0,1])
    plt.xlabel('False Positive Rate')
    plt.ylabel('True Positive Rate')
    plt.title(title)

    plt.show()

def pretty_classification_report(estimator, X, y, class_names, how='cv')

Expand source code

def pretty_classification_report(estimator, X, y, class_names, how='cv'):
    
    y_pred = estimator.predict(X)

    if how == 'cv':
        f_score = cross_val_score(estimator, X, y, cv=5, scoring='f1').mean()
    elif how == 'test':
        f_score = f1_score(y, y_pred)
    else:
        raise ValueError('how must be either "cv" or "test"')   
    
    cm = confusion_matrix(y, y_pred)
    cm_df = pd.DataFrame(cm, index=class_names, columns=class_names)

    # heatmap of confusion matrix
    sns.heatmap(cm_df, annot=True, cbar=None, cmap="Blues", fmt='d')
    if how == 'cv':
        plt.title(f'Cross-validated f-score: {f_score:.3f}')
    elif how == 'test':
        plt.title(f'Test f-score: {f_score:.3f}')
    plt.ylabel('True Class'), plt.xlabel('Predicted Class')
    plt.tight_layout()
    plt.show()

    if how == 'cv':
        roc_auc = cross_val_score(estimator, X, y, cv=5, scoring='roc_auc').mean()
    elif how == 'test':
        roc_auc = roc_auc_score(y, y_pred)
    else:
        raise ValueError('how must be either "cv" or "test"')

    # roc graph
    fpr, tpr, thresholds = roc_curve(y, y_pred)
    roc_auc = auc(fpr, tpr)
    assert roc_auc == roc_auc_score(y, y_pred)

    RocCurveDisplay(fpr=fpr, tpr=tpr, roc_auc=roc_auc, estimator_name='').plot()
    plt.title(f'ROC curve: {roc_auc:.3f}')

    plt.show()