Plot top k accuracies

plot_top_k_accuracies

import matplotlib.pyplot as plt
from sklearn.datasets import load_digits
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split

plt.xkcd()

X, y = load_digits(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)

clf = LogisticRegression()

clf.fit(X_train, y_train)

y_prob = clf.predict_proba(X_test)


def top_k_accuracy(y_prob, y_true, k=5):
    top_k_classes = y_prob.argsort(axis=1)[:, -k:]
    return sum(y in top for y, top in zip(y_true, top_k_classes))/len(y_true)


def plot_top_k_accuracy(y_prob, y_true, k=5):
    top_accuracies = [100*top_k_accuracy(y_prob, y_test, i) for i in range(1, k+1)]
    plt.plot(list(range(1, k+1)), top_accuracies)
    plt.title("Top k accuracy")
    plt.xlabel("k")
    plt.ylabel("Accuracy (%)");

plot_top_k_accuracy(y_prob, y_test, 5)