learning_curve.py

""" Exploring learning curves for classification of handwritten digits """

import matplotlib.pyplot as plt
import numpy
from sklearn.datasets import *
from sklearn.cross_validation import train_test_split
from sklearn.linear_model import LogisticRegression

data = load_digits()
# print data.DESCR
num_trials = 10
train_percentages = range(5,95,5)
test_accuracies = numpy.zeros(len(train_percentages))

# train a model with training percentages between 5 and 90 (see train_percentages) and evaluate
# the resultant accuracy.
# You should repeat each training percentage num_trials times to smooth out variability
# for consistency with the previous example use model = LogisticRegression(C=10**-10) for your learner
test_accuracies=[]
for j in train_percentages:
    train_size = j / 100.0
    cumulative_train = 0
    cumulative_test = 0
    for i in range(10): #do 10 trials
        X_train, X_test, y_train, y_test = train_test_split(data.data, data.target, train_size=train_size)
        model = LogisticRegression(C=10**-20)
        model.fit(X_train, y_train)
        cumulative_train = model.score(X_train,y_train) + cumulative_train
        cumulative_test = model.score(X_test,y_test) + cumulative_test
    test_accuracies.append(cumulative_test)


fig = plt.figure()
plt.plot(train_percentages, test_accuracies)
plt.xlabel('Percentage of Data Used for Training')
plt.ylabel('Accuracy on Test Set')
plt.show()