def get_test_set_accuracy_with_leaf_thresholds():
train_dataset = dtlearn.get_dataset_from_file(TRAIN_DATASET_FILE_PATH)
test_dataset = dtlearn.get_dataset_from_file(TEST_DATASET_FILE_PATH)
leaf_thresholds = [2, 5, 10, 20]
accuracy_with_leaf_thresholds = []
for leaf_threshold in leaf_thresholds:
print "Running test for leaf threshold : " + str(leaf_threshold)
training_dataset_dtree = dtree.learn_dtree(train_dataset, leaf_threshold)
test_set_accuracy = dtree.test_dtree(training_dataset_dtree, test_dataset)
accuracy_with_leaf_thresholds.append(test_set_accuracy*100)
print "Test set accuracy for leaf threshold " + str(leaf_threshold) + " is " + "{0:.2f}".format(test_set_accuracy)
# Plot the graph
plt.figure()
plt.plot(leaf_thresholds, accuracy_with_leaf_thresholds, label="accuracy vs stopping threshold",marker='D', color='r')
plt.xlabel("Stopping threshold")
plt.ylabel("Test set accuracy (%)")
plt.title("Test set accuracy vs Stopping threshold")
plt.xlim(0, 22)
plt.ylim(0, 100)
plt.grid(True)
plt.legend(loc="upper left")
plt.savefig("graphs/accuracy_vs_m.png")
def get_learning_curve_with_number_of_instances():
train_dataset = dtlearn.get_dataset_from_file(TRAIN_DATASET_FILE_PATH)
test_dataset = dtlearn.get_dataset_from_file(TEST_DATASET_FILE_PATH)
leaf_threshold = 4
training_set_sizes_repetitions = {}
training_set_sizes_repetitions[25] = 10
training_set_sizes_repetitions[50] = 10
training_set_sizes_repetitions[100] = 10
training_set_sizes_repetitions[200] = 1
test_set_accuracies_with_size = {}
avg_test_data_accuracy, min_test_data_accuracy, max_test_data_accuracy = [], [], []
for size in sorted(training_set_sizes_repetitions.keys()):
num_repetitions = training_set_sizes_repetitions[size]
test_set_accuracies = []
for counter in xrange(0, num_repetitions):
stratified_training_data_set = get_stratified_random_data_set(train_dataset, size)
training_dataset_dtree = dtree.learn_dtree(stratified_training_data_set, leaf_threshold)
test_set_accuracy = dtree.test_dtree(training_dataset_dtree, test_dataset)
test_set_accuracies.append(test_set_accuracy*100)
test_set_accuracies_with_size[size] = test_set_accuracies
avg_test_data_accuracy.append(sum(test_set_accuracies)/float(len(test_set_accuracies)))
min_test_data_accuracy.append(min(test_set_accuracies))
max_test_data_accuracy.append(max(test_set_accuracies))
training_data_set_sizes = sorted(training_set_sizes_repetitions.keys())
print "\n\nTraining data set sizes : " + str(training_data_set_sizes)
print "Average test set accuracy : " + str(avg_test_data_accuracy)
print "Minimum test set accuracy : " + str(min_test_data_accuracy)
print "Maximum test set accuracy : " + str(max_test_data_accuracy)
print "Test Set accuracies vs Training dataset size : " + str(test_set_accuracies_with_size)
# Plot the graph
plt.figure()
plt.plot(training_data_set_sizes, avg_test_data_accuracy, label="avg accuracy vs size", marker='H')
plt.plot(training_data_set_sizes, min_test_data_accuracy, label="min accuracy vs size", marker='H')
plt.plot(training_data_set_sizes, max_test_data_accuracy, label="max accuracy vs size", marker='H')
plt.xlabel("Training set size")
plt.ylabel("Test set accuracy (%)")
plt.title("Test accuracy vs Training set size")
plt.xlim(0, 201)
plt.ylim(0, 100)
plt.grid(True)
plt.legend(loc="lower right")
plt.savefig("graphs/accuracy_vs_size.png")