def assignment3p1():
print('Assignment 3.1')
print('Use PCA to reduce the datasets to 2D and plot the boundaries.')
for dataset_name in dataset.DatasetNames:
samples, labels = dataset.load_dataset(dataset_name)
training_samples, training_labels, test_samples, test_labels = \
dataset.split_dataset(samples=samples,
labels=labels,
train_fraction=0.5,
balance_classes=True,
seed=0)
pca = decomposition.PCA(n_components=2)
pca.fit(training_samples)
training_samples = pca.transform(training_samples)
test_samples = pca.transform(test_samples)
classifier = bc.BayesClassifier.train(samples=training_samples,
labels=training_labels,
naive=False)
test_predictions = classifier.classify(samples=test_samples)
test_accuracy = bc.evaluate_accuracy(predictions=test_predictions,
labels=test_labels)
# Plot the classification of the test samples.
fig, (ax1, ax2) = plt.subplots(1, 2)
plotting.plot_samples_2d(ax=ax1,
samples=test_samples,
labels=test_labels)
plotting.plot_gaussians(ax=ax1,
labels=test_labels,
mu=classifier.mu,
sigma=classifier.sigma)
ax1.legend()
ax1.set_title('Ground truth')
plotting.plot_samples_2d(ax=ax2,
samples=test_samples,
labels=test_predictions)
plotting.plot_gaussians(ax=ax2,
labels=test_predictions,
mu=classifier.mu,
sigma=classifier.sigma)
plotting.plot_boundaries(ax=ax2, classifier=classifier, grid_size=1000)
ax2.legend()
ax2.set_title('Prediction')
fig.suptitle('Assignment 3.1\n'
'Dataset: {}\n'
'Accuracy: {:.3f}'.format(dataset_name.value,
test_accuracy))
plt.show()
plt.close()
def assignment1():
print('Assignment 1')
print('Compute the Maximum Likelihood estimation '
'of a synthetic Gaussian dataset.')
# Generate a dataset.
samples, labels = make_blobs(n_samples=200,
centers=5,
n_features=2,
random_state=0)
mu, sigma = bc.maximum_likelihood_estimator(samples, labels, naive=False)
plt.figure()
ax = plt.gca()
plotting.plot_samples_2d(ax=ax, samples=samples, labels=labels)
plotting.plot_gaussians(ax=ax, labels=labels, mu=mu, sigma=sigma)
plt.title('Assignment 1')
ax.legend()
plt.show()
plt.close()
def assignment6p1():
print('Assignment 6.1')
print('Boosted trees, using Sklearn implementation: '
'plot the boundaries.')
for dataset_name in dataset.DatasetNames:
samples, labels = dataset.load_dataset(dataset_name)
pca = decomposition.PCA(n_components=2)
pca.fit(samples)
samples = pca.transform(samples)
(training_samples, training_labels, test_samples,
test_labels) = dataset.split_dataset(samples=samples,
labels=labels,
train_fraction=0.7,
balance_classes=True,
seed=0)
weak_classifier = \
tree_cls.SklearnDecisionTreeClassifierWrapper.train(
samples=training_samples,
labels=training_labels,
weights=None)
weak_predictions = weak_classifier.classify(samples=test_samples)
weak_accuracy = np.mean(weak_predictions == test_labels)
boost_classifier = boost_cls.BoostClassifier.train(
classifier_class=tree_cls.SklearnDecisionTreeClassifierWrapper,
samples=training_samples,
labels=training_labels,
num_iters=10)
boost_predictions = boost_classifier.classify(samples=test_samples)
boost_accuracy = np.mean(boost_predictions == test_labels)
# Plot the classification of the test samples.
fig, (ax1, ax2, ax3) = plt.subplots(1, 3)
plotting.plot_samples_2d(ax=ax1,
samples=test_samples,
labels=test_labels)
ax1.legend()
ax1.set_title('Ground truth')
plotting.plot_samples_2d(ax=ax2,
samples=test_samples,
labels=weak_predictions)
plotting.plot_boundaries(ax=ax2,
classifier=weak_classifier,
grid_size=1000)
ax2.legend()
ax2.set_title('Weak classifier')
plotting.plot_samples_2d(ax=ax3,
samples=test_samples,
labels=boost_predictions)
plotting.plot_boundaries(ax=ax3,
classifier=boost_classifier,
grid_size=1000)
ax3.legend()
ax3.set_title('Boosted classifier')
fig.suptitle('Assignment 5.2\n'
'Dataset: {}\n'
'Weak classifier accuracy: {:.3f}\n'
'Boosted classifier accuracy: {:.3f}'.format(
dataset_name.value, weak_accuracy, boost_accuracy))
plt.show()
plt.close()
def assignment2():
print('Assignment 2')
print('Compute the prior and classify using the Bayesian rule.')
dataset_size = 200
# Generate a dataset.
samples, labels = make_blobs(n_samples=2 * dataset_size,
centers=5,
n_features=2,
random_state=0)
training_samples = samples[:dataset_size]
test_samples = samples[dataset_size:]
training_labels = labels[:dataset_size]
test_labels = labels[dataset_size:]
bayes_classifier = bc.BayesClassifier.train(samples=training_samples,
labels=training_labels,
naive=False)
# Evaluate on the training data.
training_predictions = bayes_classifier.classify(training_samples)
training_accuracy = bc.evaluate_accuracy(predictions=training_predictions,
labels=training_labels)
print('Training accuracy: {:.3f}'.format(training_accuracy))
# Evaluate on the test data.
test_predictions = bayes_classifier.classify(test_samples)
test_accuracy = bc.evaluate_accuracy(predictions=test_predictions,
labels=test_labels)
print('Test accuracy: {:.3f}'.format(test_accuracy))
# Plot the classification of the test samples.
fig, (ax1, ax2) = plt.subplots(1, 2)
plotting.plot_samples_2d(ax=ax1, samples=test_samples, labels=test_labels)
plotting.plot_gaussians(ax=ax1,
labels=test_labels,
mu=bayes_classifier.mu,
sigma=bayes_classifier.sigma)
ax1.legend()
ax1.set_title('Ground truth')
plotting.plot_samples_2d(ax=ax2,
samples=test_samples,
labels=test_predictions)
plotting.plot_gaussians(ax=ax2,
labels=test_predictions,
mu=bayes_classifier.mu,
sigma=bayes_classifier.sigma)
plotting.plot_boundaries(ax=ax2,
classifier=bayes_classifier,
grid_size=1000)
ax2.legend()
ax2.set_title('Prediction')
fig.suptitle('Assignment 2 - Test samples\n'
'Accuracy: {:.3f}'.format(test_accuracy))
plt.show()
plt.close()