class TestBoost(unittest.TestCase):
def setUp(self):
self.clf = AdaBoost(n_learners=3)
self.clf.fit(X, y)
def testAll(self):
# =====================================================
# prediction test
# =====================================================
pred = self.clf.predict(X)
for (p, yi) in zip(pred, y):
self.assertAlmostEqual(p, yi)
# =====================================================
# score test
# =====================================================
score = self.clf.score(X, y)
self.assertAlmostEqual(score, 1.0)
# =====================================================
# alphas test
# =====================================================
alphas = self.clf.alpha
self.assertAlmostEqual(alphas[0], 0.42364893019360172)
self.assertAlmostEqual(alphas[1], 0.64964149206513044)
self.assertAlmostEqual(alphas[2], 0.92291334524916524)
# =====================================================
# stage_score test
# =====================================================
staged_score = self.clf.staged_score(X, y)
self.assertAlmostEqual(staged_score[0], 0.7)
self.assertAlmostEqual(staged_score[1], 0.7)
self.assertAlmostEqual(staged_score[2], 1.0)
def Question3():
depths = [1, 2, 3, 4]
num_learners = [1, 10, 50, 100, 300]
data = FoursAndNines("../data/mnist.pkl.gz")
print 'Size of training set', data.x_train.shape[0]
print 'Size of validation set', data.x_valid.shape[0]
accuracy = []
clf = AdaBoost(n_learners=max(num_learners), base=MultinomialNB())
print 'Fitting classifier...'
clf.fit(data.x_train, data.y_train)
print 'Calculating scores...'
temp_learners = clf.learners
for num_learner in num_learners:
clf.learners = temp_learners[:num_learner]
print 'Number of learners', len(clf.learners)
score = clf.score(data.x_valid, data.y_valid)
print 'Score at iteration %d is %f' % (num_learner, score)
accuracy.append(score)
plt.plot(num_learners, accuracy, 'o')
plt.title(
'Accuracy of Multinomial Naive Bayes\nvs the number of weak learners')
plt.ylabel('Accuracy')
plt.xlabel('Number of weak learners')
plt.savefig('q3.png')
def Question2():
depths = [1, 2, 3, 4]
num_learners = [1, 10, 50, 100, 300]
data = FoursAndNines("../data/mnist.pkl.gz")
print 'Size of training set', data.x_train.shape[0]
print 'Size of validation set', data.x_valid.shape[0]
for depth in depths:
print 'Tree depth', depth
accuracy = []
clf = AdaBoost(n_learners=max(num_learners),
base=DecisionTreeClassifier(max_depth=depth,
criterion="entropy"))
print 'Fitting classifier...'
clf.fit(data.x_train, data.y_train)
print 'Calculating scores...'
temp_learners = clf.learners
for num_learner in num_learners:
clf.learners = temp_learners[:num_learner]
print 'Number of learners', len(clf.learners)
# score = clf.score(data.x_valid, data.y_valid)
score = clf.score(data.x_train, data.y_train)
print 'Score at iteration %d with depth %d is %f' % (num_learner,
depth, score)
accuracy.append(score)
plt.plot(num_learners, accuracy, 'o', label=str(depth))
plt.title(
'Accuracy of decision tree boosting vs the\nnumber of weak learners at a certain depth with training set'
)
plt.ylabel('Accuracy')
plt.xlabel('Number of weak learners')
plt.legend(title='Depth', numpoints=1)
plt.savefig('q2_train_accuracy.png')
def setUp(self):
self.clf = AdaBoost(n_learners=3)
self.clf.fit(X, y)