def train(self, X, y):
"""
Parameters
----------
X : samples, shape=(num_samples, num_features)
y : labels, shape=(num_samples, )
Train this classifier over the sample (X,y)
After finish the training return the weights of the samples in the last iteration.
"""
n_samples = X.shape[0]
D = np.array([1.0 / n_samples] * n_samples)
# D = np.array([[1.0 / m] * m] * self.T)
for t in range(self.T):
gc.log(f'At T = {t}')
self.h[t] = self.WL(D, X, y)
y_hat = self.h[t].predict(X)
mask = y != y_hat
epsilon = np.matmul(D, mask)
self.w[t] = 1 / 2 * np.log((1 - epsilon) / epsilon)
D *= np.exp(-(y * y_hat * self.w[t]))
D /= np.sum(D)
def big_test(self):
gc.log("Big test")
fig = plt.figure()
fig.suptitle("SVM vs Perceptron, Accuracy Test")
plt.xlabel('Train Set Size')
plt.ylabel('Accuracy (%)')
k, n_iter = 10000, 500
M, accurs = [5, 10, 15, 25, 70], [[], []]
for m in M:
svm_accu_sum = 0
perc_accu_sum = 0
for i in range(n_iter):
while True:
train_X = self.draw_m_points(m).T
train_y = self.true_label(train_X)
if np.unique(train_y).shape[0] == 2:
break
while True:
test_X = self.draw_m_points(k).T
test_y = self.true_label(test_X)
if np.unique(test_y).shape[0] == 2:
break
svm = self._svm.fit(train_X, np.ravel(train_y.T))
perc_w = self._perc.fit(train_X, train_y)
svm_accu = self.get_svm_accu(svm, test_X, test_y)
perc_accu = self.get_perc_accu(perc_w, test_X, test_y)
svm_accu_sum += svm_accu
perc_accu_sum += perc_accu
svm_accu_avg = svm_accu_sum / n_iter
perc_accu_avg = perc_accu_sum / n_iter
accurs[0].append(svm_accu_avg)
accurs[1].append(perc_accu_avg)
plt.plot(M, accurs[0], label='SVM')
plt.plot(M, accurs[1], label='Perceptron')
plt.legend()
self.plot_to_file('q5', 2)
train_errs = [ada.error(train_images, train_labels, t) for t in T_range]
test_errs = [ada.error(test_images, test_labels, t) for t in T_range]
fig = plt.figure()
fig.suptitle("Train vs Test error, Face Classifier")
plt.xlabel('# of Hypotheses (T)')
plt.ylabel('Error rate (%)')
plt.plot(T_range, train_errs, label='Train Error')
plt.plot(T_range, test_errs, label='Test Error')
# plt.ylim(top=0.06)
plt.legend()
plt.savefig(FIG_DIR3 + 'q17')
'TODO complete this function'
def Q18():
'TODO complete this function'
if __name__ == '__main__':
start_time = time.time()
# Q4()
# Q5()
# learner, test_X, test_y, train_X, train_y = Q8()
# Q9(learner, test_X, test_y)
# Q10(learner, train_X, train_y)
# Q12()
Q17()
gc.log('Execution took %s seconds' % (time.time() - start_time))
'TODO complete this function'
def compare_many(self):
gc.log("Comparing many")
for m in [5, 10, 15, 25, 70]:
self.compare_one(m)
def __init__(self):
gc.log("Creating comparer")
self._perc = pc.Perceptron()
self._svm = svm.SVC(C=1e10, kernel='linear')
self._mu = np.zeros([DIM])
self._sig = np.eye(DIM)