classes = [[x for x, y in zip(X, Y) if y == c] for c in np.unique(Y)]
self.model = self.stat(classes)
return self
# calculate the gaussian probability given mean and sd
def gaussian(self, x, mu, sd):
if sd == 0: # when sd is 0 emulate a Dirac delta function
if x == mu:
return 1;
else:
return 0;
exponent = np.exp(-((x - mu)**2/(2*sd**2)))
gaussian_prob = exponent/(np.sqrt(2*np.pi)*sd)
if gaussian_prob == 0:
return 0
return np.log(gaussian_prob)
# predict an unlabeled dataset
def predict(self, X):
prob_mat = [[sum(self.gaussian(i, *s) for s, i in zip(stat, x)) for stat in self.model] for x in X]
return np.argmax(prob_mat, axis=1)
if __name__ == '__main__':
train_x, train_y, test_x, test_y = util.load_mnist()
model = GaussianNB()
model.fit(train_x, train_y)
pred = model.predict(test_x)
cm = util.confusion_matrix(test_y, pred)
print(cm)
print(util.accuracy(cm))
def test_confusion():
x = util.confusion_matrix("toy.csv", "a", "b")
assert np.array_equal(x[0], np.matrix([[0.75, 0.25], [0, 1.0]])) == True
assert x[1] == ["0", "6"]
writer.add_scalar('loss/text', mean_losses[0], global_idx)
writer.add_scalar('loss/gate', mean_losses[1], global_idx)
writer.add_scalar('loss_total', mean_losses[2], global_idx)
writer.add_scalar('grad_norm/enc', mean_metrics[0], global_idx)
writer.add_scalar('grad_norm/dec', mean_metrics[1], global_idx)
writer.add_scalar('alpha/enc', mean_metrics[2], global_idx)
writer.add_scalar('alpha/dec', mean_metrics[3], global_idx)
writer.add_scalar('gate_val', mean_metrics[4], global_idx)
mean_losses = np.zeros_like(mean_losses)
mean_metrics = np.zeros_like(mean_metrics)
summ_counter = 0
writer.add_figure(
'batch_CM',
confusion_matrix(text_out, text_data, [*_symbol_to_id]),
global_idx)
writer.add_text(
f'text/input',
text_id_to_string(text_data[:4], text_len=text_len[:4]),
global_idx)
gate_out[:, -1] = 1
text_len = (gate_out.squeeze(-1) > .9).to(torch.float)
text_len = text_len / (
torch.arange(gate_out.size(1), device=device) + 1.)
text_len = text_len.argmax(-1) + 1
text_out = text_out.argmax(-1)
writer.add_text(
def test_confusion():
x = util.confusion_matrix('toy.csv', 'a', 'b')
assert np.array_equal(x[0], np.matrix([[0.75, 0.25], [0, 1.0]])) == True
assert x[1] == ['0', '6']