def show_digits(epoch=0):
"""Display a 2D manifold of the digits"""
n = 15 # 15x15 digits
digit_size = 28
figure = np.zeros((digit_size * n, digit_size * n))
grid_x = norm.ppf(np.linspace(0.05, 0.95, n))
grid_y = norm.ppf(np.linspace(0.05, 0.95, n))
for i, yi in enumerate(grid_x):
for j, xi in enumerate(grid_y):
z_sample = np.array([[xi, yi]])
if use_gpu:
z_sample = dezero.cuda.as_cupy(z_sample)
with dezero.no_grad():
x_decoded = vae.decoder(z_sample)
if use_gpu:
x_decoded.data = dezero.cuda.as_numpy(x_decoded.data)
digit = x_decoded.data.reshape(digit_size, digit_size)
figure[i * digit_size: (i + 1) * digit_size,
j * digit_size: (j + 1) * digit_size] = digit
plt.figure(figsize=(10, 10))
plt.axis('off')
plt.imshow(figure, cmap='Greys_r')
plt.show()
def test_config(self):
# 学習時
self.assertTrue(Config.enable_backprop)
x = Variable(np.ones((100, 100, 100)))
y = square(square(square(x)))
y.backward()
with no_grad():
x = Variable(np.array(2.0))
y = square(x)
self.assertFalse(Config.enable_backprop)
self.assertTrue(Config.enable_backprop)
def test_MNIST(self):
max_epoch = 5
batch_size = 100
hidden_size = 1000
train_set = dezero.datasets.MNIST(train=True)
test_set = dezero.datasets.MNIST(train=False)
train_loader = DataLoader(train_set, batch_size)
test_loader = DataLoader(test_set, batch_size, shuffle=False)
#model = MLP((hidden_size, 10))
model = MLP((hidden_size, hidden_size, 10), activation=F.relu)
optimizer = optimizers.SGD().setup(model)
if dezero.cuda.gpu_enable:
train_loader.to_gpu()
model.to_gpu()
for epoch in range(max_epoch):
sum_loss, sum_acc = 0, 0
for x, t in train_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
model.cleargrads()
loss.backward()
optimizer.update()
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('epoch: {}'.format(epoch + 1))
print('train loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(train_set), sum_acc / len(train_set)))
sum_loss, sum_acc = 0, 0
with dezero.no_grad():
for x, t in test_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('test loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(test_set), sum_acc / len(test_set)))
def test_SoftmaxCrossEntorpy(self):
max_epoch = 0
batch_size = 30
hidden_size = 10
lr = 1.0
train_set = Spiral(train=True)
test_set = Spiral(train=False)
train_loader = DataLoader(train_set, batch_size)
test_loader = DataLoader(test_set, batch_size, shuffle=False)
model = MLP((hidden_size, hidden_size, hidden_size, 3))
optimizer = optimizers.SGD(lr).setup(model)
for epoch in range(max_epoch):
sum_loss, sum_acc = 0, 0
for x, t in train_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
model.cleargrads()
loss.backward()
optimizer.update()
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('epoch: {}'.format(epoch + 1))
print('train loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(train_set), sum_acc / len(train_set)))
sum_loss, sum_acc = 0, 0
with dezero.no_grad():
for x, t in test_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('test loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(test_set), sum_acc / len(test_set)))
model.clear_grads()
loss.backward()
optimizer.update()
sum_loss += float(loss.data) * len(batch_t)
avg_loss = sum_loss / data_size
print('epoch %d, loss %.2f' % (epoch + 1, avg_loss))
# Plot boundary area the model predict
h = 0.001
x_min, x_max = x[:, 0].min() - .1, x[:, 0].max() + .1
y_min, y_max = x[:, 1].min() - .1, x[:, 1].max() + .1
xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))
X = np.c_[xx.ravel(), yy.ravel()]
with dezero.no_grad():
score = model(X)
predict_cls = np.argmax(score.data, axis=1)
Z = predict_cls.reshape(xx.shape)
plt.contourf(xx, yy, Z)
# Plot data points of the dataset
N, CLS_NUM = 100, 3
markers = ['o', 'x', '^']
colors = ['orange', 'blue', 'green']
for i in range(len(x)):
c = t[i]
plt.scatter(x[i][0], x[i][1], s=40, marker=markers[c], c=colors[c])
plt.show()
for epoch in range(max_epoch):
sum_loss, sum_acc = 0, 0
for x, t in train_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
model.cleargrads()
loss.backward()
optimizer.update()
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print("epoch: {}".format(epoch + 1))
print("train loss: {:.4f}, accuracy: {:.4f}".format(
sum_loss / len(train_set), sum_acc / len(train_set)))
sum_loss, sum_acc = 0, 0
with dezero.no_grad(): # 기울기 불필요 모드
for x, t in test_loader: # 테스트용 미니배치 데이터
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t) # 테스트 데이터의 인식 정확도
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print("test loss: {:.4f}, accuracy: {:.4f}".format(
sum_loss / len(test_set), sum_acc / len(test_set)))
def test_no_grad(self):
with no_grad():
assert not Config.enable_backprop
for epoch in range(max_epoch):
sum_loss, sum_acc = 0, 0
for x, t in train_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
model.cleargrads()
loss.backward()
optimizer.update()
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('epoch: {}'.format(epoch + 1))
print('train loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(train_set), sum_acc / len(train_set)))
sum_loss, sum_acc = 0, 0
with dezero.no_grad(): # when gradients are not needed
for x, t in test_loader:
y = model(x)
loss = F.softmax_cross_entropy(y, t)
acc = F.accuracy(y, t)
sum_loss += float(loss.data) * len(t)
sum_acc += float(acc.data) * len(t)
print('test loss: {:.4f}, accuracy: {:.4f}'.format(
sum_loss / len(test_set), sum_acc / len(test_set)))