def softmax(x):
"""
Softmax function.
Score to Probability.
Returns float [0.0, 1.0]
>>> x = np.array([0])
>>> softmax(x)
array([1.])
>>> x = np.array([0, 0])
>>> softmax(x)
array([0.5, 0.5])
>>> x = np.array([[0, 0], [0, 0]])
>>> softmax(x)
array([[0.5, 0.5],
[0.5, 0.5]])
"""
if x.ndim == 2:
x = x - x.max(axis=1, keepdims=True)
x = np.exp(x)
x = x / x.sum(axis=1, keepdims=True)
elif x.ndim == 1:
x = x - np.max(x)
x = np.exp(x) / np.sum(np.exp(x))
return x
def softmax(x):
if x.ndim == 2:
x = x - x.max(axis=1, keepdims=True)
x = np.exp(x)
x /= x.sum(axis=1, keepdims=True)
elif x.ndim == 1:
x = x - np.max(x)
x = np.exp(x) / np.sum(np.exp(x))
return x
def forward(self, x, t):
self.t = t
self.y = 1 / (1 + np.exp(-x))
self.loss = cross_entropy_error(np.c_[1 - self.y, self.y], self.t)
return self.loss
def sigmoid(x):
"""
Return the sigmoid of x.
One of the activation functions.
>>> sigmoid(0)
0.5
"""
return 1 / (1 + np.exp(-x))
def fit(self,
xs,
ts,
max_epoch=10,
batch_size=20,
time_size=35,
max_grad=None,
eval_interval=20):
data_size = len(xs)
max_iters = data_size // (batch_size * time_size)
self.time_idx = 0
self.ppl_list = []
self.eval_interval = eval_interval
model, optimizer = self.model, self.optimizer
total_loss = 0
loss_count = 0
start_time = time.time()
for _ in range(max_epoch):
for iters in range(max_iters):
batch_x, batch_t = self.get_batch(xs, ts, batch_size,
time_size)
# 勾配を求め、パラメータを更新
loss = model.forward(batch_x, batch_t)
model.backward()
params, grads = remove_duplicate(model.params,
model.grads) # 共有された重みを1つに集約
if max_grad is not None:
clip_grads(grads, max_grad)
optimizer.update(params, grads)
total_loss += loss
loss_count += 1
# パープレキシティの評価
if (eval_interval
is not None) and (iters % eval_interval) == 0:
ppl = np.exp(total_loss / loss_count)
elapsed_time = time.time() - start_time
print(
'| epoch %d | iter %d / %d | time %d[s] | perplexity %.2f'
% (self.current_epoch + 1, iters + 1, max_iters,
elapsed_time, ppl))
self.ppl_list.append(float(ppl))
total_loss, loss_count = 0, 0
self.current_epoch += 1
def forward(self, x):
out = 1 / (1 + np.exp(-x))
self.out = out
return out
def sigmoid(x):
return 1 / (1 + np.exp(-x))
