def linear_simple(x, W, b=None):
x, W = as_variable(x), as_variable(W)
t = matmul(x, W)
if b is None:
return t
y = t + b
t.data = None
return y
def softmax_cross_entropy_simple(x, t):
x, t = as_variable(x), as_variable(t)
N = x.shape[0]
p = softmax_simple(x)
p = clip(p, 1e-15, 1.0) # To avoid log(0)
log_p = log(p)
tlog_p = log_p[np.arange(N), t.data]
y = -1 * sum(tlog_p) / N
return y
def softmax_simple(x, axis=1):
x = x - x.max(axis=axis, keepdims=True)
# x = x
x = as_variable(x)
y = exp(x)
sum_y = sum(y, axis=axis, keepdims=True)
return y / sum_y
def forward(self, x):
y = x - x.max(axis=self.axis, keepdims=True)
# xのtypeは<class 'numpy.ndarray'>
y = as_variable(y)
y = exp(y)
sum_y = y.sum(axis=self.axis, keepdims=True)
return y / sum_y
def reshape(x, shape):
if x.shape == shape:
return as_variable(x)
return Reshape(shape)(x)
def accuracy(y, t):
y, t = as_variable(y), as_variable(t)
pred = y.data.argmax(axis=1).reshape(t.shape)
result = (pred == t.data)
acc = result.mean()
return Variable(as_array(acc))
def sum_to(x, shape):
if x.shape == shape:
return as_variable(x)
return SumTo(shape)(x)
def broadcast_to(x, shape):
if x.shape == shape:
return as_variable(x)
return BroadcastTo(shape)(x)