def forward(self, x):
if self.h is None:
batch_size = x.data.shape[0]
self.h = Variable(np.zeros((batch_size, self.hidden_size)))
a = self.l1(x) + self.l2(self.h)
self.h = F.tanh(a)
return self.h
def forward(x):
x = model.embed(x)
x = tanh(x)
x = model.linear(x)
x = sigmoid(x)
return x
def test_forward(self):
x = Variable(np.random.rand(1))
y = F.tanh(x)
expected = np.tanh(x.data)
self.assertTrue(np.allclose(y.data, expected))
def sigmoid(x):
return 0.5 * (tanh(x) + 1)
import numpy as np
import heapq
import matplotlib.pyplot as plt
import chainer0
from chainer0 import Function, Variable
import chainer0.functions as F
from chainer0.computational_graph import get_dot_graph
x = Variable(np.array([1.0]), name='x')
#y = F.sin(x)
#y = (y + F.exp(x) - 0.5) * y
#y.backward()
y = F.tanh(x)
y.backward()
for i in range(3):
gx = x.grad_var
x.cleargrad()
gx.backward()
txt = get_dot_graph(gx)
print(txt)
def forward(self, x):
x = F.tanh(self.l1(x))
x = F.tanh(self.l2(x))
return self.l3(x)