def get_batch(self, x, t, batch_size, time_size):
batch_x = np.empty((batch_size, time_size), dtype='i')
batch_t = np.empty((batch_size, time_size), dtype='i')
data_size = len(x)
jump = data_size // batch_size
offsets = [i * jump for i in range(batch_size)] # バッチの各サンプルの読み込み開始位置
for time in range(time_size):
for i, offset in enumerate(offsets):
batch_x[i, time] = x[(offset + self.time_idx) % data_size]
batch_t[i, time] = t[(offset + self.time_idx) % data_size]
self.time_idx += 1
return batch_x, batch_t
def backward(self, dout=1):
_, T = self.xs_shape
dxs = np.empty(self.xs_shape, dtype='f')
dout *= 1 / T
for t in range(T):
layer = self.layers[t]
dxs[:, t] = layer.backward(dout)
return dxs
def backward(self, dout=1):
layers, xs = self.cache
_, T, _ = xs.shape
dxs = np.empty(xs.shape, dtype='f')
dout *= 1 / T
for t in range(T):
layer = layers[t]
dxs[:, t, :] = layer.backward(dout)
return dxs
def forward(self, xs):
N, T, _ = xs.shape
_, M = self.W.shape
self.layers = []
out = np.empty((N, T, M), dtype='f')
for t in range(T):
layer = Affine(self.W, self.b)
out[:, t, :] = layer.forward(xs[:, t, :])
self.layers.append(layer)
return out
def forward(self, xs):
N, T = xs.shape
V, D = self.W.shape
out = np.empty((N, T, D), dtype='f')
self.layers = []
for t in range(T):
layer = Embedding(self.W)
out[:, t, :] = layer.forward(xs[:, t])
self.layers.append(layer)
return out
def forward(self, xs):
N, T = xs.shape # バッチサイズ、T個の時系列データ
_, D = self.W.shape # vocab_size, wordvec_size
out = np.empty((N, T, D), dtype='f')
self.layers = []
for t in range(T):
layer = Embedding(self.W)
out[:, t, :] = layer.forward(xs[:, t])
self.layers.append(layer)
return out
def backward(self, dout):
N, T, _ = dout.shape
D, _ = self.W.shape
dxs = np.empty((N, T, D), dtype='f')
self.dW, self.db = 0, 0
for t in range(T):
layer = self.layers[t]
dxs[:, t, :] = layer.backward(dout[:, t, :])
self.dW += layer.dW
self.db += layer.db
return dxs
def forward(self, xs):
N, T, _ = xs.shape
H, _ = self.Wh.shape
self.layers = []
hs = np.empty((N, T, H), dtype='f')
if not self.stateful or self.h is None:
self.h = np.zeros((N, H), dtype='f')
for t in range(T):
layer = GRU(self.Wx, self.Wh)
self.h = layer.forward(xs[:, t, :], self.h)
hs[:, t, :] = self.h
self.layers.append(layer)
return hs
def forward(self, xs):
Wx, Wh, b = self.params
N, T, D = xs.shape
D, H = Wx.shape
self.layers = []
hs = np.empty((N, T, H), dtype='f')
if not self.stateful or self.h is None:
self.h = np.zeros((N, H), dtype='f')
for t in range(T):
layer = RNN(*self.params)
self.h = layer.forward(xs[:, t, :], self.h)
hs[:, t, :] = self.h
self.layers.append(layer)
return hs
def backward(self, dhs):
N, T, _ = dhs.shape
D = self.Wx.shape[0]
dxs = np.empty((N, T, D), dtype='f')
self.dWx, self.dWh = 0, 0
dh = 0
for t in reversed(range(T)):
layer = self.layers[t]
dx, dh = layer.backward(dhs[:, t, :] + dh)
dxs[:, t, :] = dx
self.dWx += layer.dWx
self.dWh += layer.dWh
self.dh = dh
return dxs
def backward(self, dhs):
Wx, Wh, b = self.params
N, T, H = dhs.shape
D = Wx.shape[0]
dxs = np.empty((N, T, D), dtype='f')
dh, dc = 0, 0
grads = [0, 0, 0]
for t in reversed(range(T)):
layer = self.layers[t]
dx, dh, dc = layer.backward(dhs[:, t,:] + dh, dc)
dxs[:, t, :] = dx
for i, grad in enumerate(layer.grads):
grads[i] += grad
for i, grad in enumerate(grads):
self.grads[i][...] = grad
self.dh = dh
return dxs
def forward(self, xs):
_, Wh, _ = self.params
N, T, _ = xs.shape
H = Wh.shape[0]
self.layers = []
hs = np.empty((N, T, H), dtype='f')
if not self.stateful or self.h is None:
self.h = np.zeros((N, H), dtype='f')
if not self.stateful or self.c is None:
self.c = np.zeros((N, H), dtype='f')
for t in range(T):
layer = LSTM(*self.params)
self.h, self.c = layer.forward(xs[:, t, :], self.h, self.c)
hs[:, t, :] = self.h
self.layers.append(layer)
return hs