def output(self, x):
if self.on_memory:
ret_tensor = self.onehot_list[x.flatten()].reshape((x.shape[0], x.shape[1], self.vocab_size))
else:
ret_tensor = onehot_tensor(x, self.vocab_size)
if self.zero_index != None:
mask = T.neq(x, self.zero_index)
ret_tensor *= mask[:, :, None]
return ret_tensor
def output(self, x):
if self.on_memory:
ret_tensor = self.onehot_list[x.flatten()].reshape(
(x.shape[0], x.shape[1], self.vocab_size))
else:
ret_tensor = onehot_tensor(x, self.vocab_size)
if self.zero_index != None:
mask = T.neq(x, self.zero_index)
ret_tensor *= mask[:, :, None]
return ret_tensor
def compute_tensor(self, x):
if self.cached:
if x.ndim == 1:
ret_tensor = self.onehot_list[x]
else:
ret_tensor = self.onehot_list[x.flatten()].reshape((x.shape[0], x.shape[1], self.vocab_size))
else:
ret_tensor = onehot_tensor(x, self.vocab_size)
if self.zero_index != None:
mask = T.neq(x, self.zero_index)
if x.ndim == 1:
ret_tensor *= mask[:, None]
else:
ret_tensor *= mask[:, :, None]
if self.mask:
if x.ndim == 1:
ret_tensor *= self.mask[:, None]
else:
ret_tensor *= self.mask[:, :, None]
return ret_tensor
def compute_tensor(self, x):
if self.cached:
if x.ndim == 1:
ret_tensor = self.onehot_list[x]
else:
ret_tensor = self.onehot_list[x.flatten()].reshape(
(x.shape[0], x.shape[1], self.vocab_size))
else:
ret_tensor = onehot_tensor(x, self.vocab_size)
if self.zero_index != None:
mask = T.neq(x, self.zero_index)
if x.ndim == 1:
ret_tensor *= mask[:, None]
else:
ret_tensor *= mask[:, :, None]
if self.mask:
if x.ndim == 1:
ret_tensor *= self.mask[:, None]
else:
ret_tensor *= self.mask[:, :, None]
return ret_tensor
def output(self, x):
if self.on_memory:
return self.onehot_list[x.flatten()].reshape((x.shape[0], x.shape[1], self.vocab_size))
else:
return onehot_tensor(x, self.vocab_size)
