def ShiftRight(params, inputs, **kwargs):
"""Layer to shift the tensor to the right by padding on axis 1."""
del params, kwargs
pad_widths = [(0, 0), (1, 0)]
pad_widths += [(0, 0) for _ in range(len(inputs.shape) - 2)]
padded = np.pad(inputs, pad_widths, mode='constant')
return padded[:, :-1, ...]
def ShiftRight(x, **unused_kwargs):
"""Layer to shift the tensor to the right by padding on axis 1."""
pad_widths = [(0, 0)] * len(x.shape)
pad_widths[1] = (1, 0) # Padding on axis=1
padded = np.pad(x, pad_widths, mode='constant',
constant_values=x.dtype.type(0))
return padded[:, :-1]
def ShiftRight(x, mode='train', **unused_kwargs):
"""Layer to shift the tensor to the right by padding on axis 1."""
if mode == 'predict':
# Do nothing in predict mode, as then the sequence length is 1.
return x
pad_widths = [(0, 0)] * len(x.shape)
pad_widths[1] = (1, 0) # Padding on axis=1
padded = np.pad(x, pad_widths, mode='constant',
constant_values=x.dtype.type(0))
return padded[:, :-1]
def call(self, x, params=(), **kwargs):
assert self._padding == 'VALID'
# Left pad with 0s. Applying an unmasked valid convolution on top of this
# yields a causal convolution.
# TODO(ddohan): Support strided and dilated convolutions.
rate = 1
effective_kernel_size = int((self._kernel_size[0] - 1) * rate + 1)
pad = effective_kernel_size - 1
x_leftpad = np.pad(x,
pad_width=[[0, 0], [pad, 0], [0, 0]],
mode='constant')
res = super(CausalConv, self).call(x_leftpad, params)
return res
def DiagonalGate(x, params, **kwargs):
"""Split channels in 3 parts. Shifts 1st and 3rd sections to left/right."""
del params
del kwargs
# x : [batch, 1, length, depth]
x = np.pad(
x, [(0, 0), (0, 0), (1, 1), (0, 0)], mode='constant', constant_values=0.0)
depth = x.shape[-1] // 3
assert 3 * depth == x.shape[-1], ('Depth must be divisible by 3', depth,
x.shape)
xs = [
x[:, :, :-2, :depth], x[:, :, 1:-1, depth:2 * depth],
x[:, :, 2:, 2 * depth:3 * depth]
]
return np.concatenate(xs, axis=3)
def ShiftRight(x, **unused_kwargs):
"""Layer to shift the tensor to the right by padding on axis 1."""
if not isinstance(x, (list, tuple)): # non-chunked inputs
pad_widths = [(0, 0), (1, 0)]
padded = np.pad(x, pad_widths, mode='constant')
return padded[:, :-1]
# Handling chunked inputs. Recall that the list of chunks represents a big
# sequence (the concatenation of the chunks). We want to shift that sequence,
# so we put a 0 in the beginning of the first chunk and the last element of
# that chunk is used as the new first element of the next chunk, and so on.
padded = []
last_value = np.zeros_like(x[0][:, -1])
for chunk in x:
padded_chunk = np.concatenate([last_value[:, np.newaxis], chunk],
axis=1)
last_value = chunk[:, -1]
padded.append(padded_chunk[:, :-1])
return padded
def apply_fun(params, inputs, **kwargs): del params, kwargs pad_widths = [(0, 0), (1, 0)] pad_widths += [(0, 0) for _ in range(len(inputs.shape) - 2)] padded = np.pad(inputs, pad_widths, mode='constant') return padded[:, :-1, ...]
def pad_input(x):
pad_widths = [(0, 0)] * len(x.shape)
pad_widths[-2] = (0, pad_len) # Padding on axis=-2
return np.pad(x, pad_widths, mode='constant',
constant_values=x.dtype.type(0))
def ShiftRight(x, **unused_kwargs):
"""Layer to shift the tensor to the right by padding on axis 1."""
pad_widths = [(0, 0), (1, 0)]
pad_widths += [(0, 0) for _ in range(len(x.shape) - 2)]
padded = np.pad(x, pad_widths, mode='constant')
return padded[:, :-1, ...]
