def initialize_set_shape(self):
parname, parshape = unpack(self.parent.items())
# Shape should be (batch_size, num_channels, x, y)
pool_size = totuple(self.pool_size)
pool_stride = totuple(self.pool_stride)
if self.ignore_border:
newx = (parshape[2] - pool_size[0]) // pool_stride[0] + 1
newy = (parshape[3] - pool_size[1]) // pool_stride[1] + 1
else:
if pool_stride[0] > pool_size[0]:
newx = (parshape[2] - 1) // pool_stride[0] + 1
else:
newx = max(
0,
(parshape[2] - 1 - pool_size[0]) // pool_stride[0] + 1) + 1
if pool_stride[1] > pool_size[1]:
newy = (parshape[3] - 1) // pool_stride[1] + 1
else:
newy = max(
0,
(parshape[3] - 1 - pool_size[1]) // pool_stride[1] + 1) + 1
outshape = (parshape[0], parshape[1], newx, newy)
self.outshape = outshape
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data
] #len(self.data[0])=100 I imagine because of [X]
label_batches = [mat[start:end] for mat in self.labels]
#mask = self.create_mask(batches[0].swapaxes(0, 1))
len_batches = len(batches[0].shape) # 1 because (20,)
if (len_batches <= 1):
mask = self.create_mask(batches[0])
else:
mask = self.create_mask(batches[0].swapaxes(0, 1))
batches = [self.zero_pad(batch) for batch in batches]
#label_mask = self.create_mask(label_batches[0].swapaxes(0, 1))
len_label_batches = len(label_batches[0].shape)
if (len_label_batches <= 1):
label_mask = self.create_mask(label_batches[0])
else:
label_mask = self.create_mask(label_batches[0].swapaxes(0, 1))
label_batches = [self.zero_pad(batch) for batch in label_batches]
if self.cond:
return totuple([batches[0], mask, label_batches[0], label_mask])
else:
return totuple([batches[0], mask])
'''
def initialize_set_shape(self):
parname, parshape = unpack(self.parent.items())
# Shape should be (batch_size, num_channels, x, y)
pool_size = totuple(self.pool_size)
pool_stride = totuple(self.pool_stride)
if self.ignore_border:
newx = (parshape[2] - pool_size[0]) // pool_stride[0] + 1
newy = (parshape[3] - pool_size[1]) // pool_stride[1] + 1
else:
if pool_stride[0] > pool_size[0]:
newx = (parshape[2] - 1) // pool_stride[0] + 1
else:
newx = max(0, (parshape[2] - 1 - pool_size[0]) //
pool_stride[0] + 1) + 1
if pool_stride[1] > pool_size[1]:
newy = (parshape[3] - 1) // pool_stride[1] + 1
else:
newy = max(0, (parshape[3] - 1 - pool_size[1]) //
pool_stride[1] + 1) + 1
outshape = (parshape[0], parshape[1], newx, newy)
self.outshape = outshape
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
mask = self.create_mask(batches[0].swapaxes(0, 1))
if self.load_spk_info:
batches = [self.zero_pad(batch) for batch in batches[:-1]]
spk = batches[-1]
return totuple(batches + [spk, mask])
else:
batches = [self.zero_pad(batch) for batch in batches]
return totuple(batches + [mask])
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
label_batches = [mat[start:end] for mat in self.labels]
mask = self.create_mask(batches[0].swapaxes(0, 1))
batches = [self.zero_pad(batch) for batch in batches]
label_mask = self.create_mask(label_batches[0].swapaxes(0, 1))
label_batches = [self.zero_pad(batch) for batch in label_batches]
if self.cond:
return totuple([batches[0], mask, label_batches[0], label_mask])
else:
return totuple([batches[0], mask])
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
label_batches = [mat[start:end] for mat in self.labels]
mask = self.create_mask(batches[0].swapaxes(0, 1))
batches = [self.zero_pad(batch) for batch in batches]
label_mask = self.create_mask(label_batches[0].swapaxes(0, 1))
label_batches = [self.zero_pad(batch) for batch in label_batches]
if self.cond:
return totuple([batches[0], mask, label_batches[0], label_mask])
else:
return totuple([batches[0], mask])
def slices(self, start, end):
batch = np.array(self.data[self.idx[start:end]], dtype=theano.config.floatX)
batch -= self.X_mean
batch /= self.X_std
batch = np.asarray([segment_axis(x, self.inpsz, 0) for x in batch])
batch = batch.transpose(1, 0, 2)
return totuple(batch)
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
#mask = tolist(self.create_mask(batches[0]).swapaxes(0, 1))
#batches = [self.zero_pad(batch) for batch in batches]
x, x_mask = self.create_mask_and_zero_pad(batches[0])
y, y_mask = self.create_mask_and_zero_pad(batches[1])
#return totuple(batches + mask)
return totuple([x, y, y_mask])
def slices(self, start, end):
batch = np.array(self.data[start:end], dtype=theano.config.floatX)
batch -= self.X_mean
batch /= self.X_std
batch = np.asarray([segment_axis(x, self.frame_size, 0) for x in batch])
ipdb.set_trace()
batch = batch.transpose(1, 0, 2)
return totuple(batch)
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
#mask = tolist(self.create_mask(batches[0]).swapaxes(0, 1))
#batches = [self.zero_pad(batch) for batch in batches]
x, x_mask = self.create_mask_and_zero_pad(batches[0])
y, y_mask = self.create_mask_and_zero_pad(batches[1])
#return totuple(batches + mask)
return totuple([x, y, y_mask])
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
if self.use_spec:
batches[0] = self._use_spec(batches[0])
batches[0] = self._log_magnitude(batches[0])
batches[0] = self._concatenate(batches[0])
else:
batches[0] -= self.X_mean
batches[0] /= self.X_std
if self.use_window:
batches[0] = self._use_window(batches[0])
else:
batches[0] = np.asarray([segment_axis(x, self.frame_size, 0) for x in batches[0]])
mask = self.create_mask(batches[0].swapaxes(0, 1))
if self.load_spk_info:
batches = [self.zero_pad(batch) for batch in batches[:-1]]
spk = batches[-1]
return totuple(batches + [spk, mask])
else:
batches = [self.zero_pad(batch) for batch in batches]
return totuple(batches + [mask])
def slices(self, start, end):
batch = np.array(self.data[start:end], dtype=theano.config.floatX)
if self.use_spec:
batch = self._use_spec(batch)
batch = self._log_magnitude(batch)
batch = self._concatenate(batch)
else:
batch -= self.X_mean
batch /= self.X_std
if self.use_window:
batch = self._use_window(batch)
else:
batch = np.asarray([segment_axis(x, self.frame_size, 0) for x in batch])
batch = batch.transpose(1, 0, 2)
return totuple(batch)
def slices(self, start, end):
batch = np.array(self.data[start:end], dtype=theano.config.floatX)
if self.use_spec:
batch = self.apply_fft(batch)
batch = self.log_magnitude(batch)
batch = self.concatenate(batch)
else:
batch -= self.X_mean
batch /= self.X_std
if self.use_window:
batch = self.apply_window(batch)
else:
batch = np.asarray(
[segment_axis(x, self.frame_size, 0) for x in batch])
batch = batch.transpose(1, 0, 2)
return totuple(batch)
def __init__(self,
parshape=[],
outshape=None,
filtershape=None,
tied_bias=True,
step_size=(1, 1),
border_mode='valid',
**kwargs):
super(Conv2DLayer, self).__init__(**kwargs)
# Shape should be (batch_size, num_channels, x, y)
if (outshape is None and filtershape is None) or\
(outshape is not None and filtershape is not None):
raise ValueError("Either outshape or filtershape should be given,\
but don't provide both of them.")
self.outshape = outshape
self.filtershape = filtershape
self.tied_bias = tied_bias
self.step_size = totuple(step_size)
self.border_mode = border_mode
for i, par in enumerate(tolist(self.parent.keys())):
if len(parshape) != 0:
self.parent[par] = parshape[i]
def __init__(self,
parshape=[],
outshape=None,
filtershape=None,
tied_bias=True,
step_size=(1, 1),
border_mode='valid',
**kwargs):
super(Conv2DLayer, self).__init__(**kwargs)
# Shape should be (batch_size, num_channels, x, y)
if (outshape is None and filtershape is None) or\
(outshape is not None and filtershape is not None):
raise ValueError("Either outshape or filtershape should be given,\
but don't provide both of them.")
self.outshape = outshape
self.filtershape = filtershape
self.tied_bias = tied_bias
self.step_size = totuple(step_size)
self.border_mode = border_mode
for i, par in enumerate(tolist(self.parent.keys())):
if len(parshape) != 0:
self.parent[par] = parshape[i]
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
mask = tolist(self.create_mask(batches[0].swapaxes(0, 1)))
batches = [self.zero_pad(batch) for batch in batches]
return totuple(batches + mask)
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
mask = self.create_mask(batches[0].swapaxes(0, 1))
batches = [self.zero_pad(batch) for batch in batches]
return totuple([batches[0], mask])
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
return totuple(batches)
def slices(self, start, end):
batches = [mat[start:end] for mat in self.data]
return totuple(batches)
