I found the _pad_2d still didn't work in the stable version v0.1.1 as mentioned in #127 when training an unconditional Wavenet. I think the following snip in train.py
if is_mulaw_quantize(hparams.input_type):
padding_value = P.mulaw_quantize(0, mu=hparams.quantize_channels)
x_batch = np.array([_pad_2d(np_utils.to_categorical(
x[0], num_classes=hparams.quantize_channels),
max_input_len, padding_value) for x in batch], dtype=np.float32)
is somewhat wrong and should be
if is_mulaw_quantize(hparams.input_type):
padding_value = P.mulaw_quantize(0, mu=hparams.quantize_channels - 1)
x_batch = np.array([_pad_2d(to_categorical(
x[0], num_classes=hparams.quantize_channels),
max_input_len, 0, padding_value) for x in batch], dtype=np.float32)
in the branch master.
I am wondering which version I should use to train an unconditional Wavenet. I am a little confused because you mentioned that the v0.1.1 is the working version and branch master may not be working.
I found the
_pad_2d
still didn't work in the stable version v0.1.1 as mentioned in #127 when training an unconditional Wavenet. I think the following snip intrain.py
is somewhat wrong and should be
in the branch master.
I am wondering which version I should use to train an unconditional Wavenet. I am a little confused because you mentioned that the v0.1.1 is the working version and branch master may not be working.