Hello all.
I know that this problem could be solved from a version perspective but it seems to me that can be quickly solved through optimization. I’m having some problems integrating two VAEs (one for image and other for audio) in the same py file.
When running in separate files (with tf 2.5 and CUDA) and with the eager execution option in the older code, it trains and generates without any problem. But when I mix them I have a problem. The VAE that is causing problems is the same as in https://github.com/musikalkemist/generating-sound-with-neural-networks/blob/main/14%20Sound%20generation%20with%20VAE/code/autoencoder.py and the bug is as follows:
Traceback (most recent call last):
File "/home/luis/Desktop/keras/full/rt.py", line 745, in <module>
audio_autoencoder = VAE(
File "/home/luis/Desktop/keras/full/rt.py", line 539, in __init__
self._build()
File "/home/luis/Desktop/keras/full/rt.py", line 624, in _build
self._build_encoder()
File "/home/luis/Desktop/keras/full/rt.py", line 689, in _build_encoder
bottleneck = self._add_bottleneck(conv_layers)
File "/home/luis/Desktop/keras/full/rt.py", line 738, in _add_bottleneck
x = Lambda(sample_point_from_normal_distribution, name="encoder_output")([self.mu, self.log_variance])
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/base_layer.py", line 969, in __call__
return self._functional_construction_call(inputs, args, kwargs,
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/base_layer.py", line 1107, in _functional_construction_call
outputs = self._keras_tensor_symbolic_call(
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/base_layer.py", line 840, in _keras_tensor_symbolic_call
return self._infer_output_signature(inputs, args, kwargs, input_masks)
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/base_layer.py", line 885, in _infer_output_signature
outputs = nest.map_structure(
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/util/nest.py", line 867, in map_structure
structure[0], [func(*x) for x in entries],
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/util/nest.py", line 867, in <listcomp>
structure[0], [func(*x) for x in entries],
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/keras_tensor.py", line 590, in keras_tensor_from_tensor
out = keras_tensor_cls.from_tensor(tensor)
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/keras/engine/keras_tensor.py", line 182, in from_tensor
type_spec = type_spec_module.type_spec_from_value(tensor)
File "/home/luis/anaconda3/envs/tensorflow/lib/python3.9/site-packages/tensorflow/python/framework/type_spec.py", line 579, in type_spec_from_value
raise TypeError("Could not build a TypeSpec for %r with type %s" %
TypeError: Could not build a TypeSpec for <KerasTensor: shape=(None, 128) dtype=float32 (created by layer 'tf.__operators__.add')> with type KerasTensor
It seems the lambda layer is causing some problems, any tf/keras expert to help on this?
Thanks!