check autoencoder tutorial about aitom HOT 3 CLOSED

Tutorial works as expected on keras 2.1.* (probably should add this to README if this is only intended to work with older versions of keras)

However with keras 2.2.* it generates the following error
Traceback (most recent call last): File "006_autoencoder.py", line 90, in <module> SEG.decode_all_images(d, data_dir) File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/aitom/classify/deep/unsupervised/autoencoder/seg_src.py", line 61, in decode_all_images autoencoder = KM.load_model(op_join(data_dir, 'model', 'model-autoencoder.h5')) File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/keras/engine/saving.py", line 419, in load_model model = _deserialize_model(f, custom_objects, compile) File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/keras/engine/saving.py", line 258, in _deserialize_model .format(len(layer_names), len(filtered_layers)) ValueError: You are trying to load a weight file containing 7 layers into a model with 0 layers

Possible cause of this error include:

1. the use of keras Input layer
2. the presence of the input_shape parameter in function calls to keras.layers

Please refer to this issue for more details

Another possible workaround is proposed here. This workaround doesn't require changing the code line by line and add the additional parameter, but it seems a bit "hacky".

from aitom.

hi @buptwhr can you check the problem detected by @yg422 ?

from aitom.

@yg422 I've noticed this problem, so I put keras==2.1.0 in requirements.txt. As for the issue you've mentioned, I read it before. But I don't know how to avoid Input or input_shape.
By using autoencoder.load_weights() rather than KM.load_model(), I can solve the problem. But there are so many KM.load_model() and save_model in the project. I am not sure it'a stable choice.
I suggest using keras==2.1.0 to make minimal change to the original code.

from aitom.