h5py version (integrity failure)

[JouyonP]

I followed the installation instructions, but kept getting the following message.

[DEBUG] Loading application... integrity failure

There was no problem with my CUDA settings, and tensorflow was picking up my GPU just fine, so I tried taking stuff out of try/exceptions and found out that keras was failing on loading the models.

original_keras_version = f.attrs['keras_version'].decode('utf8')

AttributeError: 'str' object has no attribute 'decode'

Googled it, and found that it works at h5py version 2.10.0.
I think it would be nice if h5py versions were also included in conda settings.

Thank you anyways for your awesome work.
I'll try it out now since my bug is fixed.

[cemicel]

@JouyonP thanks for pointing it out!
@prasunroy
I would suggest do not do like this:

    NET_C.load_weights('models/colornet_weights.h5')
except:
    RUN_FLAG = False

Print at least something, because integrity failure message is not helpful at all.

    NET_C.load_weights('models/colornet_weights.h5')
except:
    import traceback
    traceback.print_exc()
    # or a simple print like could load model would be really helpful

    RUN_FLAG = False

[fronos]

Just update tensorflow via command
pip install -U tensoflow-gpu