keras / tensorflow version compatibility

[rekham1077]

Hello,

I am using deepcpg-1.0.7. I am using it through a conda environment with python 3.5, keras 2.0.2, and tensorflow 1.0.1. I have tried later versions of all three as well and I just keep getting error after error running the dcpg_train.py.

I'm attaching the current error with the listed versions of dependencies above.

[rekham1077]

Update: I got the dcpg_train.py script to work! I ended up using a conda environment with python 3.6.3, deepcpg 1.0.7, keras 2.2.0, tensorflow 1.10.0. Other dependencies were installed when I added deepcpg.

I had tried using later versions of keras and tf but I get different errors with each version, and the same thing for trying lower versions. The version compatibility issue also seems to be mostly an issue with the dcpg_train.py step but not the other scripts as far as I know.

[Toseph]

Chiming in on this thread to mention I had encountered the same issue as @rekham1077 while using the latest python installation (v3.8.3) and tensorflow (v2.2.0). The scripts also assume by default anaconda is being used (in my installation it isn't) so we've had to update the shebang so it is available for virtualenv use by users as well.

Changed:
#!/anaconda3/envs/deepcpg/bin/python
To:
#!/usr/bin/python env