dependencis

[Kamulegeya-Fahad]

Thanks for the tool! When running the tutorial, I found that the current installation instructions lead to a compatibility issue between packages.

This issue is highlighted here: https://stackoverflow.com/questions/67549661/importerror-cannot-import-name-layernormalization-from-tensorflow-python-ker

I made a quick fix by changing the environment.yaml to have python=3.7, but it would be great to have an updated environment that works with the latest code. Any chance this could be provided? Thanks!

[davek44]

Hi, I'm not sure that I understand. Are you having a problem installing the dependencies? Which package is causing the problem? I don't understand what this might have to do with layer normalization.

[Kamulegeya-Fahad]

It seems Keras LayerNormalization is incompatible with Python 3.8 in the created basenji environment, after downgrading to python 3.7 the environment works fine.

[davek44]

I'm still a little confused; LayerNormalization is a Tensorflow class, not an import. This sounds like the type of thing that might be bespoke to your setup and/or will work itself out soon. If you have a fix that's working for you, I'm going to leave things as is unless others report similar problems.

[gfudenberg]

Do basenji tutorials currently run for you with the recommended conda environment install? I'm getting similar errors that are also fixed by modifying the python to be =3.7 environment.yml.

On Mon, Oct 25, 2021, 11:47 AM David Kelley @.***> wrote:

I'm still a little confused; LayerNormalization is a Tensorflow class, not an import. This sounds like the type of thing that might be bespoke to your setup and/or will work itself out soon. If you have a fix that's working for you, I'm going to leave things as is unless others report similar problems.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/calico/basenji/issues/97#issuecomment-951206739, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEV7GZIYEJB6OMJAYNJ5DS3UIWQ2PANCNFSM5GMNPERA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[davek44]

I haven't reinstalled in awhile, so not sure. Which tutorial in particular is giving errors? I didn't think any of them use LayerNormalization.

[QuantumAlmonds]

basenji_train.py in train_test.ipynb in the tutorials directory does not seem to work with the python version in the prespecified environment.

[davek44]

Hi Dan, we're in the process of cleaning up the installation to try to avoid problems like this in the future. In the meantime, can you elaborate on the error that you're seeing?

[QuantumAlmonds]

When fit_tape() is called during model training, there is a dimension mismatch where an expected dimension of (,3) is misaligned with a passed dimension of (). This was done with the basenji_train.py module via the tutorial.

[davek44]

Hi, I tracked down the problem. I revised the way DNA is stored in the TFRecords, but forgot to update this. Delete your data/heart_l131k and data/heart_l131k.tgz and redownload from the cloud bucket (or create using the preprocess tutorial).