search

AnacondaRecipes / tensorflow_recipes

Tensorflow conda recipes
27 stars 26 forks source link

Upgrade `keras` and `tensorboard` to 2.9.0 #41

Closed sumit0190 closed 2 years ago

sumit0190 commented 2 years ago

Upgrading tensorflow to 2.9.1 requires version 2.9.0 for keras, tensorflow-estimator and tensorboard, which this PR tries to address.

Concourse for keras: https://concourse.build.corp.continuum.io/teams/main/pipelines/keras (NOTE: This fails on certain platforms because of the issue described here).

Concourse for tensorboard: https://concourse.build.corp.continuum.io/teams/main/pipelines/tensorboard

Concourse for tensorflow-estimator: https://concourse.build.corp.continuum.io/teams/main/pipelines/tensorflow-estimator

keras and tensorflow-estimator need tensorflow as a run_constrained dependency since tensorflow doesn't have a build with that version number (because it requires keras...a cyclic dependency of sorts).

(NOTE: Keeping the base branch master-2.8.2 for now, as it is the closest to where we want to be when we are ready to upgrade to 2.9.1.)

tobijk commented 2 years ago

We will also need to update tensorflow-estimator.

sumit0190 commented 2 years ago

@tobijk Added an update for tensorflow-estimator. I had to remove its test section because I wasn't sure what we are supposed to test, and a "real" test would require importing tf and verifying that the estimator methods exist. We could add a pip check like I did with keras.

tobijk commented 2 years ago

I'm still not so happy about the run_constrained part. Putting it there would mean that installing TF becomes optional, but AFAIK keras does not work without TF?

EDIT: Never mind, one can use keras without TF :)