tensorflow / tensorflow/probability

gradient descent and performance issues with Tf 2.2

Open
#948 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Jupyter Notebook
Stars
4.4k
Forks
1.1k
PR merge metrics
No merged PRs in 30d

Description

I am running an algorithm with tensorflow 2.2 and tensorflow-probability 0.10 using a gaussian process layer which is based on the example provided in google colab.

I am getting the following warnings which seem to be impacting my performance further. Before the upgrade I could get near 0 loss with 270 epochs. Now it has stopped improving altogether after a few hundred epochs and gets permanently stuck at 34. I believe that the problem may have something to do with this warning
WARNING:tensorflow:Gradients do not exist for variables ['variational_gaussian_process_6/observation_noise_variance:0'] when minimizing the loss.

I suspect that while the VGP distribution was made tape safe the VGP layer still doesnt appear to be tape safe. I am wondering if this is true and if so how I can get around this problem

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the Google Colab Gaussian process example and reproduce it under TensorFlow 2.2 and tensorflow-probability 0.10. Inspect the VGP layer around the reported observation_noise_variance gradient warning and compare its behavior with the VGP distribution. Done means identifying whether the warning causes the stalled loss and documenting a verified workaround or fix.

Written by the indexing model from the issue text.

Assessment

Tech stack
jupyter-notebook
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.