tensorflow / tensorflow/recommenders
tfrs.metrics.FactorizedTopK with tensorflow_macos and tensorflow_metal
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 2k
- Forks
- 300
- PR merge metrics
- No merged PRs in 30d
Description
Metrics computed by FactorizedTopK seems very off when running on M1 Max.
Tried it on the "Quickstart" movielens example with Adam(0.005) and received the following evaluation results:
{'factorized_top_k/top_5_categorical_accuracy': 0.007550000213086605,
'factorized_top_k/top_10_categorical_accuracy': 0.018650000914931297,
'factorized_top_k/top_15_categorical_accuracy': 0.03215000033378601,
'factorized_top_k/top_20_categorical_accuracy': 0.2814500033855438,
'factorized_top_k/top_25_categorical_accuracy': 0.31675001978874207,
'factorized_top_k/top_30_categorical_accuracy': 0.7000000476837158,
'factorized_top_k/top_35_categorical_accuracy': 0.6884000301361084,
'factorized_top_k/top_40_categorical_accuracy': 0.6511000394821167,
'loss': 28395.984375,
'regularization_loss': 0,
'total_loss': 28395.984375}
tensorflow-macos==2.10.0
tensorflow-recommenders ==0.7.2
tensorflow-metal==0.6.0
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the Quickstart MovieLens example and reproduce FactorizedTopK evaluation on an M1 Max using tensorflow-macos 2.10.0, tensorflow-recommenders 0.7.2, and tensorflow-metal 0.6.0. Compare the results with a supported non-metal TensorFlow environment and trace the metric computation. Done means the platform-specific discrepancy is explained and a regression test or documented limitation is identified.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, tensorflow
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100