Fix metric test failures #414
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The TF 2.7.1 upgrade seems to have broken the behaviour of
tf.random.randomUniform
and friends. This causes several metric tests which used random inputs to fail due to incorrect shapes being passed in. I've replaced them all withtf.random.statelessRandomUniform
which works correctly.I'm not sure what the real root cause is here, as the TF 2.7.1 release notes don't mention any RNG changes. TF 2.8 is moving away from those RNG ops though, so maybe a backported security fix was a little too broad.
This PR also actually bumps to protobuf 3.19.4 (which is identical to 3.19.2 in the Java side of things, but is the latest version), and removes
-XX:MaxPermSize=256m
from the test command line arguments. The perm gen was removed in Java 8, the option is removed in Java 17 and so supplying it causes the test JVM to error out on startup.