Test Running Error about neural-tangents HOT 5 CLOSED

Thanks again for reporting this silent bug! It's now fixed with 11d70e1 (numpy 1.19 compatible), and 148ba2b (makes the for shell loop exit with code 1 if it fails mid-loop; here's a sample of a failing test https://travis-ci.org/github/google/neural-tangents/builds/705595567/config and passing master: https://travis-ci.org/github/google/neural-tangents/builds/705590810).

from neural-tangents.

Thanks for reporting this - looks like JAX or our code specifically doesn't go along with numpy 1.19, could you try with

pip install numpy=1.18.5

?

Interestingly, our tests have been also failing but reported as passing... https://travis-ci.org/github/google/neural-tangents/jobs/704287384

from neural-tangents.

Thanks for the reply, Roman! If you look into the execution log of travis CI, those tests actually failed.

from neural-tangents.

But as you mentioned, Numpy version is indeed the problem. Take Monte Carlo tests for example, all of them failed with Numpy 1.19, but they work fine with 1.18.

from neural-tangents.

Interestingly, our tests have been also failing but reported as passing... https://travis-ci.org/github/google/neural-tangents/jobs/704287384

Wrong exiting code (0) which misleads Travis CI might be the problem? https://stackoverflow.com/questions/24972098/unit-test-script-returns-exit-code-0-even-if-tests-fail

from neural-tangents.