New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 701238 link

Starred by 1 user
Status: Fixed
Owner:
█ kateso...@chromium.org
Last visit > 30 days ago
Closed: May 2017
Cc:
█ wrengr@chromium.org
st...@chromium.org
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 1
Type: Bug



Sign in to add a comment

[Predator] Flaky test of loglinear model training.

Project Member Reported by kateso...@chromium.org, Mar 14 2017 
[Predator] Flaky test of loglinear model training.

https://build.chromium.org/p/chromium.infra/builders/infra-continuous-precise-64/builds/6132

This flaky test is because when we multiply meta weight with meta feature, we assert the lengths are the same. That is causing problems because we drop zero weights in meta weight to have dense representation of sparse array, so when the trained weights have zeros, assertion exceptions will raise.

Comment 1 by kateso...@chromium.org, Mar 14 2017 

the flaky test is ``testTrainWeights`` in training_test.py. It is because
the training data set of loglinear model is randomly generated.

In
https://cs.chromium.org/chromium/infra/appengine/findit/crash/loglinear/test/...,
line 55.

If the trained weight happened to be 0 (the training set is randomly generated,
so the training results vary every time), the length of meta weight would be 0,
however the length of meta feature is still 1, so the assertion will fail.

Comment 3 by kateso...@chromium.org, May 15 2017

Status: Fixed (was: Assigned)
►

Sign in to add a comment