Closed engintoklu closed 1 year ago
Merging #74 (8078df0) into master (e8060ff) will decrease coverage by
0.03%
. The diff coverage is62.50%
.
@@ Coverage Diff @@
## master #74 +/- ##
==========================================
- Coverage 77.85% 77.83% -0.03%
==========================================
Files 49 49
Lines 7330 7332 +2
==========================================
Hits 5707 5707
- Misses 1623 1625 +2
Impacted Files | Coverage Δ | |
---|---|---|
src/evotorch/neuroevolution/supervisedne.py | 87.69% <62.50%> (-2.79%) |
:arrow_down: |
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.
This pull request fixes/improves
SupervisedNE
in two ways:1) The
get_minibatch()
method ofSupervisedNE
no longer uses recursion to handle the end of the data loader's minibatches. The new implementation catches onlyStopIteration
and non-recursively restarts the iterator of the data loader. Any other type of error is now deliberately unhandled to avoid unwanted infinite recursion and to allow the user to see the details of the error.2) The example notebook
Training_MNIST30K.ipynb
demonstrating the usage ofSupervisedNE
is updated so that its hyperparameter configuration follows what is reported in the technical report of EvoTorch. This way, results reported in the report can be reproduced.