Fix LSTM benchmark to evaluate on test set

[YounesBouhadjar]

Fixes #262
The benchmark is corrected to evaluate LSTM on the test set, using the hidden states warmed up at the end of training.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 77.63%. Comparing base (aefdbaf) to head (a2a7026).
Report is 5 commits behind head on dev.

Additional details and impacted files

@@           Coverage Diff           @@
##              dev     #263   +/-   ##
=======================================
  Coverage   77.63%   77.63%           
=======================================
  Files          43       43           
  Lines         805      805           
  Branches      119      119           
=======================================
  Hits          625      625           
  Misses        133      133           
  Partials       47       47           

Flag	Coverage Δ
unittests	77.63% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jasonlyik]

Thanks @YounesBouhadjar, I think there was just one small issue -

When the hidden states are populated in the last epoch of the training at L139, there is still an optimizer step / weight update after that.

So I believe that the hidden states initialized for testing would be using the weights after the second-to-last update, rather than the weights after the final update.

I changed it to re-run the forward pass on the train data before the benchmark run on the test set, and also re-did the benchmark, but the difference in results is negligible.

[jasonlyik]

Update corrects the issue with the test dataset for the LSTM Mackey-Glass benchmark.