Dev/record #59

[long715]

Implementation of Record class in _example_trainingloops.py

• replace and record all logs from training and evaluation on one csv file (using Record)
• plots dynamically in training (and selected evaluations)
• saves avg plots for all attributes in csv file

Changes made to Plot.py:

• commented out line 75 because its causing "AttributeError: no attribute save.csv"

To work on:

• [x] storing models
• [x] intermediate saves
• [ ] separating evaluation and training data sets?

[long715]

Everything is looking good! Added my own commits to fix a bug in the Record class. Currently coming across a bug that creates multiple blank windows – specifically for the TD3 or policy example. Are you experiencing the same?

Yep, Im getting this bug as well. Seems to happen after 1200 steps, next episode log after the logs of actor/critic losses. I found a quick fix by commenting out the plot.post() command at line 118 for policy example. Tested it with another max_steps_exploration value and same thing happens where the figures appear after the actor/critic writes. Not sure why this is happening, the plot instance data should be independent of whats in the record file :'

[retinfai]

True. It's probably something to do with matplotlib being trash. I reckon we remove that for now, and bundle this problem into the new issue #63 and we can merge the current integration you've done. Thoughts?

[long715]

True. It's probably something to do with matplotlib being trash. I reckon we remove that for now, and bundle this problem into the new issue #63 and we can merge the current integration you've done. Thoughts?

Sounds good! I'll quickly fix it up now 🙏