Here's the corresponding blog post.
To train a model from scratch, run
I recommend a value of N = 50 or 100 for best results, though training does take some time with those values.
python a2c.py --n 100
Better graphs, Tensorboard visualizations, testing, and saved model files on the way.
|N||Max Reward||Iterations before overfit|
|50||397||Less than 1020626|
|100||428||Less than 1031646|