Hi, so I’m using memoization policy but after the first action, it fails to find the next action. As suggested in a previous post, I have set augmentation factor to 0, but to no avail.
When testing my bot, I literally took out the flattened version of stories.md that was dumped, and tried to follow it step by step. Still, the bot cannot find the action though I am reading from the generated stories (after checkpoint gluing).
Augmentation factor will glue stories to create larger ones. setting it to 0 makes sure you only have stories that you have written in your training data. Did you try to adjust the max_history with the --history flag
Personally, i don’t like checkpoints as it doesn’t fully represent the training data. I use interactive learning to generate my stories full length stories.