Story Blocks never process before Interactive Training

Rasa Open Source
1

I’ve been trying to use interactive learning recently and have found that when trying to process the story blocks that it gets stuck as the trackers keep on increasing, and I am never able to actually use the interactive learning.

Processed trackers: 100%|██████| 1039/1039 [00:03<00:00, 335.92it/s, # actions=375] 2019-08-20 20:12:52 INFO rasa.core.agent - Persisted model to ‘C:\Users\SHAYVA~1\AppData\Local\Temp\tmp07n7t83l\core’ Core model training completed. NLU data/configuration did not change. No need to retrain NLU model. Your Rasa model is trained and saved at ‘C:\Users\ShayVanDam\Projects\RASA\chtbt\models\20190820-201252.tar.gz’. Bot loaded. Visualisation at http://localhost:5006/visualization.html. Type a message and press enter (press ‘Ctr-c’ to exit). Processed Story Blocks: 100%|█████| 155/155 [00:01<00:00, 100.23it/s, # trackers=1] Processed Story Blocks: 100%|█████| 155/155 [01:21<00:00, 1.90it/s, # trackers=26] Processed Story Blocks: 77%|▊| 120/155 [01:49<14:38, 25.11s/it, # trackers=318452]

Is there any reason this may happen, or any way I can avoid this? I recently added a lot of stories and nlu data recently too. Is it possible the stories are stuck in some sort of loop due to checkpoints?

Let me know if anymore info is needed.

2

What version of rasa do you use? How many checkpoints do you use? What do you mean by stuck?

3

I’m using the latest version of Rasa. It was getting stuck on the processing story blocks as it would stay at 96% and the trackers would keep on increasing to an unmanageable number. I believe it was my stories where I would have something like:

Story

checkpoint_1

  • intent_1
    • utter_1
  • intent_2
    • utter_2

checkpoint_1

Strangely this didn’t affect the training of the model but only happened when trying to do interactive learning.

4

number of trackers can blow up depending on your use of checkpoints, but normal training and interactive training should be the same

1 Like
5

and if you blow it up despite of checkpoints, what can you do that the training continues and gets completed?