rasa: agent.handle_message() returns empty text after a few conversations

Rasa Core version: 0.9.0a2 Rasa NLU version: 0.12.2 Python version: 3.6 Tensorflow version: tensorflow-gpu 1.7.0 Operating system (windows, osx, …): Ubuntu 17.10 Issue: agent.handle_message() returns only empty dict/text after a few conversations. See example below:

George: hi Bot: hey there None! George: my name is eric Bot: Good to know you, None George: i am called eric Bot: Good to know you, None George: my parents call me eric Bot: Good to know you, eric George: bye Bot: KEINE ANTWORT ## means dict/text is empty, i.e. None George: that’s correct Bot: KEINE ANTWORT ## means dict/text is empty, i.e. None George: hi
Bot: KEINE ANTWORT ## means dict/text is empty, i.e. None George: my parents call me eric Bot: KEINE ANTWORT ## means dict/text is empty, i.e. None George:

Content of domain file (if used & relevant): see attached zip files

Thx in advance for your help!
KR Eric
[test_rasa.zip](https://github.com/RasaHQ/rasa_core/files/1939666/test_rasa.zip)

About this issue

Original URL
State: closed
Created 6 years ago
Comments: 20 (8 by maintainers)

Most upvoted comments

@tmbo: thx pip install -U git+https://github.com/rasahq/rasa_core.git --user worked (i had to use --user in addition)

@tmbo: the max_history and featurizer issue wasn’t necessary until 0.9.0a3, only from 0.9.0a5 onwards, plus to complement the limited documentation for all the poor folks out there in the world:

training the dialogue BEFORE 0.9.0a5:

agent = Agent(domain_file, policies=[MemoizationPolicy(), KerasPolicy(), fallback])
agent.train(training_data_file,
            max_history=5,
            epochs=500,
            batch_size=10,
            validation_split=0.2)

training the dialogue FROM 0.9.0a5 ONWARDS:

agent = Agent(domain_file, policies=[MemoizationPolicy(max_history=5), KerasPolicy(), fallback])
training_data=agent.load_data(training_data_file)
agent.train(training_data,
            epochs=500,
            batch_size=10,
            validation_split=0.2)

ewagner70 on May 15, 2018

I’m having same problem. I don’t have any slots in my stories. I’m using sample stories and domain from the documentation. My NLU version is 0.12.3. Other than that, all versions are same.

saqib-ahmed on May 11, 2018