Learning mode

[ghost]

Hi,

Hope you are all well !

I was wondering if it is possible to add a learning mode to gpt2bot too ?

I found this example but the code is not really as neat as yours ^^.

• https://github.com/Existencce/GPT2-Telegram-Chatbot#for-quick-command-reference

Thanks for any inputs or insights on that that question.

Cheers, X

[polakowo]

I can't fully understand what learning mode is, the code is a little bit of a mess. The model in this example doesn't learn anything new, it simply adds some context prior to the dialogue.

[ghost]

That's what I though while testing it... Thanks for the confirmation :-)

I guess that a learning mode could be added with a separate skill by using something like https://deeppavlov-agent.readthedocs.io/ with https://github.com/deepset-ai/haystack for the QA system.

Have you ever tested those stacks ?

I would be curious to see what kind of results it gives, as I tested haystack with fairly good results so far, with gpt2bot+deeppavlov agent+haystack

Another (sorry if a dumb one!!! ^^) question, is it possible to apply reinforcement learning to a chatbot ?

[polakowo]

Can't comment on these tools as I haven't used them in this setting.

You may find papers training an RL agent for conversation. You would need to establish a reward function such as human likeness, answer coherence, etc. Can't tell whether it's effective though.

[breadbrowser]

https://machinelearning.veravandeseyp.com/

[breadbrowser]

https://parl.ai/