Policy Predictions for End-to-End #7136
Conversation
wochinge
force-pushed
the
end-to-end-policy-predictions
branch
2 times, most recently
from
955ad22 to
36e73a7
Compare
| from unittest.mock import Mock, patch | ||
|
|
||
| import numpy as np | ||
| import pytest | ||
| from _pytest.monkeypatch import MonkeyPatch |
There was a problem hiding this comment.
sorry, typing got a bit out of hand here 🙈
wochinge
force-pushed
the
end-to-end-policy-predictions
branch
from
36e73a7 to
4f5c697
Compare
wochinge
force-pushed
the
end-to-end-policy-predictions
branch
from
db327b1 to
47dacd6
Compare
|
I'll have an engineer review this as soon as you're ok with the changes in general @Ghostvv |
|
part of #6670 |
wochinge
force-pushed
the
end-to-end-policy-predictions
branch
4 times, most recently
from
20e64bd to
da6c99d
Compare
wochinge
requested review from
a team,
kalkbrennerei,
degiz and
Ghostvv
and removed request for
a team and
kalkbrennerei
|
@degiz Do you have time? It looks bigger than it is. Mostly typing and renames. |
|
@wochinge can review that tomorrow in the morning, would it be fine? |
|
Sure, absolutely no rush! |
| @staticmethod | ||
| def is_not_memo_policy( | ||
| policy_name: Text, max_confidence: Optional[float] = None | ||
| policy_name: Optional[Text], max_confidence: Optional[float] = None |
There was a problem hiding this comment.
Should the default value of max_confidence be 0.0?
There was a problem hiding this comment.
No as this would otherwise evaluate to True for the call here
rasa/core/policies/ensemble.py
Outdated
| ) -> bool: | ||
| """Checks if the policy is the `MemoizationPolicy` or one of its subclasses. |
There was a problem hiding this comment.
That's not exactly what it's doing right? There's also a check of max_confidence argument.
There was a problem hiding this comment.
re-named the function and changed the docstring. Better?
Co-authored-by: Alexander Khizov <[email protected]>
Proposed changes:
Policys now return aPolicyPredictionobject instead of the plain probabilities. This allows us to return multiple pieces after a prediction. This is not only useful for this PR but also for this PRPolicyEnsemblenow returns aPolicyPredictionobject instead ofTuple[probabilities, policy_name]. The reason is the same for the change in thePolicyinterface: This allows us to encapsulate more information and pass this up to theMessageProcessortest_policiesmodulepredict_next_with_trackerto avoid having to reimplement some predict logic withinserverfollowup_actionStatus (please check what you already did):
black(please check Readme for instructions)