Unify quantitative actions interface by shaharbar1 · Pull Request #111 · PlaytikaOSS/pybandits

shaharbar1 · 2025-11-18T07:19:19Z

Changes:

Moved _normalize_field method to PyBanditsBaseModel for default value normalization of optional fields.
Unified quantitative actions interface to use callables for general form, rather than floats for zooming.
Improved tests for action selection and quantitative actions, ensuring proper handling of constraints and expected results.

AnastasiiaKabeshova

Great code organization improvement :)

Just one question for myself: Why was Optuna replaced? (Performance, dependency reduction, or other reasons?)

pybandits/utils.py

AnastasiiaKabeshova · 2026-01-08T14:46:44Z

pybandits/utils.py

+        return result.x
+    else:
+        logger.warning(f"Optimization failed: {result.message}")
+        return None


When optimization fails, maximize_by_quantity returns None
Did you check that all callers handle None properly?
Whether a more specific exception would be better?

Good point, I added an appropriate exception handling.

ronshiff1 · 2026-01-15T12:44:31Z

pybandits/base.py

-QuantitativeProbabilityWeight = Tuple[Tuple[Tuple[Float01, ...], ProbabilityWeight], ...]
-QuantitativeMOProbability = Tuple[Tuple[Tuple[Float01, ...], List[Probability]], ...]
-QuantitativeMOProbabilityWeight = Tuple[Tuple[Tuple[Float01, ...], List[ProbabilityWeight]], ...]
+QuantitativeProbability = Callable[[np.ndarray], Probability]


The name of the type is misleading...
maybe it should be: QuantitativeProbabilityFunction?

@ronshiff1 Maybe ContinuousProbability? I just didn't want it to be too long.

ronshiff1 · 2026-01-15T13:53:27Z

pybandits/quantitative_model.py

@@ -68,6 +78,12 @@ class QuantitativeModel(BaseModelSO, ABC):
    def sample_proba(self, **kwargs) -> List[QuantitativeProbability]:


Please explain why list. For Multi objective?

If I understand correctly - it's for different segments?
I think that it is very specific to the zooming algorithm, and it should be function which receives an input, and outputs one function, and not a list
the zooming model should create this function and handle the segmentation internally (I also commented on the first sample_proba implementation)

The list is not because of zooming. It returns a list whose length is as the number of samples.

I fixed the docstring to make it explicit.

pybandits/quantitative_model.py

ronshiff1 · 2026-01-19T12:09:02Z

pybandits/quantitative_model.py

+                    for segment in segment_probabilities.keys():
+                        if quantity in segment:
+                            segment_probas_for_segment = segment_probabilities[segment]
+                            return segment_probas_for_segment[sample_idx][output_index]  # Probability or weight


Why not directly return segment_probas_for_segment[sample_idx] which is tuple?
(so def create_probability_or_weight_function(
sample_idx: NonNegativeInt
) -> Tuple[QuantitativeProbability, QuantitativeWeight]:

Then in the cmab class you would need to separate them to list of probabilities and list of weights.
So I preferred to do it in advance.

ronshiff1 · 2026-01-26T10:42:30Z

pybandits/base.py

        return annotation

+    @classmethod
+    def _normalize_field(cls, v: Any, field_name: str) -> Any:


why do we need this?

I use it in validators of "pre" behavior to get the default value of an ungiven attribute

pybandits/quantitative_model.py

ronshiff1 · 2026-02-02T09:39:19Z

pybandits/quantitative_model.py

        rewards: Union[List[BinaryReward], List[List[BinaryReward]]]
            The reward for each sample.
        """
        super()._quantitative_update(quantities, rewards)


this:
quantities: Optional[List[Union[float, List[float], None]]],
rewards: Union[List[BinaryReward], List[List[BinaryReward]]],
doesn't conform with the ZoomingModel _quantitative_update:
def _quantitative_update(self, quantities: List[Union[float, np.ndarray]], rewards: List[BinaryReward], **kwargs):

Wow! Nice catch. Fixed to: Optional[List[Union[float, List[float], None]]]

### Changes: * Moved `_normalize_field` method to `PyBanditsBaseModel` for default value normalization of optional fields. * Unified quantitative actions interface to use callables for general form, rather than floats for zooming. * Improved tests for action selection and quantitative actions, ensuring proper handling of constraints and expected results.

ronshiff1 · 2026-02-06T13:16:37Z

pybandits/mab.py

        self,
        p: ActionRewardLikelihood,
        actions: Optional[Dict[ActionId, BaseModel]] = None,
    ) -> ActionId:


say, shouldnt it be
-> Union[ActionId, QuantitativeActionId]?

ronshiff1 · 2026-02-06T16:02:17Z

docs/src/tutorials/cmab_zooming.ipynb

    "\n",
    "    # Observe reward\n",
    "    rewards = [\n",
    "        reward_function(chosen_action, chosen_quantity, current_context[0])\n",


isn't current_context[0] a bug?
it uses the first context for all samples

shaharbar1 added the enhancement New feature or request label Nov 18, 2025

shaharbar1 force-pushed the feature/unified_quantitative_interface branch 5 times, most recently from 4fee41e to c052743 Compare November 18, 2025 11:36

shaharbar1 requested review from AnastasiiaKabeshova, j3rom3c and ronshiff1 November 18, 2025 11:39

shaharbar1 force-pushed the feature/unified_quantitative_interface branch 11 times, most recently from 4669026 to b7e5bec Compare November 25, 2025 12:37

shaharbar1 force-pushed the feature/unified_quantitative_interface branch 2 times, most recently from 1664e90 to 5a43378 Compare December 1, 2025 07:10

shaharbar1 force-pushed the feature/unified_quantitative_interface branch from 5a43378 to a9d8084 Compare December 25, 2025 07:30

AnastasiiaKabeshova requested changes Jan 8, 2026

View reviewed changes

ronshiff1 reviewed Jan 15, 2026

View reviewed changes

shaharbar1 force-pushed the feature/unified_quantitative_interface branch from a9d8084 to 7f6a131 Compare January 19, 2026 08:16

ronshiff1 reviewed Jan 19, 2026

View reviewed changes

ronshiff1 reviewed Jan 26, 2026

View reviewed changes

ronshiff1 reviewed Feb 2, 2026

View reviewed changes

pybandits/quantitative_model.py Show resolved Hide resolved

ronshiff1 reviewed Feb 2, 2026

View reviewed changes

ronshiff1 self-requested a review February 5, 2026 14:31

ronshiff1 approved these changes Feb 5, 2026

View reviewed changes

shaharbar1 force-pushed the feature/unified_quantitative_interface branch from 7f6a131 to 2c00401 Compare February 5, 2026 20:52

ronshiff1 reviewed Feb 6, 2026

View reviewed changes

		@@ -68,6 +78,12 @@ class QuantitativeModel(BaseModelSO, ABC):
		def sample_proba(self, **kwargs) -> List[QuantitativeProbability]:

Conversation

shaharbar1 commented Nov 18, 2025

Changes:

Uh oh!

AnastasiiaKabeshova left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ronshiff1 Jan 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shaharbar1 Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ronshiff1 Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ronshiff1 Feb 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ronshiff1 Jan 15, 2026 •

edited

Loading

shaharbar1 Jan 19, 2026 •

edited

Loading

ronshiff1 Jan 19, 2026 •

edited

Loading

ronshiff1 Feb 6, 2026 •

edited

Loading