seldonian.models.objectives.US_estimate

US_estimate(model, theta, episodes, **kwargs)

Get the expected return of the PRIMARY reward for behavior episodes whose actions (cr,cf) fall within the theta bounding box. This function is the objective used in this example: https://seldonian.cs.umass.edu/Tutorials/examples/diabetes/ It is hardcoded for that specific application and in its current form should not be used for other problems. See https://arxiv.org/abs/1611.03451 for the general form.

Parameters:
  • model – SeldonianModel instance

  • theta (numpy ndarray) – The parameter weights

  • episodes – List of episodes

Returns:

The US estimate

Return type:

float