Skip to content

Ratio Score, score type cannot handle legitimate cases where observation and prediction are of different sign (+,-). #172

Description

@russelljjarvis

Background:

Ratio Score score type cannot handle legitimate cases where observation and prediction are of different sign (+,-).

In optimization, when fitting with sweep traces, RatioScore is a more appropriate score choice than ZScore since number of observations is n=1.

@rgerkin

if observation and prediction are of different sign, add the absolute value of the negative one to the positive one.

Pseudo code attempted solution:

observation = -1
prediction = 1
new_observation = np.abs(observation) + prediction
new_observation = 2
intended_score = 1/2 

observation = 1
prediction = -1
new_observation = observation + np.abs(prediction)
new_observation = 2
intended_score = 1/2 actual score 2.


* this approach lead to poor optimization, I think because,
I didn't force score to be 1/2 instead of 2.0 for each of the different cases.

Also in optimization lower scores are better, in this context lower sciunit score with get worse with greater distance from 1.0. I need to make sure that is reflected somehow in a derivative sciunit score I can use with optimization.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions