We need a set of principled tests for a variety of graph model and algorithm properties. This issue is to be one of many capturing specific testing use cases.
This test must provide a quantitative measure of the sensitivity of cred scores to the inclusion or exclusion of a particular edge time. Since edge types have weight parameters, this actually best defined as sensitivity of the edge weight parameters (those exposed in the existing prototype).
The utilities and scripts for such a computational experiment should be part of the "research infrastructure" because this experiment will need repeating when other aspects of the data model or weighting heuristics have proposed changes.
We need a set of principled tests for a variety of graph model and algorithm properties. This issue is to be one of many capturing specific testing use cases.
This test must provide a quantitative measure of the sensitivity of cred scores to the inclusion or exclusion of a particular edge time. Since edge types have weight parameters, this actually best defined as sensitivity of the edge weight parameters (those exposed in the existing prototype).
The utilities and scripts for such a computational experiment should be part of the "research infrastructure" because this experiment will need repeating when other aspects of the data model or weighting heuristics have proposed changes.
see: https://github.com/sourcecred/sourcecred/issues/1116#issuecomment-471644958