PrincetonUniversity / PsyNeuLink / 34630995006

84%

master: 84%

Build Type

Pull #3642

github

Committed by aryan-pan

Commit Message

Fit a group with participants their factory scores by estimator

Covers the case the group-level rejection points users towards: a factory that builds
participant models scored by a trained estimator, fitted in one process and across a
cluster, which have to agree.

Neither participant model is given an execution mode to compile with, which is what
makes this a test rather than a demonstration: scoring by estimator compiles nothing, so
a model that fell back to simulating would be refused instead of quietly fitting.

The estimator is trained per session rather than kept as a file. It takes a fraction of a
second to make, and a saved one is weights for the architecture the sbi version in use
builds, which a checked-in copy would silently outlive.

Pull Request Pull Request #3642: Hierarchical PEC Fitting #2: Neural Likelihood Estimation

Coverage Stats

11581 of 15088 branches covered (76.76%)

Branch coverage included in aggregate %.

82 of 315 new or added lines in 4 files covered. (26.03%)

2 existing lines in 1 file now uncovered.

39652 of 45973 relevant lines covered (86.25%)

0.86 hits per line