poisson race simulator

[hayden-johnson]

This PR provides an initial implementation of the Poisson race model described in Appendix A of:

A comparison of two response time models applied to perceptual matching

Key Differences from the Original Formulation

Stimulus-dependent rate parameters are not implemented. This would require trial-by-trial stimulus identifiers for log-probability evaluation.

See PR 840 in HSSM.

[AlexanderFengler]

Thanks @hayden-johnson left few smaller comments but overall looks good already.

[AlexanderFengler]

@hayden-johnson I think you mentioned some small tutorial, was it this PR, maybe just forgot to git add the file? Didn't see any .ipynb floating around when reviewing (if I missed it just let me know).

[hayden-johnson]

@hayden-johnson I think you mentioned some small tutorial, was it this PR, maybe just forgot to git add the file? Didn't see any .ipynb floating around when reviewing (if I missed it just let me know).

Ah so I added it to PR 840 in this commit, so I could show the pipeline from simulation to parameter recovery.

Should I add a notebook here as well?

[AlexanderFengler]

Hey @hayden-johnson,

thanks this is basically good, just a few minor adjustments to accommodate the drift in the underlying repo code.

Relevant changes that happened in the design of ssm-simulators in the meanwhile.

1. theta_processor.py is now deprecated so we only need to care about the modular version
2. modular_theta_processor.py is renamed to modular_parameter_simulator_adapter.py.
3. The model configs now explicitly carry the model wise parameter transforms that are applied: The pattern is that configs now have a "parameter_transforms" dictionary that has "sampling" (for parameter sampling stage, most models don't have any) and "simulation" (for the adaptation of parameters from passing into the python simulator to passing correctly down to the c-level simulator. Here's an example:

def get_race_no_z_2_config():
    """Get configuration for race model with 2 choices and no z."""
    return {
        "name": "race_no_z_2",
        "params": ["v0", "v1", "a", "t"],
        "param_bounds": [
            [0.0, 0.0, 1.0, 0.0],
            [2.5, 2.5, 3.0, 2.0],
        ],
        "boundary_name": "constant",
        "boundary": bf.constant,
        "n_params": 4,
        "default_params": [0.0, 0.0, 2.0, 1e-3],
        "nchoices": 2,
        "choices": [0, 1],
        "n_particles": 2,
        "simulator": cssm.race_model,
        "parameter_transforms": {
            "sampling": [],
            "simulation": [
                SetZeroArray("z", shape=(None, 2)),
                ColumnStackParameters(["v0", "v1"], "v", delete_sources=False),
                ExpandDimension(["t", "a"]),
            ],
        },
    }

Lastly, if you can, please take the current model out of the race.py file and add a poisson_race.py file instead. I think it's sufficiently different to warrant it's own file even though it's called "race" as well.

Let me know if you have other questions, either way I think it's going to take you only ~10min to work through the conflicts and make the adjustments. This is 99% there.

[AlexanderFengler]

This looks good now afaic, thanks @hayden-johnson.
Just run pre-commit locally then should be good to go.
ruff format . seems to be failing on git actions.