Python: TypeError in lm_spike when initializing sampler

When attempting to instantiate lm_spike, a TypeError occurs.

System info:

OS: Ubuntu 18.04.6 LTS
Python: 3.8.8
- BayesBoom-0.1.9
- pybind11-2.10.1
- numpy-1.19.2
- pandas-1.1.5

Minimal example:

import numpy as np
import pandas as pd
from BayesBoom.spikeslab import dot, lm_spike

data = pd.DataFrame(np.random.random(3000).reshape(1000,3), columns=['y', 'X1', 'X2'])
model = lm_spike(f'y ~ {dot(data)}', niter=100, data=data)

Expectation:

An lm_spike model is instantiated (though not useful for this dummy data).

Actual result:

(...)/site-packages/BayesBoom/spikeslab/spikeslab.py in __init__(self, formula, niter, data, prior, ping, seed, **kwargs)
    161             prior = RegressionSpikeSlabPrior(x=X, y=y, **kwargs)
    162 
--> 163         sampler = boom.BregVsSampler(
    164             self._model,
    165             prior.slab(self._model.Sigsq_prm),

TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. _boom.BregVsSampler(model: _boom.RegressionModel, slab: _boom.MvnGivenScalarSigma, residual_precision_prior: _boom.GammaModelBase, spike: _boom.VariableSelectionPrior, seeding_rng: _boom.RNG = <_boom.RNG object at 0x7f63d5604670>)

Invoked with: <_boom.RegressionModel object at 0x7f639e0bf4b0>, <_boom.MvnGivenScalarSigma object at 0x7f639e0aa930>, <_boom.ChisqModel object at 0x7f639e0b95b0>, <_boom.VariableSelectionPrior object at 0x7f639efb2770>

Aside from looking for an obvious error on my part (step 1), I have also tried simple modifications of lm_spike to isolate the issue. Notably:

Passing residual_precision_prior=None to BregVsSampler
- Succeeds in instantiating sampler, but leads to an error later. (Expected.)
Constructing a GammaModel instance and passing that as residual_precision_prior
- Causes a TypeError very similar to the one from the minimal example (replacing the reference to ChisqModel with one to GammaModel.

The first of these confirms that the issue is specifically caused by the provided GammaModelBase (ChisqModel) instance, while the second demonstrates that it is not a unique issue affecting the chi-squared model but one which also applies to other implementations of GammaModelBase.

steve-the-bayesian / BOOM

Python: TypeError in lm_spike when initializing sampler #68