Setting priors raises ValueError: need at least one array to concatenate

Hi! I was trying to follow the docs to set custom priors on some made up data but kept hitting this numpy error in weird places. Including an example, diagnostics, and the results of my investigation here:

Example:

import bambi as bmb
import pandas as pd
import numpy as np

df = pd.DataFrame(
    dict(y=[0, 1, 0, 0, 1, 0], x_1=np.random.uniform(size=(6,)), x_2=np.random.uniform(size=(6,))),
)


priors = {
    "x_1": bmb.Prior("TruncatedNormal", sigma=1, mu=0, lower=0),
    "x_2": bmb.Prior("TruncatedNormal", sigma=1, mu=0, upper=0),
}
model = bmb.Model(
    formula="y ~ x_1 + x_2", data=df, family="bernoulli", priors=priors
)

model.build()

...
~/.pyenv/versions/3.8.10/envs/jupyter/lib/python3.8/site-packages/numpy/core/shape_base.py in hstack(tup)
    343         return _nx.concatenate(arrs, 0)
    344     else:
--> 345         return _nx.concatenate(arrs, 1)
    346 
    347 

<__array_function__ internals> in concatenate(*args, **kwargs)

ValueError: need at least one array to concatenate

Diagnostics:

import sys

print("sys: ", sys.version)
print("np: ", np.__version__)
print("pd: ", pd.__version__)
print("bmb: ", bmb.__version__)

sys:  3.8.10 (default, Jun 24 2021, 16:50:28) 
[Clang 12.0.5 (clang-1205.0.22.11)]
np:  1.21.2
pd:  1.3.3
bmb:  0.6.3

What i tried

1. no results in project tracker
2. Added/removed/changed the automatic_priors setting.
3. Added/removed/changed the auto_scale setting.
4. I put one term’s prior into a tuple, and then the model built, but it does not look like the prior took:

priors = {
    ("x_1",): bmb.Prior("TruncatedNormal", sigma=1, mu=0, lower=0),
    "x_2": bmb.Prior("TruncatedNormal", sigma=1, mu=0, upper=0),
}
model = bmb.Model(
    formula="y ~ x_1 + x_2", data=df, family="bernoulli", priors=priors
)

model.build()
model.plot_priors()

You can see that x_1 is not given the correct prior whereas x_2 is.

1. Setting auto_scale=False on the model allows everything to build, but the priors are not correct:

model = bmb.Model(
    formula="y ~ x_1 + x_2", data=df, family="bernoulli", priors=priors, auto_scale=False
)
model.build()
model.plot_priors()

This feels like a numpy incompatibility- are these any known such issues?

[tomicapretto]

This is fixed in #430

[tomicapretto]

Glad it helped! And thanks for raising this issue. I wasn’t aware of it. I’ll try to fix it for the next release 😃