How do I include constant_data in my model?

Questions v5
,
1

Hello,

I’m trying to work with the constant data group of the inference object, only, the model does not seem to save mine.

Following this part of the doc, pymc.Data — PyMC dev documentation, I instantiated and fit the following model:

with pm.Model(coords=coords, rng_seeder=RANDOM_SEED) as pooled_model:
    item_idx = pm.Data('item_idx',items, dims="obs_id", mutable=False)
    a = pm.Normal("a", 0.0, sigma=10.0, dims="ITEM_NUMBER")

    theta = a[item_idx]
    sigma = pm.HalfCauchy("error", 0.5)

    y = pm.Normal("y", theta, sigma=sigma, observed=training_data['eaches'], dims="obs_id")

But there is no constant_data in the inference object. After sampling, I get the following…
image

What am I doing wrong to miss this?

2

What version of pymc are you using? When I run this (a simplified version of your code):

import pymc as pm
coords = {
    "obs_id": [0,1,2,3,4],
}
with pm.Model(coords=coords) as rugby_model:
    item_idx = pm.Data('item_idx',[0,1,2,3,4], dims="obs_id", mutable=False)
    a = pm.Normal("a", 0.0, sigma=10.0, shape=5)

    theta = a[item_idx]
    sigma = pm.HalfCauchy("error", 0.5)

    y = pm.Normal("y", theta, sigma=sigma, observed=[3,2,6,8,4])

    idata= pm.sample()

I get this:

In [6]: idata
Out[6]: 
Inference data with groups:
	> posterior
	> log_likelihood
	> sample_stats
	> observed_data
	> constant_data

The constant data is this:

In [7]: idata.constant_data
Out[7]: 
<xarray.Dataset>
Dimensions:   (obs_id: 5)
Coordinates:
  * obs_id    (obs_id) int64 0 1 2 3 4
Data variables:
    item_idx  (obs_id) int32 0 1 2 3 4
Attributes:
    created_at:                 2022-05-17T21:29:44.899657
    arviz_version:              0.11.4
    inference_library:          pymc
    inference_library_version:  4.0.0b6
3

I’m using pymc=‘4.0.0b6’. When I sample, i’m using:

pooled_trace = pymc.sampling_jax.sample_numpyro_nuts(tune=1000, chains = 4, target_accept=0.9)

Could using jax that have anything to do with it?

1 Like
4

Ah, you’re using jax. Does constant data appear if you use pm.sample()? If so, it may be similar to (but different from) this issue.

5

pm.sample doesn’t even work. Chain 1 always fails. It might be the sizes of the data but I’m not sure. The value error states the matrix contains zeros on the diagonal.

6

Odd. Can you run my toy code and see what happens?

7

It runs fine and has constant_data…must have something to do with jax.

1 Like
8

If you want to open an issue about this, please do. If not, I can do it and link to this thread.

9

I think I did this correctly. If you want to check, here it is… JAX/NUMPYRO does not include constant_data · Issue #5781 · pymc-devs/pymc (github.com)

Thank you for the sample code.

2 Likes
10

Thanks!

11

And I would suggest opening a new thread about the chain 1 failures using pm.sample() (if you’re interested in resolving that).