probcomp/bayeslite

gracefully report error in CGPM for latent categoricals without specified number of categories

Open

#461 opened on Jul 15, 2016

View on GitHub
 (0 comments) (0 reactions) (0 assignees)Python (915 stars) (63 forks)batch import
help wantedk min/hr

Description

CREATE GENERATOR g FOR p USING cgpm (
    LATENT y CATEGORICAL,
    MODEL y GIVEN x USING whatever,
    MODEL z GIVEN y USING frobnosticator
)

In order to use the latent categorical (or, 'nominal') variable y as an input of the frobnosticator model, CGPM wants to know fictitious distribution parameters, namely the number of categories (or, 'names' or 'labels'), but there is no way to ascertain this because x is latent and does not actually appear in the table. The _default_categorical (or, _default_nominal) method will fail trying to select a nonexistent variable from the table -- or maybe quietly return 1 because of the usual sqlite3 quotation misfeature.

Not tested; observed by code inspection.

Alternative fix: Obviate the need to specify fictitious distribution parameters. This requires changes to CGPM.

Contributor guide

Tech stack
python
Domain
backenddata
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
Python
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
45
Research direction
Look at the default categorical method in the CGPM implementation (likely in cgpm.py). The issue occurs when a latent categorical variable is used as an input without specifying the number of categories. The method tries to fetch a nonexistent column from the table. Add a check for latent variables and raise a clear error message. Alternatively, consider modifying the CGPM to infer the number of categories from the model parameters if possible.