Girth Release Update

The refactoring of the dependent package girth required several updates to the import statements. A hard requirement on girth > 0.8.0 is required for girth_mcmc > 0.5.1.

Added Partial Credit Model

Added partial credit model to IRT models. This supports both univariate and multivariate estimation.

Univariate Signature

girth_model = GirthMCMC(model='PCM', model_args=(n_categories,))

Multivariate Signature

girth_model = GirthMCMC(model='PCM_MD', model_args=(n_categories, n_factors))

Multidimensional GRM IRT

Added multidimensional graded response model as an estimation option.

Run option looks like:

girth_model = GirthMCMC(model='GRM_MD', model_args=(n_categories, n_factors),
                        options={'n_tune': 1000, 'n_samples': 1000})

Multidimensional 2PL IRT Seeding

Added an initial guess parameter for the multidimensional 2pl model to help with convergence.

Added as a girth_mcmc option:

options = {'initial_guess': True}

Multidimensional 2PL IRT

Added multidimensional 2PL IRT for MCMC and Variational samplers.

August Release

• Added missing data support

December Release

Alpha Version: Support For MCMC / Variational Inference

Unidimensional

• Rasch Model
• 1PL Model
• 2PL Model
• 3PL Model
• Graded Response Model