Implementation of Diffusion decision model with across-trial variability in boundary separation

PyPI Python License: MIT tests

Diffusion decision model with across-trial variability in boundary separation (sa), drift (sv), and non-decision time (st). This codebase presents the DDM-SA as a fully differentiable PyTensor likelihood for gradient-based Bayesian estimation.

The likelihood is analytic (Navarro–Fuss density; drift variability integrated in closed form, uniform variability by Gauss–Legendre quadrature), so NUTS gets exact gradients. It is validated against the Fortran implementation the model was originally developed in; the Fortran programs, the data, and reference results are in this repository so every validation is reproducible from a clone.

Install

From PyPI:

pip install saddm               # core: numpy, scipy, pytensor
pip install "saddm[sampling]"   # + pymc, arviz, numpyro (pinned jax)
pip install "saddm[hssm]"       # + hssm

From a clone, for development or to run the verification and examples:

pip install -e ".[test]"        # + pytest, numba
pip install -e ".[hssm]"        # everything the examples need

The sampling and hssm extras pin jax==0.5.3 / numpyro==0.19.0 deliberately: newer jax silently freezes numpyro's NUTS at its initial point.

Quickstart

from saddm import make_ddmsa_model, sample_ddmsa, sample_ddmsa_exact

data  = sample_ddmsa_exact(a=1.1, z=0.5, v=1.5, t=0.25,
                           sv=0.8, sa=0.5, st=0.08, n_trials=2000)
idata = sample_ddmsa(make_ddmsa_model(data), backend="numpyro")

saddm.ddmsa_logp(rt, response, a, z, v, t, sv, sa, st, sz) is the per-trial log-likelihood; every parameter may be a scalar or a per-trial vector.

With HSSM, register saddm.ddmsa_logp as a loglik_kind="analytical" likelihood; examples/estimate_HSSM_saddm.py is the minimal adapter.

Layout

path contents
saddm/ ddmsa.py: the likelihood and PyMC glue.
tests/ test_ddmsa.py: verification suite — s = 1 closed forms, agreement with the Numba/Fortran reference, finite-difference gradients, corner finiteness, per-trial broadcasting, backend agreement, static-zero collapse (run directly with --sample for an end-to-end NUTS check). reference.py is the Numba port of the Fortran density it is held to; test_reference.py covers that port.
verification/ parameter_recovery.py: 100-config NUTS recovery study. recovery_figure.py, compare_to_fortran.py, likelihood_figure.py: analysis and figures (read results/reference/ by default; set RESULTS for a fresh run).
examples/ HSSM applications: flat fit on cavanagh_theta, the per-subject + k-sweep replication of the Fortran intertemporal-choice analysis, hierarchical variants, and the random-effects figure.
fortran/ The Fortran programs that produced the benchmarks, with build notes.
data/itc_amasino/ Amasino et al. (2019) trials, the Fortran benchmarks, and the exact k-sweep permutation files.
results/reference/ Reference outputs: recovery, ITC, hierarchical, and cavanagh results with figures. Everything else under results/ is gitignored, and every script writes there by default.

Citation

@unpublished{shevlin2026little,
  author = {Shevlin, Blair R. K. and Fernandez, Kiant{\'e} and Ratcliff, Roger and Krajbich, Ian},
  title  = {A little goes a long way: Fitting one-shot decisions with cognitive models},
  note   = {Manuscript in preparation},
  year   = {2026},
}

Blair R. K. Shevlin* and Kianté Fernandez* contributed equally.

:Author: Kianté Fernandez, Blair R K Shevlin, Roger Ratcliff, Ian Krajbich

:Contact: kiante@ucla.edu, blair.shevlin@mssm.edu, ratcliff.22@osu.edu, krajbich@ucla.edu

Project links

Data verified by PyPI on Sep 4, 2026
Data provided by the project maintainers, verified at the time the release was uploaded to PyPI.

Key dates

PyPI data
Data sourced directly from PyPI's database.
  • Released:
Latest release

1 maintainer

PyPI data
Data sourced directly from PyPI's database.
Avatar for kiante from gravatar.com kiante

Credits

Author: Kianté Fernandez, Blair R K Shevlin, Roger Ratcliff, Ian Krajbich

GitHub Statistics

Data verified by PyPI on Sep 4, 2026
The GitHub source repository was provided by the project maintainers and verified by PyPI at the time of upload. Stars, forks, and open issues/PRs are derived from that repository and have not been independently verified.

License

MIT License (MIT)

Requires

Python >=3.11

Provides Extra

sampling hssm test

Classifiers

Report project as malware

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

saddm-0.2.0.tar.gz (18.3 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

saddm-0.2.0-py3-none-any.whl (10.5 kB view details)

Uploaded Python 3

File details

Details for the file saddm-0.2.0.tar.gz.

File metadata

  • Download URL: saddm-0.2.0.tar.gz
  • Upload date:
  • Size: 18.3 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for saddm-0.2.0.tar.gz
Algorithm Hash digest
SHA256 f7e4e007668b1bf7b3ecb39477471af7517bde4469371bc3f280ae4b6d80ecf8
MD5 012b77c5254c14306c57b1a81cdfae22
BLAKE2b-256 afa9bee5776f9ba6dfbf112c196a81703848cb62bce9c89bb95760da60a0bbe7

See more details on using hashes here.

Provenance

The following attestation bundles were made for saddm-0.2.0.tar.gz:

Publisher: publish.yml on kiante-fernandez/saddm

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file saddm-0.2.0-py3-none-any.whl.

File metadata

  • Download URL: saddm-0.2.0-py3-none-any.whl
  • Upload date:
  • Size: 10.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for saddm-0.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 c683e6e40f41570508a428680b9d48c7386ca6c0058b69695eb7d389c445d509
MD5 bd5e0e434ee7c78c0cca8f93e2e61db9
BLAKE2b-256 83aadc5623096fdd4ee2092f1ebcd19bf75556885b7c2a360838680e0cef5c80

See more details on using hashes here.

Provenance

The following attestation bundles were made for saddm-0.2.0-py3-none-any.whl:

Publisher: publish.yml on kiante-fernandez/saddm

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

This release

0.2.0 This release

2 files

0.1.1

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page