deepSTRF documentation

deepSTRF is a PyTorch library for system identification of sensory neurons — predicting trial-resolved neural responses (spikes, calcium fluorescence, EEG, intracellular potential, …) from naturalistic stimuli with deep neural network models. It bundles a growing zoo of publicly available recordings, a unified four-slot model template, pretrained checkpoints on the Hugging Face Hub, a NaN-aware metrics suite, and a thin opt-in training utility.

This site is the project’s reference documentation. Source code, issue tracker, and pull requests live on GitHub.

Note

The current release focuses on auditory datasets and models. The video API base classes are exposed under deepSTRF.datasets.video and deepSTRF.models.video, but the specific video loaders are parked on the archive/video-api-v0 branch and will be revived once rewritten against the modernized base class.

Getting started

If this is your first time here, the fastest path is:

  1. Install with pip install -e ".[dev]" — see Installation instructions.

  2. Skim the four design notes that define deepSTRF’s public contracts: data, models, metrics, and fitter. They are short and compose into the entire API surface.

  3. Open one of the runnable example notebooks — for instance crcns_aa_tutorial to load a dataset end-to-end, or load_pretrained_statenet_ns1 to grab a published Hugging Face checkpoint and score it. Each notebook opens in Colab in one click.

Tour of the documentation

The sidebar groups pages into five sections:

Quickstart

Installation, the four design notes that define the data / model / metrics / training contracts, a short walkthrough, supported file formats, and the publications index.

Datasets

The dataset zoo, split by modality. Each dataset has its own page with the original publication, licensing, the response shape, and instructions on how to obtain the data (auto-download where possible).

Models

The model zoo, the standalone AdapTrans module documentation, and the registry of pretrained checkpoints we publish on the Hugging Face Hub.

Examples

Runnable notebooks, also available under the examples/ folder of the repo and one click from Colab. Grouped into Start here (the gentlest end-to-end tutorials), Dataset inspection (per-dataset visual walkthroughs), and Advanced / analyses (parameterized STRFs, gradient-attribution receptive fields, learnable front-ends).

API

Auto-generated reference for every public class and function under deepSTRF.*.

Citing deepSTRF

If deepSTRF is useful for your work, please cite the relevant paper (BibTeX in the README citation section):

  • Rançon, Masquelier & Cottereau, A general model unifying the adaptive, transient and sustained properties of ON and OFF auditory neural responses, PLOS Computational Biology, 20 (8), 2024. DOI: 10.1371/journal.pcbi.1012288

  • Rançon, Masquelier & Cottereau, Temporal recurrence as a general mechanism to explain neural responses in the auditory system, Communications Biology, 8:1456, 2025. DOI: 10.1038/s42003-025-08858-3

A running list of downstream publications built on deepSTRF lives on the Publications page — pull requests welcome to add yours.