Skip to content

CHANGELOG.md

Latest commit

 

History

History
39 lines (32 loc) · 2.02 KB

CHANGELOG.md

File metadata and controls

39 lines (32 loc) · 2.02 KB
Raw

Changelog

All notable changes to deepSTRF are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning. While the version is 0.x, the public API may still change between minor releases.

Unreleased

0.1.0 - 2026-06-03

First public release, available on PyPI: pip install deepSTRF.

Added

  • Datasets — a zoo of auditory neural-recording datasets (NS1, CRCNS AA1/AA2/AA4, NAT4, CRCNS-AC1, Espejo, Downer 2025, Wingert 2026, Le 2025, Alice EEG) on a common NeuralDataset API with shape (B, N, R, T), dict-based batches, download=True auto-download where data is publicly mirrored, and a filter API (select_pop_by_*, select_stims_by_*).
  • Waveform input — optional raw-waveform branch on every audio dataset, with a wav2spec front-end zoo (CausalMel, SincNet, LEAF, gammatone/-gram).
  • Models — a four-slot encoding template (wav2spec → prefiltering → core → readout): Linear/LN, ConvNet2D, Transformer, StateNet (GRU/Mamba/S4/LMU), DNet, NetworkReceptiveField. Strictly causal in eval mode; output rank (B, N, R=1, T). Pluggable parametric STRF kernels and parametric activations.
  • Metrics — NaN-aware functional metrics (corrcoef, normalized corrcoef, FVE, Sahani–Linden SNR, CCmax, coherence) and Poisson/MSE losses.
  • Training — an opt-in Fitter (early stopping + best-checkpoint selection, optional per-cell restoration), multi-seed sweeps (fit_multi_seed), and optional Weights & Biases / TensorBoard loggers.
  • Pretrained weights — load checkpoints from the Hugging Face Hub via from_pretrained; save/push via save_pretrained / push_to_hub.
  • Packaging — published to PyPI via GitHub Actions Trusted Publishing (OIDC); ships inline type hints (PEP 561 py.typed).