DAS 0.33.2

Fix opening recordings with video and pose data on current NumPy by requiring xarray-behave 0.37.8 or later (below 0.38). The xarray-behave fix extracts the scalar tracking-frame index explicitly, avoiding the startup TypeError.

Upgrade in the existing DAS Conda environment:

uv pip install --upgrade-package das das --torch-backend=auto

DAS 0.33.1 — Conda GUI installation fixes

This patch fixes the supported installation path for the stable DAS GUI.

• Install PySide6 and FFmpeg together through conda-forge so Windows uses compatible Qt/ICU libraries.
• Require xarray-behave>=0.37.7,<0.38, restoring the DAS 0.33 Train/Predict dialogs and excluding the incompatible 0.38 alpha API.
• Install numcodecs<0.16 through Conda for Python 3.14.
• Select pthreads OpenBLAS on macOS to avoid a duplicate OpenMP runtime with PyTorch.
• Add GUI startup and Train/Predict integration tests across Windows, macOS, and Linux on Python 3.12–3.14.

Validation: all nine post-merge install/test jobs passed, and the release documentation deployment succeeded. The published PyPI artifacts match the verified build.

Follow the updated installation instructions to create an environment with Conda-provided Qt and FFmpeg.

DAS 0.33.0 — PyTorch backend

DAS 0.33.0 moves the supported backend from TensorFlow to PyTorch through Keras 3.

Highlights:

• Loads and predicts with legacy DAS model catalogs, including TensorFlow-era H5 models.
• Preserves the existing train, predict, evaluate, and GUI command-line interfaces.
• Uses xarray-behave 0.37.5 with PySide 6.10 support.
• Supports NumPy 2 result saving and the modern .keras model format.
• Prevents NaN gradients in the trainable STFT frontend on silent frames.
• Requires Python 3.12 or newer.

The full install matrix passed on Python 3.12–3.14 across macOS, Linux, and Windows. The legacy model menagerie and the larger support-model catalog were also verified.

Users who need the TensorFlow backend should install the final TensorFlow-backed release: das==0.32.13.

DAS 0.32.13 — final TensorFlow release

DAS 0.32.13 is the final TensorFlow-backed release.

Changes since 0.32.12:

• Removed morphology layers and morphology-based model variants.
• Removed automatic model tuning and the das tune command.

To remain on the TensorFlow-backed release after DAS moves to PyTorch, install das==0.32.13.

Validated on macOS, Linux, and Windows.

DAS 0.32.12 — final TensorFlow release

DAS 0.32.12 is the final TensorFlow-backed release.

It preserves the existing DAS interface and adds strict-first model weight loading with a warned compatibility fallback for older models. To remain on the TensorFlow-backed release after DAS moves to PyTorch, install das==0.32.12.

Validated on macOS, Linux, and Windows.

DAS v0.22.3

Merge branch 'master' of https://github.com/janclemenslab/das

Tutorial data

Recording of a Drosophila melanogaster male courting a female.

Data recorded by David Stern and publicly available here:
https://www.janelia.org/lab/stern-lab/tools-reagents-data