deepss 0.13.1

DeepSS

DeepSS

DeepSS is supervised. See dss-unsupervised for unsupervised tools to augment DeepSS.

Installation

conda create -n dss python=3.7
conda activate dss
conda install zarr
conda install tensorflow  # add -gpu to ensure GPU support
pip install deepss

DeepSS has been tested with python 3.7 and 3.8 and tensorflow versions 2.1, 2.2, and 2.3.

Tensorflow is not installed automatically to avoid interference with existing installations and to provide more control over versions. We recommend installation via conda install, but pip install should also work and typically installs a newer version. The manual install of zarr is required in some windows systems since the install via pip can be wonky.

Tutorials

There are four tutorial notebooks that illustrate all steps required for going from annotated data via training and evaluating a network to segmenting new recordings:

• Prepare training data
• Train the network
• Evaluate the network and fine tune inference parameters for predicting events like Drosophila song pulses or segments like Drosophila sine song or bird song syllables
• Inference on new data
• Realtime inference

For the tutorials to work, you first need to download some data and example models (266MB) from here and put the four folders in the same folder as the tutorials notebooks. The tutorial notebooks also have extra dependencies: conda install jupyterlab ipython tqdm ipywidgets -y

Acknowledgements

The following packages were modified and integrated into dss:

• Keras implementation of TCN models modified from keras-tcn (in dss.tcn)
• Trainable STFT layer implementation modified from kapre (in dss.kapre)
• FCN model code modified from LEAP

See the sub-module directories for the original READMEs.