Demo Notebooks

We provide a set of demo notebooks to get started with using CEBRA. To run the notebooks, you need a working Jupyter notebook server, a CEBRA installation, and the datasets required to run the notebooks, available on FigShare <https://figshare.com/s/60adb075234c2cc51fa3>_.

.. nbgallery:: :maxdepth: 2

Getting Started with CEBRA <demo_notebooks/CEBRA_best_practices.ipynb> Encoding of space, hippocampus (CA1) <demo_notebooks/Demo_hippocampus.ipynb> Decoding movie features from (V1) visual cortex <demo_notebooks/Demo_Allen.ipynb> Forelimb dynamics, somatosensory (S1) <demo_notebooks/Demo_primate_reaching.ipynb> Synthetic neural benchmarking <demo_notebooks/Demo_synthetic_exp.ipynb> Hypothesis-driven analysis <demo_notebooks/Demo_hypothesis_testing.ipynb> Consistency <demo_notebooks/Demo_consistency.ipynb> Decoding <demo_notebooks/Demo_decoding.ipynb> Topological data analysis <demo_notebooks/Demo_cohomology.ipynb> Technical: Training models across animals <demo_notebooks/Demo_hippocampus_multisession.ipynb> Technical: conv-piVAE <demo_notebooks/Demo_conv-pivae.ipynb> Technical: S1 training with MSE loss <demo_notebooks/Demo_primate_reaching_mse_loss.ipynb> Technical: Learning the temperature parameter <demo_notebooks/Demo_learnable_temperature.ipynb> Demo: Using OpenScope Data <demo_notebooks/Demo_openscope_databook.ipynb> Demo: Using Dandi Data <demo_notebooks/Demo_dandi_NeuroDataReHack_2023.ipynb> Explainability: xCEBRA on RatInABox dataset <demo_notebooks/Demo_xCEBRA_RatInABox.ipynb>

The demo notebooks can also be found on GitHub <https://github.com/AdaptiveMotorControlLab/CEBRA-demos>__.

Installation

Before you can run these notebooks, you must have a working installation of CEBRA. Please see the dedicated :doc:Installation Guide </installation> for information on installation options using conda, pip and docker.

Synthetic Experiment Demo (CEBRA, piVAE, tSNE, UMAP): This demo requires several additional packages that have differing requirements to CEBRA. Therefore, we recommend using the supplied docker container or conda cebra-full env.

Demo Data

We host prepackaged data on figshare <https://figshare.com/s/60adb075234c2cc51fa3>__. And several of the demo notebooks have an automatic data download function.

If you don't see the auto-download, and you use Google Colaboratory, you can easily add the following code into an early cell in the notebook to directly download and use:

.. code-block::

#for google colab only, run this cell to download and extract data: !wget --content-disposition https://figshare.com/ndownloader/files/36869049?private_link=60adb075234c2cc51fa3 !mkdir data !tar -xvf "/content/data.tgz" -C "/content/data"

For different paths, you can specify the CEBRA_DATADIR=... environment variable. You can do this by placing import os; os.environ['CEBRA_DATADIR'] = "path/to/your/data" at the top of your notebook.

Contributing

We welcome Demo notebooks from others! Please fork the repo <https://github.com/AdaptiveMotorControlLab/CEBRA-demos>, add your notebook, check that it works on Google Colaboratory (remove the launch button within your PR), and then open a PR! Please also edit the "gallery" list (see the soure code for this page), and finally, add an icon here <https://github.com/AdaptiveMotorControlLab/CEBRA-assets>, then a path to the icon here <https://github.com/AdaptiveMotorControlLab/CEBRA/blob/main/docs/source/conf.py>__.

For reference, the original open-source data we used in Schneider, Lee, Mathis 2023 is available at:

• Hippocampus dataset <https://crcns.org/data-sets/hc/hc-11/about-hc-11>, using a preprocessing script <https://github.com/zhd96/pi-vae/blob/main/code/rat_preprocess_data.py>.
• Primate S1 dataset <https://gui.dandiarchive.org/#/dandiset/000127>_.
• Allen Institute Neuropixels dataset <https://allensdk.readthedocs.io/en/latest/visual_coding_neuropixels.html>_ and 2P dataset <https://allensdk.readthedocs.io/en/latest/>_.