This code implements pi-VAE, VAE and all the examples in paper: Learning identifiable and interpretable latent models of high-dimensional neural activity using pi-VAE.
To check source code for pi-VAE and VAE, see folder ./code/
To check examples in paper, see folder ./examples/
We are working on adding more comments to the code.
The code is written in Python 3.6. In addition to standard scientific Python libraries (numpy, scipy, matplotlib), the code expects the tensorflow (1.13.1) and keras (2.3.1) packages.
To download this code, run git clone https://github.com/zhd96/pi-vae.git
You can find code to generate the simulated data in our paper ./examples/simulate_data.ipynb.
You can find the raw rat hippocampus data in https://portal.nersc.gov/project/crcns/download/hc-11/data/. We use session Achilles_10252013. We include the code to preprocess the rat and macaque data in ./code/. You can download the raw rat data and preprocess it yourself, or you can download our preprocessed data in https://drive.google.com/drive/folders/1lUVX1IvKZmw-uL2UWLxgx4NJ62YbCwMo?usp=sharing.
For simulated data, please use ./examples/simulate_data.ipynb to generate data first, then follow pi-vae_simulated_data_xx.ipynb to run it.
For real data, please check the corresponding ipynb in ./examples/ to run it.
We also save the model results for real and simulated data in our paper in ./results/ You can check them first to reproduce the figures in the paper.
If you use this code please cite the paper:
Zhou, D., Wei, X. Learning identifiable and interpretable latent models of high-dimensional neural activity using pi-VAE. NeurIPS 2020. https://arxiv.org/abs/2011.04798
-
AD Grosmark, JD Long, and G Buzsáki. Recordings from hippocampal area ca1, pre, during and post novel spatial learning. crcns. org, 2016. doi:10.6080/K0862DC5. (http://crcns.org/data-sets/hc/hc-11)
-
Andres D Grosmark and György Buzsáki. Diversity in neural firing dynamics supports both rigid and learned hippocampal sequences. Science, 351(6280):1440–1443, 2016.