Autolfads-tf2 - A TensorFlow 2.0 implementation of Latent Factor Analysis via Dynamical Systems (LFADS) and AutoLFADS

Overview

autolfads-tf2

A TensorFlow 2.0 implementation of LFADS and AutoLFADS.

Installation

Clone the autolfads-tf2 repo and create and activate a conda environment with Python 3.7. Use conda to install cudatoolkit and cudnn and pip install the lfads_tf2 and tune_tf2 packages with the -e (editable) flag. This will allow you to import these packages anywhere when your environment is activated, while also allowing you to edit the code directly in the repo.

git clone [email protected]:snel-repo/autolfads-tf2.git
cd autolfads-tf2
conda create --name autolfads-tf2 python=3.7
conda activate autolfads-tf2
conda install -c conda-forge cudatoolkit=10.0
conda install -c conda-forge cudnn=7.6
pip install -e lfads-tf2
pip install -e tune-tf2

Usage

Training single models with lfads_tf2

The first step to training an LFADS model is setting the hyperparameter (HP) values. All HPs, their descriptions, and their default values are given in the defaults.py module. Note that these default values are unlikely to work well on your dataset. To overwrite any or all default values, the user must define new values in a YAML file (example in configs/lorenz.yaml).

The lfads_tf2.models.LFADS constructor takes as input the path to the configuration file that overwrites default HP values. The path to the modeled dataset is also specified in the config, so LFADS will load the dataset automatically.

The train function will execute the training loop until the validation loss converges or some other stopping criteria is reached. During training, the model will save various outputs in the folder specified by MODEL_DIR. Console outputs will be saved to train.log, metrics will be saved to train_data.csv, and checkpoints will be saved in lfads_ckpts.

After training, the sample_and_average function can be used to compute firing rate estimates and other intermediate model outputs and save them to posterior_samples.h5 in the MODEL_DIR.

We provide a simple example in example_scripts/train_lfads.py.

Training AutoLFADS models with tune_tf2

The autolfads-tf2 framework uses ray.tune to distribute models over a computing cluster, monitor model performance, and exploit high-performing models and their HPs.

Setting up a ray cluster

If you'll be running AutoLFADS on a single machine, you can skip this section. If you'll be running across multiple machines, you must initialize the cluster using these instructions before you can submit jobs via the Python API.

Fill in the fields indicated by <>'s in the ray_cluster_template.yaml, and save this file somewhere accessible. Ensure that a range of ports is open for communication on all machines that you intend to use (e.g. 10000-10099 in the template). In your autolfads-tf2 environment, start the cluster using ray up <NEW_CLUSTER_CONFIG>. The cluster may take up to a minute to get started. You can test that all machines are in the cluster by ensuring that all IP addresses are printed when running example_scripts/ray_test.py.

Starting an AutoLFADS run

To run AutoLFADS, copy the run_pbt.py script and adjust paths and hyperparameters to your needs. Make sure to only use only as many workers as can fit on the machine(s) at once. If you want to run across multiple machines, make sure to set SINGLE_MACHINE = False in run_pbt.py. To start your PBT run, simply run run_pbt.py. When the run is complete, the best model will be copied to a best_model folder in your PBT run folder. The model will automatically be sampled and averaged and all outputs will be saved to posterior_samples.h5.

References

Keshtkaran MR, Sedler AR, Chowdhury RH, Tandon R, Basrai D, Nguyen SL, Sohn H, Jazayeri M, Miller LE, Pandarinath C. A large-scale neural network training framework for generalized estimation of single-trial population dynamics. bioRxiv. 2021 Jan 1.

Keshtkaran MR, Pandarinath C. Enabling hyperparameter optimization in sequential autoencoders for spiking neural data. Advances in Neural Information Processing Systems. 2019; 32.

Comments
  • Update lfads-tf2 dependencies for Google Colab compatibility

    Update lfads-tf2 dependencies for Google Colab compatibility

    Summary of changes to setup.py

    • Change pandas==1.0.0 to pandas==1.* to avoid a dependency conflict with google-colab
    • Add PyYAML>=5.1 so that yaml.full_loadworks in lfads-tf2.
    opened by yahiaali 0
  • Are more recent versions of tensorflow/CUDA supported by the package?

    Are more recent versions of tensorflow/CUDA supported by the package?

    Right now the package supports TF 2.0 and CUDA 10.0 which are more than 3 years old. Is there support planned/already established for more recent Tensorflow and CUDA versions?

    Thanks!

    opened by stes 0
  • Error: No 'git' repo detected for 'lfads_tf2'

    Error: No 'git' repo detected for 'lfads_tf2'

    Hello, I am having this issue. I have followed all the installation instructions, and I was wondering why this issue would come up. autolfads-tf2 is cloned using git, and it is inside the git folder. But it seems like train_lfads.py is not loading data. I am using Window 10.

    error

    Thank you so much in advance!

    opened by jinoh5 0
  • Add warnings and assertion to chop functions for bad overlap

    Add warnings and assertion to chop functions for bad overlap

    Add warnings and assertion to chop functions when requested overlap is greater than half of window length

    Addresses https://github.com/snel-repo/autolfads-tf2/issues/2

    opened by raeedcho 0
  •  `merge_chops` is unable to merge when the requested overlap is more than half of the window length

    `merge_chops` is unable to merge when the requested overlap is more than half of the window length

    Without really thinking a whole lot about it, I chopped data to window length 100 and overlap 80, since this would leave at most 20 points of unmodeled data at the end of the trials I'm trying to model. The chopping seems to work totally fine, but when merging the chops together, it seems that the code assumes that the overlap will be at most half the size of the window, and the math to put the chops back together breaks down in weird ways, leading to duplicated data in the final array.

    On further thought, it makes sense to some degree to limit the overlap to be at most half of the window length, since otherwise, data from more than two chops would have to be integrated together to merge everything--if this is the thought process, I think it would be a good idea to put an assertion in both functions that this is the case (or maybe at least an assertion in the merge_chops function and a warning in the chop_data function, since chopping technically works fine).

    If instead it would make sense to be able to merge chops with overlap greater than half the window size, then I think the merge_chops function needs to be reworked to be able to integrate across more than two chops

    opened by raeedcho 0
Releases(v0.1)
Owner
Systems Neural Engineering Lab
Emory University and Georgia Institute of Technology
Systems Neural Engineering Lab
Serverless proxy for Spark cluster

Hydrosphere Mist Hydrosphere Mist is a serverless proxy for Spark cluster. Mist provides a new functional programming framework and deployment model f

hydrosphere.io 317 Dec 01, 2022
BlockUnexpectedPackets - Preventing BungeeCord CPU overload due to Layer 7 DDoS attacks by scanning BungeeCord's logs

BlockUnexpectedPackets This script automatically blocks DDoS attacks that are sp

SparklyPower 3 Mar 31, 2022
This is an official source code for implementation on Extensive Deep Temporal Point Process

Extensive Deep Temporal Point Process This is an official source code for implementation on Extensive Deep Temporal Point Process, which is composed o

Haitao Lin 8 Aug 15, 2022
GULAG: GUessing LAnGuages with neural networks

GULAG: GUessing LAnGuages with neural networks Classify languages in text via neural networks. Привет! My name is Egor. Was für ein herrliches Frühl

Egor Spirin 12 Sep 02, 2022
Official implementation of Deep Reparametrization of Multi-Frame Super-Resolution and Denoising

Deep-Rep-MFIR Official implementation of Deep Reparametrization of Multi-Frame Super-Resolution and Denoising Publication: Deep Reparametrization of M

Goutam Bhat 39 Jan 04, 2023
A lightweight library to compare different PyTorch implementations of the same network architecture.

TorchBug is a lightweight library designed to compare two PyTorch implementations of the same network architecture. It allows you to count, and compar

Arjun Krishnakumar 5 Jan 02, 2023
FwordCTF 2021 Infrastructure and Source code of Web/Bash challenges

FwordCTF 2021 You can find here the source code of the challenges I wrote (Web and Bash) in FwordCTF 2021 and the source code of the platform with our

Kahla 5 Nov 25, 2022
for a paper about leveraging discourse markers for training new models

TSLM-DISCOURSE-MARKERS Scope This repository contains: (1) Code to extract discourse markers from wikipedia (TSA). (1) Code to extract significant dis

International Business Machines 6 Nov 02, 2022
An unsupervised learning framework for depth and ego-motion estimation from monocular videos

SfMLearner This codebase implements the system described in the paper: Unsupervised Learning of Depth and Ego-Motion from Video Tinghui Zhou, Matthew

Tinghui Zhou 1.8k Dec 30, 2022
Import Python modules from dicts and JSON formatted documents.

Paker Paker is module for importing Python packages/modules from dictionaries and JSON formatted documents. It was inspired by httpimporter. Important

Wojciech Wentland 1 Sep 07, 2022
A toolkit for developing and comparing reinforcement learning algorithms.

Status: Maintenance (expect bug fixes and minor updates) OpenAI Gym OpenAI Gym is a toolkit for developing and comparing reinforcement learning algori

OpenAI 29.6k Jan 08, 2023
NCVX (NonConVeX): A User-Friendly and Scalable Package for Nonconvex Optimization in Machine Learning.

The source code is temporariy removed, as we are solving potential copyright and license issues with GRANSO (http://www.timmitchell.com/software/GRANS

SUN Group @ UMN 28 Aug 03, 2022
Navigating StyleGAN2 w latent space using CLIP

Navigating StyleGAN2 w latent space using CLIP an attempt to build sth with the official SG2-ADA Pytorch impl kinda inspired by Generating Images from

Mike K. 55 Dec 06, 2022
Neural network for recognizing the gender of people in photos

Neural Network For Gender Recognition How to test it? Install requirements.txt file using pip install -r requirements.txt command Run nn.py using pyth

Valery Chapman 1 Sep 18, 2022
PyMatting: A Python Library for Alpha Matting

Given an input image and a hand-drawn trimap (top row), alpha matting estimates the alpha channel of a foreground object which can then be composed onto a different background (bottom row).

PyMatting 1.4k Dec 30, 2022
Wordplay, an artificial Intelligence based crossword puzzle solver.

Wordplay, AI based crossword puzzle solver A crossword is a word puzzle that usually takes the form of a square or a rectangular grid of white- and bl

Vaibhaw 4 Nov 16, 2022
A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation

Segnet is deep fully convolutional neural network architecture for semantic pixel-wise segmentation. This is implementation of http://arxiv.org/pdf/15

Pradyumna Reddy Chinthala 190 Dec 15, 2022
Semi-Supervised Graph Prototypical Networks for Hyperspectral Image Classification, IGARSS, 2021.

Semi-Supervised Graph Prototypical Networks for Hyperspectral Image Classification, IGARSS, 2021. Bobo Xi, Jiaojiao Li, Yunsong Li and Qian Du. Code f

Bobo Xi 7 Nov 03, 2022
"Projelerle Yapay Zeka Ve Bilgisayarlı Görü" Kitabımın projeleri

"Projelerle Yapay Zeka Ve Bilgisayarlı Görü" Kitabımın projeleri Bu Github Reposundaki tüm projeler; kaleme almış olduğum "Projelerle Yapay Zekâ ve Bi

Ümit Aksoylu 4 Aug 03, 2022
Keras Image Embeddings using Contrastive Loss

Image to Embedding projection in vector space. Implementation in keras and tensorflow of batch all triplet loss for one-shot/few-shot learning.

Shravan Anand K 5 Mar 21, 2022