42bPhD / TSCP2_pytorch

Time Series Change Point Detection based on Contrastive Predictive Coding pytorch implementation

TSCP2_pytorch

Time Series Change Point Detection based on Contrastive Predictive Coding pytorch implementation

• This repo covers an reference implementation for the following papers in PyTorch. Official Code and Paper as follow: Tensorflow2: Repository and TSCP2: Deldari, Shohreh and Smith, Daniel V. and Xue, Hao and Salim, Flora D.

Dataset

Go to Page

Differents of code between mine and author's

1. Pytorch Implementation
2. Add Early Stopping approach Using Trainer Class
3. Add Spatial Dropout approach
4. Offer a different set of hyperparameters using Hydra
5. Add Toy datasets
6. Add Attention Mechanism and Batchnorm

Requirements

Python 3.9.13

To install requirements:

pip install -r requirements.txt   or
conda create --name <env> --file conda_requirement.txt   or
conda create --name <env> --file environment.yml

Train

python main.py

Reference Papers

1. Time Series Change Point Detection with Self-Supervised Contrastive Predictive Coding
2. Temporal Convolution Network
3. Representation Learning with Contrastive Predictive Coding
4. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models
5. InfoNCE_Loss
6. Greedy_InfoMax
7. Kernel Change-point Detection with Auxiliary Deep Generative Models