Awesome

DXtreMM_TSF

Repository for paper, "Deep Extreme Mixture Model for Time series forecasting" CIKM ACM conference, 2022 paper

This model combines two modules: 1) Variation Disentangled Autoencoder based classifier 2) GPD based and Normal Forecaster modules

VD-AE classifier is the extension of work titled "Variational Disentanglement for Rare Event Modeling". GPD prior is extended for left extremes.

The complete implemetation of classifier modules is given in VD_AE_classifer folder

Training the classifier model can be done by executing

python train_VIE_multi_simulationDL_cv.py

Anomaly detection results are saved in same folder

Forecaster modules training and inference implementaion is given in extreme_gpd folder

extreme_gpd_3cls.ipynb invokes training function and infers predicton of unseen data

Step by step procedure for working with code is as follows:

1. install pytorch (you can follow instructions from [here](https://pytorch.org/get-started/locally/))
2. pip install -r requirements.txt
3. Change parameters according to your data in train_VIE_multi_simulationDL_cv.py file
4. Get classification results and use that for forecasting module

Requirements file can be installed using conda environemnt by following instructions from here or using docker as given here