JulesBelveze / time-series-autoencoder

PyTorch Dual-Attention LSTM-Autoencoder For Multivariate Time Series

LSTM-autoencoder with attentions for multivariate time series

This repository contains an autoencoder for multivariate time series forecasting. It features two attention mechanisms described in A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction and was inspired by Seanny123's repository.

Download and dependencies

To clone the repository please run:

git clone https://github.com/JulesBelveze/time-series-autoencoder.git

To install all the required dependencies please run:

python3 -m venv .venv/tsa
source .venv/tsa/bin/activate
poetry install

Usage

The project uses Hydra as a configuration parser. You can simply change the parameters directly within your .yaml file or you can override/set parameter using flags (for a complete guide please refer to the docs).

python3 main.py -cn=[PATH_TO_FOLDER_CONFIG] -cp=[CONFIG_NAME]

Optional arguments:

  -h, --help            show this help message and exit
  --batch-size BATCH_SIZE
                        batch size
  --output-size OUTPUT_SIZE
                        size of the ouput: default value to 1 for forecasting
  --label-col LABEL_COL
                        name of the target column
  --input-att INPUT_ATT
                        whether or not activate the input attention mechanism
  --temporal-att TEMPORAL_ATT
                        whether or not activate the temporal attention
                        mechanism
  --seq-len SEQ_LEN     window length to use for forecasting
  --hidden-size-encoder HIDDEN_SIZE_ENCODER
                        size of the encoder's hidden states
  --hidden-size-decoder HIDDEN_SIZE_DECODER
                        size of the decoder's hidden states
  --reg-factor1 REG_FACTOR1
                        contribution factor of the L1 regularization if using
                        a sparse autoencoder
  --reg-factor2 REG_FACTOR2
                        contribution factor of the L2 regularization if using
                        a sparse autoencoder
  --reg1 REG1           activate/deactivate L1 regularization
  --reg2 REG2           activate/deactivate L2 regularization
  --denoising DENOISING
                        whether or not to use a denoising autoencoder
  --do-train DO_TRAIN   whether or not to train the model
  --do-eval DO_EVAL     whether or not evaluating the mode
  --data-path DATA_PATH
                        path to data file
  --output-dir OUTPUT_DIR
                        name of folder to output files
  --ckpt CKPT           checkpoint path for evaluation 

Features

• handles multivariate time series
• attention mechanisms
• denoising autoencoder
• sparse autoencoder

Examples

You can find under the examples scripts to train the model in both cases:

• reconstruction: the dataset can be found here
• forecasting: the dataset can be found here