Univariate-Time-Series-Prediction-using-Deep-Learning

Univariate Time Series Prediction using Deep Learning and PyTorch

0. Introduction

This repository provides Univariate Time Series Prediction using deep learning models including DNN, CNN, RNN, LSTM, GRU, Recursive LSTM, and Attention LSTM.

The dataset used is Appliances Energy Prediction Data Set and can be found here.

1. Quantitative Analysis

According to the table below, CNN using 1D Convolutional layer outperformed the other models.

Model	MAE↓	MSE↓	RMSE↓	MPE↓	MAPE↓	R Squared↑
DNN	31.0077	4039.9806	57.8505	-16.4529	27.5759	0.4355
CNN	28.4919	3869.6289	56.6529	-11.5615	24.3810	0.4567
RNN	30.7757	3997.9815	57.8951	-19.2878	28.4873	0.4297
LSTM	29.8795	3949.6140	57.5196	-17.5516	27.2467	0.4393
GRU	29.9521	3939.7874	57.4498	-17.9298	27.4501	0.4402
Recursive LSTM	29.8795	3949.6140	57.5196	-17.5516	27.2467	0.4393
Attention LSTM	30.6609	3923.0855	57.2503	-17.8343	28.1153	0.4372

2. Qualitative Analysis

It definitely suffers from the typical lagging issue.

3. Run the Codes

If you want to train Attention LSTM,

1) Train

python main.py --model 'attention'

2) Test

python main.py --model 'attention' --mode 'test'

To handle more arguments, you can refer to main.py

Development Environment

- Windows 10 Home
- NVIDIA GFORCE RTX 2060
- CUDA 10.2
- torch 1.6.0
- torchvision 0.7.0
- etc