https://github.com/nullhawk/time-series-prediction
Predicting time sequence data using LSTM
https://github.com/nullhawk/time-series-prediction
deep-learning lstm pytorch
Last synced: about 1 month ago
JSON representation
Predicting time sequence data using LSTM
- Host: GitHub
- URL: https://github.com/nullhawk/time-series-prediction
- Owner: nullHawk
- Created: 2024-11-26T09:22:21.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-11-26T12:01:23.000Z (over 1 year ago)
- Last Synced: 2025-03-29T07:45:21.387Z (about 1 year ago)
- Topics: deep-learning, lstm, pytorch
- Language: Python
- Homepage:
- Size: 9.77 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Time Series Prediction with LSTM
This project demonstrates a time-series prediction model for predicting a sine wave using PyTorch. It employs an LSTM-based architecture to learn and predict the behavior of a sine wave, showcasing the capability of recurrent neural networks for sequence modeling.
## Features
- Generates synthetic sine wave data for training and testing.
- Trains an LSTM-based model to predict future values of the sine wave.
- Visualizes predictions after each training step.
- Saves predictions as `.pdf` files for review.
## Requirements
- Python 3.x
- PyTorch
- NumPy
- Matplotlib
Install the required packages using:
```bash
pip install torch numpy matplotlib
```
## Usage
### 1. Generate Data
The sine wave data is generated with added random shifts. The `generate_sinewave.py` script provides:
- `generate()`: Creates training and testing data.
- `plot()`: Visualizes the sine wave.
### 2. Train the Model
Run `train.py` to train the LSTM model:
```bash
python train.py
```
During training:
- The model learns the sine wave pattern.
- Predictions for unseen data are plotted and saved at each training step (e.g., `predict1.pdf`, `predict2.pdf`, etc.).
### 3. Model Architecture
The `LSTMPredictor` class in `model.py` defines a custom LSTM model:
- **Two LSTM Cells**: For capturing complex temporal patterns.
- **Linear Layer**: Maps LSTM outputs to the target space.
### Results
Below is an example visualization of the model's predictions:
- **Solid Line**: Actual sine wave.
- **Dashed Line**: Predicted future values.
The model progressively learns the pattern as training progresses.
## File Structure
```
├── train.py # Main script to train the model
├── model.py # LSTM model definition
├── generate_sinewave.py # Data generation and plotting
├── predictX.pdf # Generated prediction plots (after each training step)
```
## Example Output
Training logs:
```
Step: 0
Loss: 0.0456
Test loss: 0.0312
Step: 1
Loss: 0.0213
Test loss: 0.0187
...
```
Plots (`predictX.pdf`) show how the model's predictions improve over steps.
---
Feel free to clone this repository and experiment with time-series prediction using LSTM!