This is a Python program designed to help you predict time series data using deep learning models, specifically using LSTM (Long Short-Term Memory) units. This program works with any dataset that includes time-series data and allows you to customize various aspects of the prediction process.
- pip install -r requirements.txt
- Load your data from a CSV file.
- Select specific time frames for analysis.
- Choose the number of columns to use from your dataset.
- Customize deep learning model settings such as epochs, batch size, and more.
- Reduce memory usage by adjusting data types.
- Validate the dataset to ensure it contains the correct data types.
- Predict future values based on historical data.
- Download the program files to your local machine.
- Prepare your CSV data file. Ensure it has a datetime index and the columns you wish to analyze.
- Modify the parameters in the
multi_RNNclass instantiation in the script to match your data and preferences:
df: Path to your CSV file.time_stamp_start: Start date for the data analysis.time_stamp_end: End date for the data analysis.- Other parameters like
epochs,batch_size,LSTM_units, etc., as needed.
-
Run the script from your command line: python multi_RNN.py
-
Check the output files for predictions and model performance plots.
You can customize the program by changing the parameters in the multi_RNN class:
- Change
columns_left_in_dfto select how many features (columns) you want to include from your dataset. - Adjust
float_16,float_32, orfloat_64to manage memory usage depending on your dataset size. - Modify
LSTM_unitsandlengthto tweak the LSTM model configuration.
The program will output:
- Predictions as CSV files.
- Plots comparing predicted values with actual values.
- Training and validation loss plots to evaluate model performance.
- And a h5 model.
This project is licensed under the MIT License - see the LICENSE file for details.
Thank you for using our Time Series Prediction Program! We hope it assists you effectively in your data analysis tasks.