Implementation of Deep-Forecast using PyTorch

• Deep Forecast: Deep Learning-based Spatio-Temporal Forecasting

• Adapted from original implementation

Setup

• Clone this repository : git clone https://github.com/Wizaron/deep-forecast-pytorch.git

• Download and install Anaconda or Miniconda

• Go to the "reseg-pytorch/code/pytorch" : cd reseg-pytorch/code/pytorch

• Create environment : conda env create -f conda_environment.yml

• Activate environment : source activate deep-forecast-pytorch

Code Structure

• "data" : Stores data and scripts to prepare dataset for training.

• "lib" : Stores miscellaneous scripts for training and testing.

• "arch.py" : Defines network architecture

• "model.py" : Defines model (Minibatching mechanism, optimization, criterion, fit, predict, etc.)

• "prediction.py" : Metrics and plots to evaluate the performance of the trained model

• "data.py" : Creates training, validation and testings datasets

• "loader.py" : Creates Dataset loader for PyTorch

• "train.py" : Main training script.

• "test.py" : Main testing script.

• "settings.py" : Defines hyper-parameters of the model.

Data

• Data is downloaded from IEM

• Download data and save it under "data/raw"

• To prepare dataset, run the scripts in "data/scripts"

Training and Testing

• Train : python train.py --data [PATH OF PREPARED DATASET]

• Test : python test.py --data [PATH OF PREPARED DATASET] --model [PATH OF THE SAVED MODEL]

• For more info : python train.py --help, python test.py --help

train.py

• It saves models and logs under "models"

• At the end of the training, it saves predictions under "outputs"

test.py

• It saves predictions under the directory of the model.

Sample Outputs