wavecast: A forecasting model based on WaveNet architecture

This work has been done as part of Master's thesis titled "WaveNet Architectures for Time Series Forecasting".

Author: Naveen Kaushik

Supervisor: Dr. Christoph Bergemeir

Published date: July 15, 2020

Installation and Usage

Software Requirements:

python>=3.6

tensorflow==1.13

keras==2.2.4

Usage:

python wavecast.py --dataset_name "dummy_ts_data" --train_file "train.csv" --test_file "test.csv" --training_steps 72 --forecast_horizon 12 --output_file "forecast.csv" --frac 1.0 --result_file "results.csv"

The parameters used are explained as follows:

• dataset_name - a unique string to identify the dataset
• train_file - the training data file for training of the model
• test_file - the test data file for evaluating the model
• training_steps - number of timesteps to be considered for training window
• forecast_horizon - the forecasting horizon for the dataset
• output_file - file to write the forecasted values
• frac - fraction of input data to be used
• result_file - file to write the results based on evaluation metric and hyperparameter values

Most of the datasets used in the experiments are available here.