Part of my master project, which includes the following algorithms
- ARIMA
- Gaussian Process
- Multi-Task Gaussian Process
- Multi-Task Index Gaussian Process (Use index of each task)
- Deep Gaussian Process with Multi-Task Output
- Deep Sigma Point Process with Multi-Task Output
- Sparse Multi-Task Index Gaussian Process
- Sparse Matern Graph Gaussian Process
- Deep Graph Kernel
- Deep Graph Kernel + Deep Graph Infomax Pretraining
- Cluster Multi-Task GP (Pyro + Gpytorch)
- Non-Linear Deep Multi-Task GP
- Non-Linear Deep Sigma Point Process
- Cluster Non-Linear Deep Multi-Task GP
- Cluster Non-Linear Deep Sigma Point Process
- Learning Graph GP
- Graph Propagation Deep GP
- Interaction Net Deep GP
- DSPP Graph Propagation GP
- Interaction Net DSPP
- Non-Linear Deep Multi-Task GP Multi-Output
- Non-Linear Deep Sigma Point Process Multi-Output
See main.py
for examples. Running a Test for Data-Splitting Algorithm. The data should be stored in data/{metal_name}
.
In order to run the experiments, we assume to have a raw_data
folder that contains folders that named after the commodities, which have {commodity name}_feature.csv
and {commodity name}_raw_prices.csv
(this should be raw prices not log of it) storted within. To create a preprocessed data that is saved in folder data
, we run save_date_common("raw_data", "data")
from utils.data_preprocessing
.
We can run the test by:
python -m pytest
One may be interested in training the GP within google colabs, we have provided a simple way to zip the necessary files/folder
sh upload/zip_folders.sh
where we can upload to the colabs, extract the file and then perform the training.