We are training a deep learning model to find the optimal trading strategy on the BTC dataset ¹. An important assumption that we are making is that there are no fees.

Make sure that you have Anaconda or Miniconda installed, or something similar.

$ git clone https://github.com/TorNATO-PRO/crypto-forecasting.git

$ cd crypto-forecasting

$ conda env create -f environment.yml

$ conda activate crypto

(crypto) $ python main.py

Ensure that Docker is installed.

$ git clone https://github.com/TorNATO-PRO/crypto-forecasting.git

$ cd crypto-forecasting

$ docker build -it .

$ docker run crypto

• Create basic directory structure
• Implement Baseline
• Implement Oracle
• Added NNI for hyperparameter tuning
• Create custom model(s) for data
• Develop loss function(s) for comparing performance
• Figure out a way to measure power consumption and execution time
• Train on all datasets
• Train on more datasets

.
├── LICENSE
├── README.md
├── assets
│   └── datasets
│       ├── BTC-USD.csv
│       └── ETH-USD.csv
├── environment.yml
├── main.py
├── requirements.txt
└── src
    ├── __init__.py
    ├── data_loader
    │   ├── __init__.py
    │   └── load.py
    ├── models
    │   ├── __init__.py
    │   ├── baseline
    │   │   └── baseline.py
    │   ├── custom
    │   │   └── custom.py
    │   ├── loss.py
    │   ├── oracle
    │   │   ├── data
    │   │   │   └── oracle_best.pth
    │   │   └── oracle.py
    │   └── utils.py
    └── tuning
        ├── __init__.py
        ├── custom_hyp_conf.yml
        ├── custom_trial.py
        ├── oracle_hyp_conf.yml
        ├── oracle_trial.py
        ├── search_space_custom.json
        └── search_space_oracle.json