nkid00 / futurecrypto

Predict the future BTCUSDT cryptocurrency exchange prices using neural network.

FutureCrypto

Predict the future BTCUSDT cryptocurrency exchange prices using neural network.

Usage

1. Install the dependences:

   $ pip install -U numpy tensorflow matplotlib

2. Build the BTCUSDT dataset using CryptocurrencyPriceDataset.

3. Create a symbolic directory link named data targeting the data directory containing the datasets.

4. Modify ./config.py if needed.

5. Run this script to preprocess the data:

   $ python ./preprocess.py

   Preprocessed data is saved in ./preprocessed/BTCUSDT.npz.

6. Run this script to train the model:

   $ python ./train.py

   Trained model is saved on every epoch in ./model/<Time>_<Sum of previous epoches>_<Current epoch>/.

   Or run this script to restore semi-trained model from the latest epoch:

   $ python ./restore.py

7. Run this script to plot the prediction (and real values):

   $ python ./plot.py

Model Structure

┌────────────┐
│ InputLayer │
└─────┬──────┘
      │
┌─────▼─────┐
│    GRU    │
└─────┬─────┘
      │
┌─────▼─────┐
│   LTSM    │
└─────┬─────┘
      │
┌─────▼─────┐
│   Dense   │
└─────┬─────┘
      │
┌─────▼─────┐
│   Dense   │
└───────────┘