Predict the future BTCUSDT
cryptocurrency exchange prices using neural network.
-
Install the dependences:
$ pip install -U numpy tensorflow matplotlib
-
Build the
BTCUSDT
dataset using CryptocurrencyPriceDataset. -
Create a symbolic directory link named
data
targeting thedata
directory containing the datasets. -
Modify
./config.py
if needed. -
Run this script to preprocess the data:
$ python ./preprocess.py
Preprocessed data is saved in
./preprocessed/BTCUSDT.npz
. -
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
-
Run this script to plot the prediction (and real values):
$ python ./plot.py
┌────────────┐
│ InputLayer │
└─────┬──────┘
│
┌─────▼─────┐
│ GRU │
└─────┬─────┘
│
┌─────▼─────┐
│ LTSM │
└─────┬─────┘
│
┌─────▼─────┐
│ Dense │
└─────┬─────┘
│
┌─────▼─────┐
│ Dense │
└───────────┘