Limit Order Book(L2) Prediction

This repo trys to predict price jumps from features derived from L2 order book information.

Order book information contains ask, bid prices and corresponding quantities at each level.
Note! price jumps : current bid price > previous bid price (within short period of time)

1. Example JuypterNotebook-Rnn
2. "./data" contains sample data from upbit, tick-tick information of L2 orderbook (KRW-ADA)
3. features are from https://github.com/dzitkowskik/StockPredictionRNN/blob/master/docs/project.pdf

Getting Started

from nn import NeuralNetwork
from rnn import RNN

timestep = 50
n_cross_validation = 3
# for order book info only
data = data_prep.get_test_data(timestep, predict_step=5, filename="upbit_l2_orderbook_ADA")

# input_shape <- (timestep, n_features)
# output_shape <- n_classes
nn = NeuralNetwork(RNN(input_shape=data.x.shape[1:], output_dim=data.y.shape[1]), class_weight={0: 1., 1: 1., 2: 1.})

print("TRAIN")
nn.train(data)

print("TEST")
nn.test(data)

print("TRAIN WITH CROSS-VALIDATION")
nn.run_with_cross_validation(data, n_cross_validation)

Prerequisites

keras, tensorflow, sklearn, nuumpy, pandas ...

pip install -r requirements.txt

Impements LSTM / Conv2DLSTM

[LSTM]https://github.com/miroblog/limit_orderbook_prediction/blob/master/rnn.py
[CNN-LSTM] https://github.com/miroblog/limit_orderbook_prediction/blob/master/cnn_lstm.py

Authors

• Lee Hankyol - Initial work - L2-Prediction

License

This project is licensed under the MIT License - see the LICENSE.md file for details