RNN-Trading-Bot
About
C++ implementation of artificial neural network (ann) that analyzes BTC-e trade data to predict future prices and generate trades accord to those predictions. Uses python3 to executes trades output by the neural network on the BTC-e exchange.
trade_stream.py
⋅⋅⋅Sends HTTP request to BTC-e public API every 2 seconds to gather recent trade data
trading_floor.py
⋅⋅⋅Checks output.sqlite for new records inserted by neural-network-predictor and executes orders to BTC-e if various criteria are met
ann-predictor
⋅⋅⋅Inserts records into output.sqlite containing the current price, predicted price, and type of trade to execute
Neural Network Configuration
| File | Variable | Range | Description |
|---|---|---|---|
| main.cpp | eta_ | 0.0...1.0 | Overall net training rate |
| main.cpp | alpha_ | 0.0...n | Momentum multiplier |
| main.cpp | topology | n/a | Defines number of node per layer |
| training.h | subset_length | n/a | Defines number of records per training set |
Usage
$ mkdir build
$ cd build
$ cmake .. && make
$ ./ann_predictor
$ python3 pyTrader/trade_stream.py
$ python3 pyTrader/trading_floor.py
Notes
- trade_stream.py, trading_floor.py, and ann-predictor all must be run in separate terminal windows
- Execute programs in following order: ann-predictor, trade_stream.py, trading_floor.py
- neural-network-predictor will automatically proceed with execution when subset_length * 10 records are present in input.sqlite
- trading_floor.py will automatically proceed with execution once records are present in output.sqlite