Practical LSTM Time Series Prediction for Forex with TensorFlow and Algorithmic Bot

This is the companion code to Pragmatic LSTM for a Forex Time Series. So, if you want to understand the intention of the code, I highly recommend reading the article series first.

Setting Up The Environment

The model training and prediction have been tested on both Ubuntu Linux 20.04 and Windows 10 and both work as expected. To prepare your machine to run the code, follow these steps:

Install Conda or update your Conda installation to the latest
Make sure you have the latest Nvidia driver if you are planning to use the GPU. On Windows, the latest version of Nvidia driver was failing on some machines and the solution was to revert back to version 431.86, so keep this in mind. The details of installing TensorFlow can be found here: Official TensorFlow GPU support for Nvidia
Execute the next lines on your console:

conda create --name tf python==3.8.* ipython jupyter pandas numpy scikit-learn matplotlib flask
conda activate tf

I chose to install TensorFlow 2.3 as it has some solved bugs that hit me with earlier versions. Unfortunately, this version was not available from Conda which had TensorFlow 2.1, so I had to use pip:

pip install tensorflow-gpu==2.3

N.B. Keras became part of TensorFlow from v2, no need to install it separately.

Files Structure

│ LICENSE                       # The code licence 
│ README.md                     # This Readme file
|
├─LSTM-FX-CTrader-Client        # Directory of the client code
│     MLBot.cs                  # Sample bot C# code
│     README.md                 # Read me for the CTrader Client
|
├─LSTM-FX-Prediction-Server     # Directory of the Web Server
│ │   main.py                   # The web server loading file
│ │   README.md                 # Read me for the Prediction server
│ │
│ └───Models                    # Directory of the ML Models 
│     │   gbpusd-32-256-14.bin  # Sample MinMax Scaler
│     │
│     └───gbpusd-32-256-14      # Directory of Sample LSTM Model 
|
└─LSTM-FX-Train-Test            # Directory of the Training and Testing of Model
  │   common_variables.py       # Configures your model before training and testing
  │   multi_pred_model.ipynb    # Multiple predictions of more than one unit
  │   prep_and_split.ipynb      # Prepares your data and splits it into multiple CSVs
  │   README.md                 # Read me for the training and testing
  │   test_model.ipynb          # Tests your trained model
  │   time_series.py            # Utility reusable functions 
  │   train_model.ipynb         # Trains your model
  │
  ├───data                      # storage direcotry for the CSV
  │
  ├───models                    # Contains the trained models
  │   └───gbpusd-32-256-14      # Sample trained model
  │
  └───scalers                   # Contains the scalers used in the model training
          gbpusd-32-256-14.bin  # Sample scaler associated with the model

Tools

I used Juypter Notebook from within Visual Studio Code and I executed everything using Visual Studio Code for Windows and for Linux.

Hardware Specifications

It might help to know what hardware I used:

Laptop (for development): Dell Precision M4800, 32GB RAM, 8 Logical Processors Intel i7 2.9GHz, 2GB RAM Nvidia Quadro K2100M
Server (for running): Dell Precision Tower 7910, 24GB RAM, 28 Logical Processors Intel Xeon 2.6GHz, Nvidia GeForce RTX 2080 8GB RAM

Training and Testing Your Model

LSTM-FX-Train-Test

Backtesting Your Trading Strategy

LSTM-FX-Prediction-Server

Disclaimer

These stories are meant as a research on the capabilities of deep learning and are not meant to provide any financial or trading advice. Do not use this research and/or code with real money.

Licence

This repository and the code are licenced under the MIT licence, please check the licence before attempting to use the code.

Adriaan1 / LSTM-FX