A dissertation project looking to use machine learning and Japanese candlestick data to predict trends in forex markets.
Ryan Easter EAS16635772
This project is made up of two section:
- Object Detector for Candlestick Patterns
- Price Prediciton Scripts
- OpenCV
- Ubuntu
- GPU support (optional)
The object detector can be found on GitHub by the user AlexeyAB at this link: https://github.com/AlexeyAB/darknet Within the Readme of the project is a detailed set of configurations however, another copy can be found below.
This project requires an Ubuntu OS for easiest installation.
- Clone the darknet repo into the top level of the Project-Soros repo
- Run the make command in the darknet folder (List of configurable parameters: https://github.com/AlexeyAB/darknet#how-to-compile-on-linux-using-make)
- Download the weights from: https://github.com/DeputyShrute/Project-Soros/blob/main/darknet%20requirements/yolo-obj_last.weights
- Copy the downloaded weights file to the
backup
folder inside thedarknet
folder - Copy the
yolo-obj.cfg
from thedarknet_requirements
folder to the cfg folder insidedarknet
- Copy the
obj.data
,obj.names
andtest.txt
from thedarknet_requirements
folder to the data folder insidedarknet
- Copy the
test_img
folder into thedarknet
folder - Run the following command in the root of the
darknet
folder:./darknet detector test data/obj.data cfg/yolo-obj.cfg backup/yolo-obj_last.weights
- When the prompt for an image display, use the file location of the images from the
test_img
folder. - Depending if a GPU has been configured the output will either display in the screen or in the
predictions.png
in the root of the darknet folder.
This is basic setup for the object detector. If a GPU has been utilised in the makerfile
the detections will run faster.
All of the required scripts are within the repo and before they can be run, the requirements.txt
needs to be installed:
pip install -r requirements.txt
The software can then be started using by running:
python3 run.py
This is in the root of the Project-Soros Folder.
On the menu two options will appear:
Update all data
pulls fresh data from YahooFull Analysis
runs both the object detection and price prediction