XTX Markets Forecasting Challenge StarterKit

Instructions

The goal of this challenge is to create a program to predict stock movement based on real-market orderbook data. A dataset and some starter code has been provided to help build submissions.

data.csv

• A small sample of the dataset is included in this starter kit. The full dataset data-training.csv can be downloaded from https://challenge.xtxmarkets.com/ under the Data Download section.
• After downloading the dataset, it can be included in the root of the StarterKit directory, but it should not be included in the final submission.
• This file contains some data to be analyzed and trained with the submitted model.
• The goal of the challenge is to predict the y values (the column on the far right), given the current and all past rows.
• The submitted model will be tested on data which is similar to data.csv; however it will not contain the y column.

StarterKit Folder Structure

There are two programming languages that are currently supported for this challenge, Python and R.

In this challenge, There are three relevant folders - one named src containing all the source code needed to process data files and getting your result, another two named either python or r depending on preference of language to be used in the submission. The first step should be to choose a programming language by removing the folder that is unnecessary for the submission. For example, if the language used in a submission is Python, the r folder should be removed.

Quickstart

• Submissions will not be able to access the Internet on our submission environment.

OSX and Linux

• For Python submissions on OSX or Linux see the docs under docs/quickstart_osx_linux/python.md.
• For R submissions on OSX or Linux see the docs under docs/quickstart_osx_linux/r.md.
• We highly recommend using a virtual environment for Python development: docs/quickstart_osx_linux/virtual_env.md.

Windows

• For Python submissions on Windows see the docs under docs/quickstart_windows/python.md.
• For R submissions on Windows see the docs under docs/quickstart_windows/r.md.
• We highly recommend using a virtual environment for Python development: docs/quickstart_windows/virtual_env.md

The src folder

None of these files require changing; however, there are a couple of files in this folder that are quite important to know. Please refrain from trying to run these files manually; there is a script in the python and R folders that runs the model_tester and scorer.

src/model_tester.py

• A prediction vector will be outputted at the value of RESULT_LOCATION.
• Any changes made to this file will not persist on the testing servers.

src/scorer.py

• This program is used to score a prediction vector.
• This program will print the final r2 value to stdout after running.

The Submission Folders

python

This folder should be removed if the solution is written in R.

python/core.py

• This file does not need to be modified
• This file contains the Submission class, all code to interact with src/model_tester.py

python/submission.py

• This file should be used if the solution is written in Python.
• This file contains the starter code for the submission program.
• This extends the Submission class found in core.py.
• The function self.get_next_data() must be used to read a line of data.
• The function self.submit_prediction(pred) must be used to submit a prediction for the y value of the next row of data.
• self.get_next_data() cannot be called two or more times in a row without calling self.submit_prediction(pred).
• Messages should not be printed to stdout because model_tester.py will be looking for predictions from stdout.
• To debug, messages should be printed to stderr.

python/requirements.txt

• Any packages or dependencies necessary for the submission should be added here.
• These will be installed at runtime.

r

This folder should be removed if the solution is written in Python.

r.submission.r

• This file should be used if the solution is written in R.
• This file contains the starter code for the submission program.
• The function get_next_data() must be used to read a line of data.
• The function submit_prediction(pred) must be used to submit a prediction for the y value of the next row of data.
• get_next_data() cannot be called two or more times in a row without calling submit_prediction(pred).
• Messages should not be printed to stdout because model_tester.py will be looking for predictions from stdout.
• To debug, messages should be printed to stderr.

r/requirements.txt

• Any packages or dependencies necessary for the R submission should be added here.
• These will be installed at runtime.

Submission Instructions

For Python submissions

• Follow the steps in docs/submissions/python.md to upload a Python submission.

For R submissions

• Follow the steps in docs/submissions/r.md to upload a R submission.