AI Challenge

Welcome! you are a super star for making it here. This is your time to shine, an opportunity to show off your skills, understanding and more importantly coding abilities 😉. So relax, grab some coffee / whiskey (depending on time of day) and start developing on this take-home exercise.

We recommend that you implement deep learning models, using PyTorch exclusively for any models you develop

Setup 💿

1. Clone this repository to your local machine.
2. Install the required Python libraries:

pip install -r requirements.txt

Please feel free to install any other repositories of your choice. For any deep learning implementations, please use PyTorch

Challenge Overview 💪

The challenge is split into two parts, where first part involves the implementation of a model, training it on the data and testing it. Second part involves reporting and analysis, we recommend you create a Jupyter Notebook to showcase your analysis and interpretation of your model performance. The following sections go into greater detail for each of the parts.

Part 1: Time Series Forecasting

In this part you will:

• Implement a time series forecasting model with a transformer backbone and any prediction head design of your choice. (Implemented yourself, do not use externally implemented models)
• Train the model on the provided dataset.
• Test the model to evaluate its performance.
• Benchmark your model to predict future trend in different time-windows, Select a time window that performs the best.

---

Datasets

There are 2 datasets we have selected for this exercise, feel free to explore and choose the one that you'd like to use.

1. Yahoo finance stock data: As the name suggests this is a financial time series dataset. To download it a python script has already been written for you. You can download the data via the following:

python src/utils/download_yfinance_data.py AAPL

This will download the stock data for Apple Inc. (AAPL) from 5 days ago to yesterday in 5-minute intervals, and save it to a CSV file named AAPL_stock_data.csv. You can also specify custom dates if needed as shown by the sample command below. Though minimum, 60 days length in 5 minute intervals is what we recommend.

python src/utils/download_yfinance_data.py AAPL --start_date 2024-05-17 --end_date 2024-05-22 --interval 5m

2. CO2 Concentration Estimation: This task utilized a publicly available dataset from Imperial College’s Carbon Capture Pilot Plant, accessible through the orginal repositry or our Google Drive. This task requires estimating the CO2 concentration at six distinct sampling points in the absorber. For more details on the dataset and preprocessing steps, please refer to the official notebook. If you choose to proceed with this task, we strongly recommend using 140207_1.xlsx as the test dataset, while the remaining files serve as the training dataset. The related paper is also included in our Goolge Drive, please feel free to read it for additional background information.

---

Part 2: Interpretability & Reporting

In this part, you will:

• Analyse the performance of your model.
• Interpret the results and provide insights.
• Document your analysis and findings in a Jupyter Notebook, including visualizations and detailed explanations.

What we'd like to see 🙀

We'd like to see:

• Effective use of deep learning models, innovative learning strategies and monitoring techniques, implemented exclusively in PyTorch.
• Clear and concise code, with appropriate comments and documentation. Please share your results in a github repository with instructions on how to run and reproduce your results.
• Analysis and interpretation of your model's performance.
• Insightful visualisations that help explain your results and findings.

What you'd be assessed on 🔎

You will be assessed on:

• Model Implementation: How well you implement the time series forecasting model using PyTorch.
• Performance: The accuracy and reliability of your model on the test data.
• Analysis: The depth and clarity of your analysis in the Jupyter Notebook.
• Interpretability: How well you interpret and explain the results of your model.
• Code Quality: The readability, organization, and documentation of your code.
• Visualizations: The effectiveness of your visualizations in conveying your findings.

Good luck, and we look forward to seeing your work!

Additional support 🤝

If you have any questions, or need further clarification of any of the challenges.

Then please reach out to: 📣 [email protected] 📣