Table of Contents

• Introduction
• The Task:
• My Submission

Introduction

This was the take-home task that got me into Harrison.ai. My goal was to showcase how I would go about developing a model and explain my experimental reasoning. Below is the original notebook I submitted, exactly as it was. For context, this was done in November, 2019; EffientNetV1 had just come out and was on top of the leaderboards, UNet++ came out the year before. Vision transformers hadn't yet come out and impacted the computer vision world.

The Task:

Kaggle Competition: https://www.kaggle.com/competitions/siim-acr-pneumothorax-segmentation

Please perform the following technical task and present your findings, e.g., in a Jupyter notebook

1. Access and download the dataset from the Kaggle competition for Pneunomthorax detection: https://www.kaggle.com/c/siim-acr-pneumothorax-segmentation/overview
2. Perform appropriate exploratory data analysis and visualisation
3. Build a best-effort algorithm to detect a visual signal for Pneunomthorax as per the competition evaluation metric
4. Generate appropriate graphs / results table to assess model performance
5. Make a baseline submission to the late submission pool
6. Discuss shortcomings and the improvements you would make to the dataset, evaluation metrics and algorithm

[7. Discuss the task together in the next week.]

My Submission

Trouble viewing the embed? Open the notebook in a new tab.