Conor O'Sullivan

Why, if we look at the bigger picture, black-box models are not more accurate Conor O’Sullivan · Follow Published in Towards Data Science · 7 min read · 22 hours ago — Photo by Nathan Cima on Unsplash When I started as a data scientist, I was expecting to use state-of-the-art models. XGBoost, Neural Networks. These things are complex and interesting and surely they would drive improvements. Little did I know, the models faced a

Empirical and theoretical evidence for why Figure of Merit (FOM) is the best edge-detection evaluation metric Conor O’Sullivan · Follow Published in Towards Data Science · 12 min read · 6 hours ago — Image segmentation and edge detection are closely related tasks. Take this output from a coastal segmentation model for example: Figure 1: going from segmention mask to edge map (source: author) (dataset: LICS) (CC BY 4.0) The model will classify every pixel

Applying the H-stat with the artemis package and interpreting the pairwise, overall, and unnormalised metrics Conor O’Sullivan · Follow Published in Towards Data Science · 8 min read · 6 hours ago — (source: author) Friedman’s h-statistic (h-stat) provides a powerful window into complex machine learning models. Specifically, it helps us understand if they use interactions to make predictions. We will…

Intuition, algorithm and code for using ALEs to explain machine learning models Conor O’Sullivan · Follow Published in Towards Data Science · 10 min read · 2 hours ago — (source: author) Highly correlated features can wreak havoc on your model interpretations. They violate the assumptions of many XAI methods and make it difficult to understand the nature of a feature’s relationship with the target. At the same…

Witness the evolution of our world with 40 years of Landsat data and minimal manual work Conor O’Sullivan · Follow Published in Towards Data Science · 12 min read · 7 hours ago — (source: chatGPT) Time is a remarkable storyteller. It weaves narratives of change across our planet. Seeming inanimate objects like rivers, forests and coastlines are all malleable given enough time. Even our cities, which don’t seem to…

Create a short list of features using this statistical method (Python Tutorial) Conor O’Sullivan · Follow Published in Towards Data Science · 11 min read · 13 hours ago — In industry, you can have hundreds and even thousands of potential model features in your dataset. And, using dimensionality reduction methods, like PCA, can leave you with features that are hard to explain. Thankfully, feature clustering can help create a short list of features and