Summary:Talk:DrussGT/Understanding DrussGT/Reason behind using Manhattan distance

Distance Function Comparison in ML

The conversation is about the use of different distance functions in machine learning, specifically in the context of a gun for a video game. The participants discuss the use of Euclidean distance, Manhattan distance, and a log-based distance function. They mention that Manhattan distance performed better in the context of the game, and suggest that this may be due to the noise rejection properties of Manhattan distance and the curse of dimensionality in high-dimensional spaces. They also discuss the similarities between the L1 and L2 distance functions and the L1 and L2 norms in logistic regression and the idea of pattern matching with L1 norm.

— ChatGPT