error-quantification

AriChow/error_propagation

This project is for understanding and quantifying the errors in a machine learning or data analytic pipeline. Two approaches are explored. The first is using freezing and unfreezing of pipeline components (using optimization techniques like grid-search, random-search, Bayesian Optimization, Genetic Algorithms etc.). The second is using a gradient based approach to quantify the gradients of the expected error w.r.t the algorithms and hyperparameters.