bioeconometRician
In this post I describe how to calculate the first and second moments of a loss function for a machine learning (ML) model. The expected value of a loss function is known as its risk, and I refer to the second central moment as the loss variance. While these quantities are not perfectly knowable in the real world, in the simulation setting the researcher has full knowledge of how data is generated…
Most machine learning models are point predictors: they output a single number (a regression estimate or a most-likely class label). But deploying a model in practice almost always requires understanding how uncertain that prediction is. Heuristic measures of uncertainty that come out of a ML model like the posterior variance from a GPR, the variance of a prediction from dropout, or a classifier’s…
This post describes how to adjust the traditional sample standard deviation (SD) estimator to reduce its bias. The SD or a random variable, like the variance, is a population parameters, and is usually denoted \(\sigma\) and \(\sigma^2\), respectively. While the well-known sample variance estimator provides an unbiased estimate of its population parameter, the equivalent sample SD estimator does…
This post accompanies the release of the sntn package, which implements a scipy-like class for doing inference on a sum of a normal and a trunctated normal (SNTN) distribution (see Kim (2006) and Arnold (1993)). I’ve written about this distribution before.
Executive summary
Executive summary
Executive summary
This posts outlines the paranet package, which allows for the fitting of elastic net regularized parametric survival models with right-censored time-to-event data in python. I became interested in this topic when I realized that the a parametric modelling excercise would be useful for a business case, but was unable to find any package support in python. Currently, paranet supports three…
This post summarizes a newly released python package: trialML. This package is designed to help researchers and practitioners prepare their machine learning models for a statistical trial to establish a lower-bound on model performance. Specifically, this package helps to calibrating the operating threshold of a binary classifier and carry out a power analysis. A more formal description of these…
This post summarizes a newly released python package: SurvSet the first ever open-source time-to-event dataset repository. The goal of SurvSet is to allow researchers and practioneeres to benchmark machine learning models and assess statistical methods. All datasets in this repository are consisently formatted to enable rapid prototyping and inference. The origins of this dataset were for testing…