niacin: A Python package for text data enrichment
Dillon Niederhut · The Journal of Open Source Software · 2020
A common component of frameworks for building robust and accurate learning models is a utility for performing sets of perturbations on the input data.In machine vision, for example, rotating, cropping, stretching, warping, and flipping training images have all been shown to increase model accuracy, and reduce certain kinds of overfitting (Chen, Dobriban, & Lee, 2019).Popular machine vision libraries like PyTorch, Tensorflow, and FastAI all have built-in utilities for performing these transformations (Abadi et al., 2015;Howard & Gugger, 2020;Paszke et al., 2019).