swMATH ID: 42233
Software Authors: Alexandre Abraham, Léo Dreyfus-Schmidt
Description: Cardinal, a metric-based Active learning framework. In Active learning, a trained model is used to select samples to label to maximize its performance. Choosing the best sample selection strategy for a one-shot experiment is hard, but metrics have been proven to help by detecting strategies performing worse than random or detecting and avoiding noisy samples. Cardinal is a python framework that assists the practitioner in selecting a strategy using metrics and the researcher in developing those metrics. Cardinal caches experiments to compute insights costlessly, keeps track of logged metrics, and proposes extensive documentation. It also interfaces with other packages to use state-of-the-art strategies
Homepage: https://www.sciencedirect.com/science/article/pii/S2665963822000173
Source Code:  https://github.com/dataiku-research/cardinal
Dependencies: Python
Related Software: modAL; PyTorch Lightning; ImageNet; SuperGLUE; UCI-ml; Scikit; libact; ALiPy; scikit-activeml; JCLAL; skorch; BatchBALD; GPyTorch; Flax; JAX; Keras; PyTorch; Python; PyRelationAL
Cited in: 0 Publications