A library for meta-mapping (https://arxiv.org/abs/2005.04318).
Meta-mapping provides a new perspective on zero-shot task performance, by framing it as a problem of transforming a prior task.
HoMM provides a parsimonious homoiconic implementation of this framework.
See the examples and paper for further details. The application to the domains from the paper can be found as follows:
Polynomial domain experiments: code in the examples folder and at https://github.com/lampinen/HoMM_polynomial_analysis
Card games domain: https://github.com/lampinen/HoMM_cards
RL domain: https://github.com/lampinen/homm_grids
Visual concepts domain: https://github.com/lampinen/categorization_HoMM
This code has generally been tested with CUDA 10, Tensorflow 1.14, and Python 3.6.