ML Toys is a collection of machine learning toy problems. These toy problems are intended to facilitate the testing of new machine learning code and algorithms by providing small instances with a variety of challenges. Given an appropriate algorithm, these toy problems can be quickly solved. However, not all toy problems will be solvable with all algorithms.
ML Toys follows the factory method design pattern, and provides MLToyFactory classes for each kind of toy problem. An MLToyFactory object will generate random MLToyInstance objects unless a seed is provided. This allows an algorithm to be tested repeatedly with minimal worry about overfitting, since each fresh MLToyInstance wraps its own set of training and test data.
An example script estimating the training data average is provided.
demo model using training average for all predictions:
ConstantFunctionFactory : min/mean/max = 0.0000/0.0000/0.0000
LinearFunctionFactory : min/mean/max = 0.0016/0.1427/0.3219