The code is organized as follows:
main.py file contains the key algorithm: Which includes the decoder, the learnable task embeddings, and the training procedure. Inside the problems directory the sample problems can be found. All the experiments in the figure were generated using the problems/toy_problem.py scenario.
To run the algorithm invoke:
python main.py --problem toy_problem
Refer to python main.py -h for the parameters and explanations.
The file exp_generalization.py contains the code for generalization experiments. Finally ops directory contains all the improvement operators,
of which only SGLD is currently implemented, SGLD is also the default improvement operator used by main.py.