This repository includes the code and data used in tLSTD (IJCAI, 2016) and ATD papers (AAAI, 2017).
The atdeasycode is easy to understand but is not optimized in many aspects (i.e. frequently allocate and free memeory for gsl variables). This code is written in a straightforward way. For example, in comp_algo_domain.c file, there is no complex structure: we define necessary variables used to read data, set up parameters to sweep, initialize variables for each algorithm, and then use for loop to sweep over all possible parameter combinations. The ATD and tLSTD algorithms in algorithms.c file are written in a manner which almost exactly follow the pseudocode in our paper. It is easy to adapt any single algorithm's implementation to your own repository.
The atdoptcode is our most recently updated version. It is very well structured and written in OOD style even though we use C. It includes implementation of random MDP and boyan chain domains, and also allows users to read in datasets generated by themselves. It currently does not support sparse computation, but it is still efficient if your representation is not too large.
For more details please check readme files under the above two directories.