Datasets of planGPT

In this repository, we include all the different dataset used to train the models from the paper Learning General Policies for Planning through GPT Models. All domains are randomized, except for Floortile and Visitall, where the grid structure is kept fixed. For these two domains, we later conducted additional experiments by translating the grid along horizontal and vertical axes, achieving coverage results above 90%. During this process, we discovered around 2,000 duplicate problems between the training and testing sets due to generator issue. After correcting the datasets and repeating the experiments, we obtained identical results, confirming the robustness of our findings. After further checks, no other domains showed any overlap between the training and test problems.