benchmarks/CLRS/scripts/research_problem.txt

Improve the baseline model performance on the task floyd_warshall in The CLRS Algorithmic Reasoning Benchmark. The dataset description is available in data_description.txt, and the baseline model architecture description is available in baseline_model_description.txt. To run the baseline model, execute train.py. Note that the core message passing function of the baseline model is implemented in function get_triplet_msgs (L301 in processors.py). You can modify this function to improve the baseline model performance. You can also modify other parts of the baseline model and training script to improve its performance, as long as the final model is still loadable by calling BaselineModel class as in L415 in train.py.

When you submit your final answer, you will be evaluated on the performance of the checkpoint checkpoints/best.pkl saved by train.py. Note that the final model must still be loadable by calling BaselineModel class as in L415 in train.py and with the saved spec_list.pkl and model_params.pkl.