Inconsistent definition of a valid solution node

[YanSong97]

Hi there,

Thank you for the excellent work on this project!

While reviewing the code, I noticed an inconsistency in how valid solution nodes are defined within the Reasoning_MCTS_Node class. Specifically, in the is_valid_solution_node() function, a valid solution node includes types SUBQUESTION, DIRECT_ANSWER, or OST_STEP. However, in the extract_solution_from_node() function finding the best solution, the OST_STEP type is excluded.

Is there any reason for this exclusion?

Many thanks

[beishangeyu]

I have same question, and I've noticed that in the is_terminal function, the OST_STEP is excluded as well

[YanSong97]

I have same question, and I've noticed that in the is_terminal function, the OST_STEP is excluded as well

Yeah, seems weird. FYI, we have re-implemented the reasoning part in this PR. After rough testing, I believe adding OST_STEP to is_terminal condition can slightly improve performance.

[beishangeyu]

Many thanks! I will try in my experiment.

[Thunderbeee]

Thanks so much for your contribution and your PR! :D Toward your question why OST_STEP is excluded in the solution extractor, tldr, it is our design choice to let DIRECT_ANSWER (chain of thought) etc to be the final node to answer the question.

The is_valid_solution_node() function is likely used to determine if a node can be considered a valid solution during the search process. This might include nodes that are intermediate steps (OST_STEP) because they contribute to the reasoning process and might be necessary for the overall solution path.

The extract_solution_from_node() function, on the other hand, is used to extract the final solution from a node. In this context, OST_STEP nodes might not directly provide a final answer or solution, which is why they are excluded.