ReasoningAgent

[Hk669]

Why are these changes needed?

• basic working implementation of the ReasoningAgent
• Notebook on how to use the ReasoningAgent
• Nested Chats
• Doc strings and comments
• Tests
• Blog Post

Related issue number

Closes #114

Checks

• I've included any doc changes needed for https://ag2ai.github.io/ag2/. See https://ag2ai.github.io/ag2/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[Hk669]

Ideas for more modular implementation: Break down the search process into a swarm chat:

• ExpandAndGradeAgent uses (async) nested sequential chats:

  • In n_leaf async chats, use the ThinkAgent to expand the current leaf nodes and the GradeAgent to score each;

• SelectionAgent: use a custom reply function to select top nodes to keep; the AFTER_WORK transition is to ExpandAndGradeAgent or terminate, depending on whether the search has finished.

we can have this implemented in the new PR, once this is merged. Thanks!

[qingyun-wu]

@Hk669 could you fix the doc format issue? Thank you! Otherwise, I think it is close to being merged!

[qingyun-wu]

Ideas for more modular implementation: Break down the search process into a swarm chat:

• ExpandAndGradeAgent uses (async) nested sequential chats:

  • In n_leaf async chats, use the ThinkAgent to expand the current leaf nodes and the GradeAgent to score each;

• SelectionAgent: use a custom reply function to select top nodes to keep; the AFTER_WORK transition is to ExpandAndGradeAgent or terminate, depending on whether the search has finished.

we can have this implemented in the new PR, once this is merged. Thanks!

@Hk669 sounds great! Could you open an issue in case we forget later? Thank you!

[Hk669]

Ideas for more modular implementation: Break down the search process into a swarm chat:

• ExpandAndGradeAgent uses (async) nested sequential chats:

  • In n_leaf async chats, use the ThinkAgent to expand the current leaf nodes and the GradeAgent to score each;

• SelectionAgent: use a custom reply function to select top nodes to keep; the AFTER_WORK transition is to ExpandAndGradeAgent or terminate, depending on whether the search has finished.

we can have this implemented in the new PR, once this is merged. Thanks!

@Hk669 sounds great! Could you open an issue in case we forget later? Thank you!

Sure let me do it. Thanks!

[qingyun-wu]

LGTM! Thanks @Hk669 and @BabyCNM!