check lats

[TataKKKL]

• pass different number of simulations, number of iterations to LATS[todo]
• visit the same node for LATS, is current implementation working?[todo]

[TataKKKL]

just use websocket testing is enough

uvicorn app.main:app --host 0.0.0.0 --port 3000
python test/test-tree-search-ws-lats.py 

[H-Freax]

#78

[TataKKKL]

1. Done
   2 TODO, and need to check other implementation

[TataKKKL]

increase iterations
multiple rollout, set num_simulations > 1, need to change self.simulation and self.backpropagate implementation?

[TataKKKL]

relax score requirement?
https://github.com/PathOnAI/VisualTreeSearch-Demo/blob/main/visual-tree-search-backend/app/api/lwats/agents_async/SearchAgents/lats_agent.py#L79

[TataKKKL]

https://github.com/PathOnAI/LWATS/issues/149

[TataKKKL]

check how reward, value are updated for each node at different step

[TataKKKL]

I suspect there is an issue: after expanding nodes, I evaluate new children and use that score as both the child's value and reward. But I am not sure if I need to keep reward; maybe value is enough.
During backpropagation, I'm only updating node.value. As iterations increase and the algorithm selects the same nodes repeatedly, I'm unsure whether to re-expand nodes and how to properly handle value updates between new and existing children.

[TataKKKL]

Do we need to keep best_child? if so, do we want to use node reward? maybe node value is enough.