From 68c0e7631a747d4b27f704f199c6506943f2d264 Mon Sep 17 00:00:00 2001
From: narawat lamaiin <narawat@outlook.com>
Date: Tue, 7 May 2024 21:24:51 +0700
Subject: [PATCH] update

---
 src/interface.jl | 1 +
 src/mcts.jl      | 9 ++++++++-
 2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/src/interface.jl b/src/interface.jl
index 82a1eae..554c0b2 100644
--- a/src/interface.jl
+++ b/src/interface.jl
@@ -440,6 +440,7 @@ function conversation(a::T, userinput::Dict) where {T<:agent}
         :select=> nothing,
         :reward=> 0,
         :isterminal=> false,
+        :feedback=> nothing,
         :thoughtHistory=> OrderedDict{Symbol, Any}( # contain question, thought_1, action_1, observation_1, thought_2, ...
           # :recap=>,
           :question=> userinput[:text],
diff --git a/src/mcts.jl b/src/mcts.jl
index bec27b0..cb9ef9b 100644
--- a/src/mcts.jl
+++ b/src/mcts.jl
@@ -126,7 +126,10 @@ julia>
   [] update docstring
   [] try loop should limit to 3 times. if not succeed, skip
   [] newNodeKey ∉ keys(node.children). New state may have semantic vector close enought to one of existing child state. Which can be assume that they are the same state semantically-wise. 
-  [WORKING] add reflector()
+  [WORKING] store feedback -> state -> agent. 
+    But 1). how should i store state in agent?
+        2). how should I retrieve and use feedback?
+
 
 # Signature
 """
@@ -144,6 +147,10 @@ function expand(a::T1, node::MCTSNode, decisionMaker::Function,
       # add progressValueEstimator
       stateevaluation, statevalue = progressValueEstimator(a, newstate)
 
+      if reward < 0
+        newstate.feedback = stateevaluation
+      end
+
       if newNodeKey ∉ keys(node.children)
         node.children[newNodeKey] = MCTSNode(newNodeKey, newstate, 0, stateevaluation, statevalue, 
                             reward, isterminalstate, node, Dict{String, MCTSNode}())