update

src/interface.jl

@@ -107,17 +107,23 @@ function decisionMaker(a::T1, state::T2)::String where {T1<:agent, T2<:AbstractD
     "Thought_2": "reasoning",
     ...
     "Thought_n": "reasoning",
-    "Action_1": "action to take",
+    "Action_1": {"action": "action to take", "input": "Action input"},
     "Observation_1": "result of the action"
   }
 
   Here are some examples:
   {
-    "Question": "I would like to buy a sedan",
+    "Question": "I would like to buy a sedan.",
-    "Thought_1": "I have many car in my inventory suitable for several usage scenarios",
+    "Thought_1": "I have many cars in my inventory suitable for several usage scenarios.",
-    "Thought_2": "It would be better if I know what the user intend to do with his car",
+    "Thought_2": "It would be better if I knew what the user intends to do with his car."1,
     "Thought_3": "I will ask the user what is the intended usecase",
-    "Action_1": "Chatbox[What will you use it for?]"
+    "Action_1": {"action": "Chatbox", "input": "What will you use it for?"}
+  }
+  {
+    "Question": "I'm looking for a sedan.",
+    "Thought_1": "I have many types of sedans in my inventory, each with diverse features.",
+    "Thought_2": "It would be easier to make a recommendation if I knew what feature the user is looking for. I should ask the user.",
+    "Action_1": {"action": "Chatbox", "input": "Do you have any specific feature in mind?"}
   }
 
   $reflect
@@ -307,7 +313,7 @@ function conversation(a::T, userinput::Dict) where {T<:agent}
 
   
     
-    else    #[WORKING] new thinking 
+    else    #[PENDING] new thinking 
       initialState = Dict(
 
         # deepcopy the info to prevent modifying the info unintentionally during MCTS planning

src/mcts.jl

@@ -121,15 +121,28 @@ function expand(a::T1, node::MCTSNode, state::T2, decisionMaker::Function, state
   for sample in 1:n
     thoughtJstr = decisionMaker(a, state)
     thoughtDict = copy(JSON3.read(thoughtJstr)) 
-    latestActionKey = GeneralUtils.findHighestIndexKey(thoughtDict, "Action")
     
-    error("--> expand")  
+    """ Example of thoughtDict
-    newState = transition(node.state, action)  #[] Implement your transition function
+      Dict{Symbol, Any} with 3 entries:
+      :Thought_1     => "The customer wants to buy a bottle of wine. This is a good start!"
+      :Action_1      => Dict{Symbol, Any}(
+                          :action=>"Chatbox", 
+                          :input=>"What occasion are you buying the wine for?"
+                          )
+      :Observation_1 => ""
+    """
+    latestActionKey = GeneralUtils.findHighestIndexKey(thoughtDict, "Action")
+    _action = thoughtDict[latestActionKey]
+    action = _action[:action]
+    actioninput = _action[:input]
+
+    newState = transition(a, node.state, action, actioninput)  #[] Implement your transition function
+    
     if newState ∉ keys(node.children)
         node.children[newState] = MCTSNode(newState, 0, 0.0, Dict{T, MCTSNode}())
     end
   end
-  
+  error("--> expand")  
 end
 
 """
@@ -192,9 +205,17 @@ function backpropagate(node::MCTSNode, reward::Float64)
     end
 end
 
-"""
+""" Get a new state
 
 # Arguments
+  - `a::T1`
+    one of YiemAgent's agent
+  - `state::T2`
+    current game state
+  - `action::String`
+    name of LLM's function to be used
+  - `actioninput::String`
+    input to LLM function
   
 # Return
 
@@ -205,12 +226,14 @@ julia>
 
 # TODO
   - [] update docstring
-  - [] implement the function
+  - [WORKING] implement the function
 
 # Signature
 """
-function transition(state, action)
+function transition(a::T1, state::T2, action::T3, 
+          actioninput::T3) where {T1<:agent, T2<:AbstractDict, T3<:AbstractString}
 
+  error("--> transition")
 end
 
 """