plancraft 0.3.9__py3-none-any.whl → 0.3.11__py3-none-any.whl

plancraft/evaluator.py

@@ -198,9 +198,7 @@ class Evaluator:
         action = None
 
         # run episode until stuck or until max steps is reached
-        while (
-            not self.history.check_stuck() and self.history.num_steps < self.max_steps
-        ):
+        while self.history.num_steps < self.max_steps:
             # if the action is stop then we end the episode
             if isinstance(action, StopAction):
                 # if the action is stop and task is impossible then success

plancraft/utils.py

@@ -44,8 +44,6 @@ class History:
         self.resolution = resolution  # low, medium, high
 
         self.inventory_history = []
-        self.inventory_counters = []
-
         self.tokens_used = 0
 
         # use system prompt if provided
@@ -105,14 +103,6 @@ class History:
 
     def add_inventory_to_history(self, inventory: dict):
         self.inventory_history.append(inventory)
-        # count inventory
-        counter = Counter()
-        for slot, item in inventory.items():
-            # ignore slot 0
-            if slot == 0:
-                continue
-            counter[item["type"]] += item["quantity"]
-        self.inventory_counters.append(counter)
 
     def add_image_to_history(self, image):
         self.images.append(image)
@@ -121,11 +111,6 @@ class History:
         if observation is None:
             return
         if "inventory" in observation:
-            # clean_inv = []
-            # remove empty slots
-            # for slot, item in observation["inventory"].items():
-            #     if item["quantity"] > 0:
-            #         clean_inv.append(item)
             self.add_inventory_to_history(observation["inventory"])
         if "image" in observation:
             self.add_image_to_history(observation["image"])
@@ -140,7 +125,6 @@ class History:
         self.initial_dialogue_length = len(self.dialogue_history)
 
         self.inventory_history = []
-        self.inventory_counters = []
 
         self.tokens_used = 0
 
@@ -157,21 +141,6 @@ class History:
     def num_steps(self):
         return (len(self.dialogue_history) - self.initial_dialogue_length) // 2
 
-    def check_stuck(self, max_steps_no_change: int = 10) -> bool:
-        """
-        If inventory content does not change for max_steps_no_change steps
-        the agent is considered stuck.
-
-        With N=10, the oracle solver can still solve 100% of the examples
-        """
-        if len(self.inventory_counters) <= max_steps_no_change:
-            return False
-
-        return all(
-            c == self.inventory_counters[-max_steps_no_change - 1]
-            for c in self.inventory_counters[-max_steps_no_change - 1 :]
-        )
-
 
 def get_downloaded_models() -> dict:
     """

{plancraft-0.3.9.dist-info → plancraft-0.3.11.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: plancraft
-Version: 0.3.9
+Version: 0.3.11
 Summary: Plancraft: an evaluation dataset for planning with LLM agents
 License: MIT License
         

{plancraft-0.3.9.dist-info → plancraft-0.3.11.dist-info}/RECORD

@@ -1,8 +1,8 @@
 plancraft/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 plancraft/config.py,sha256=ShsFRlJ7plsl3ToD9fiO_4LDQuXdbjNV6Xp6o3Yk2Yg,4315
-plancraft/evaluator.py,sha256=FogXAjnldK2K1gD02f2WW1YWvjWokKGUOJxwtwzGNjk,10804
+plancraft/evaluator.py,sha256=vu8RqAsvoDtUizLpiDA9w1fmCdCb6q91DUuE_4mUhUo,10745
 plancraft/generate_dataset.py,sha256=DlrU-PmvWqSNJD1g1-8Lpb8n3N-Ogw3rje1nrRzjGKs,2382
-plancraft/utils.py,sha256=phaHzbIS85YZrBPaGG9TStHY8ZBKR1LKfuN1exfVy1U,6889
+plancraft/utils.py,sha256=0Uq-3VE-bTRstalzKknBJ-ExWf8ec_Jrg4QNEk8bJ-o,5778
 plancraft/data/test.json,sha256=4jWfYMAVuZCFmGB4iZJAjlh9_8jXECdaGp8xn7_tAM4,1317131
 plancraft/data/test.small.easy.json,sha256=5NZEJ2PqIgmHQecJOIVQyM1D6GFKyJq7GVmgRudaqQk,189304
 plancraft/data/test.small.json,sha256=eULAG1rdolRMXPrecV-7YoDIheKGyIT5MVpWdISV0wg,270089
@@ -1920,7 +1920,7 @@ plancraft/models/generators.py,sha256=F76_iPiqxUjDIrQwF58tzM0bLM91OkZJ0sBqBuki5w
 plancraft/models/oracle.py,sha256=jDCE6zVFvbwFpDzQZTkHIlRwMud1yMJ4LVIdfpt5ddU,8449
 plancraft/models/utils.py,sha256=E-sZohvolWgGbpHQKgAgkgIfUJoVnT5pMt6JP8xLHKg,4034
 plancraft/train/dataset.py,sha256=oFqEd4LG9oEQ-71teh0Wf7-jJbtybT2ZibfM2bBdBkM,5474
-plancraft-0.3.9.dist-info/METADATA,sha256=zXrxilbtrVK05xX0jjFbDAub_tToabPX0wX1TULyUjA,11147
-plancraft-0.3.9.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-plancraft-0.3.9.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
-plancraft-0.3.9.dist-info/RECORD,,
+plancraft-0.3.11.dist-info/METADATA,sha256=wgvEebVv8N2uL51t9oRGEgkniXkbUCZbshDZPY7kRIo,11148
+plancraft-0.3.11.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+plancraft-0.3.11.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
+plancraft-0.3.11.dist-info/RECORD,,