kaggle-environments 1.14.9__py2.py3-none-any.whl → 1.14.11__py2.py3-none-any.whl

kaggle_environments/__init__.py

@@ -21,7 +21,7 @@ from .main import http_request
 from . import errors
 from . import utils
 
-__version__ = "1.14.9"
+__version__ = "1.14.11"
 
 __all__ = ["Agent", "environments", "errors", "evaluate", "http_request",
            "make", "register", "utils", "__version__",

kaggle_environments/envs/llm_20_questions/llm_20_questions.py

@@ -8,7 +8,7 @@ import torch
 from .keywords import KEYWORDS_JSON
 from os import path
 from pathlib import Path
-from random import choice, random
+from random import choice
 from string import Template
 from transformers import T5Tokenizer, T5ForConditionalGeneration
 
@@ -29,9 +29,7 @@ TIMEOUT = "TIMEOUT"
 GUESS = "guess"
 ASK = "ask"
 GUESSER = "guesser"
-RANDOM_GUESSER = "random_guesser"
 ANSWERER = "answerer"
-RANDOM_ANSWERER = "random_answerer"
 
 keywords_list = json.loads(KEYWORDS_JSON)
 keyword_cat = random.choice(keywords_list)
@@ -70,19 +68,6 @@ def guesser_agent(obs):
     
     return call_llm(prompt)
 
-def random_guesser(obs):
-    if obs.turnType == GUESS:
-        return "banana"
-    if random() < .5:
-        return "Is is a person?"
-    if random() < .5:
-        return "Is it a place?"
-    return "Is it a thing?"
-
-def random_answerer():
-    if random() > .5:
-        return "yes"
-    return "no"
     
 
 def answerer_agent(obs):
@@ -99,7 +84,7 @@ def answerer_agent(obs):
         return ""
 
 
-agents = {GUESSER: guesser_agent, ANSWERER: answerer_agent, RANDOM_ANSWERER: random_answerer, RANDOM_GUESSER: random_guesser}
+agents = {GUESSER: guesser_agent, ANSWERER: answerer_agent}
 
 def guesser_action(active, inactive, step):
     inactive.observation.keyword = keyword
@@ -118,40 +103,43 @@ def guesser_action(active, inactive, step):
     if active.action and keyword_guessed(active.action):
         guessed = True
         score = 20 - int(step / 3)
-        end_game(active, inactive, score, DONE, DONE)
+        end_game(active, score, DONE)
+        end_game(inactive, score, DONE)
     return guessed
 
-def end_game(active, inactive, reward, status, inactive_status):
-    active.observation.keyword = keyword
-    active.observation.category = category
-    inactive.observation.keyword = keyword
-    inactive.observation.category = category
-    active.reward = reward
-    inactive.reward = reward
-    active.status = status
-    inactive.status = inactive_status
-
+def end_game(agent, reward, status):
+    agent.observation.keyword = keyword
+    agent.observation.category = category
+    agent.reward = reward
+    agent.status = status
 
 def answerer_action(active, inactive):
     active.observation.keyword = keyword
     active.observation.category = category
     response = active.action
+    bad_response = False
     if not response:
         response = "none"
-        end_game(active, inactive, -1, ERROR, DONE)
+        end_game(active, -1, ERROR)
+        end_game(inactive, 1, DONE)
+        bad_response = True
     elif "yes" in response.lower():
         response = "yes"
     elif "no" in response.lower():
         response = "no"
     else:
         response = "maybe"
-        end_game(active, inactive, -1, ERROR, DONE)
+        end_game(active, -1, ERROR)
+        end_game(inactive, 1, DONE)
+        bad_response = True
     active.observation.answers.append(response)
     inactive.observation.answers.append(response)
+    return bad_response
 
 def increment_turn(active, inactive, step, guessed):
     if step == 59 and not guessed:
-        end_game(active, inactive, -1, DONE, DONE)
+        end_game(active, -1, DONE)
+        end_game(inactive, -1, DONE)
     elif active.observation.turnType == "guess":
         active.observation.turnType = "ask"
     elif active.observation.turnType == "ask":
@@ -186,42 +174,47 @@ def interpreter(state, env):
     end_early = (active1 and active1.status) in (TIMEOUT, ERROR) or (active2 and active2.status in (TIMEOUT, ERROR))
     one_guessed = False
     two_guessed = False
+    one_bad_response = False
+    two_bad_response = False
 
-    if active1 is not None:
-        guessed = False
-        if active1.observation.role == GUESSER:
-            guessed = guesser_action(active1, inactive1, step)
-            one_guessed = guessed
-        else:
-            answerer_action(active1, inactive1)
+    if active1 is None or active2 is None:
+        raise ValueError
 
-        if active1.status in (TIMEOUT, ERROR):
-            end_game(active1, inactive1, 0, active1.status, DONE)
-        elif end_early:
-            end_game(active1, inactive1, 0, DONE, DONE)
-        else:
-            increment_turn(active1, inactive1, step, guessed)
-    
-    if active2 is not None:
-        guessed = False
-        if active2.observation.role == GUESSER:
-            guessed = guesser_action(active2, inactive2, step)
-            two_guessed = guessed
-        else:
-            answerer_action(active2, inactive2)
+    if active1.observation.role == GUESSER:
+        one_guessed = guesser_action(active1, inactive1, step)
+    else:
+        one_bad_response = answerer_action(active1, inactive1)
 
-        if active2.status in (TIMEOUT, ERROR):
-            end_game(active2, inactive2, 0, active2.status, DONE)
-        elif end_early:
-            end_game(active2, inactive2, 0, DONE, DONE)
-        else:
-            increment_turn(active2, inactive2, step, guessed)
+    if active2.observation.role == GUESSER:
+        two_guessed = guesser_action(active2, inactive2, step)
+    else:
+        two_bad_response = answerer_action(active2, inactive2)
+
+    if active1.status in (TIMEOUT, ERROR) or one_bad_response:
+        end_game(active1, -1, active1.status)
+        end_game(inactive1, 1, DONE)
+    elif end_early or two_bad_response:
+        end_game(active1, 1, DONE)
+        end_game(inactive1, 1, DONE)
+    else:
+        increment_turn(active1, inactive1, step, one_guessed)
+    
+    if active2.status in (TIMEOUT, ERROR) or two_bad_response:
+        end_game(active2, -1, active2.status)
+        end_game(inactive2, 1, DONE)
+    elif end_early or one_bad_response:
+        end_game(active2, 1, DONE)
+        end_game(inactive2, 1, DONE)
+    else:
+        increment_turn(active2, inactive2, step, two_guessed)
     
     # make sure to end the game if only one team guessed correctly this round
     if one_guessed and not two_guessed:
-        end_game(active2, inactive2, 0, DONE, DONE)
+        end_game(active2, 0, DONE)
+        end_game(inactive2, 0, DONE)
     elif two_guessed and not one_guessed:
-        end_game(active1, inactive1, 0, DONE, DONE)
+        end_game(active1, 0, DONE)
+        end_game(inactive1, 0, DONE)
     
     return state
 

kaggle_environments/envs/llm_20_questions/test_llm_20_questions.py

@@ -0,0 +1,53 @@
+
+from kaggle_environments import make
+
+def custom_questioner(obs):
+    if obs.turnType == "guess":
+        return "banana"
+    return "Is it a banana?"
+
+def custom_answerer():
+    return "no"
+
+def bad_answerer():
+    return "maybe?"
+
+def error_agent():
+    raise ValueError
+
+def test_llm_20_q_completes():
+    env = make("llm_20_questions", debug=True)
+    env.run([custom_questioner, custom_answerer, custom_questioner, custom_answerer])
+    json = env.toJSON()
+    assert json["name"] == "llm_20_questions"
+    assert json["statuses"] == ["DONE", "DONE", "DONE", "DONE"]
+
+def test_llm_20_q_errors_on_bad_answer():
+    env = make("llm_20_questions", debug=True)
+    env.run([custom_questioner, custom_answerer, custom_questioner, bad_answerer])
+    json = env.toJSON()
+    assert json["name"] == "llm_20_questions"
+    assert json["rewards"] == [1, 1, 1, None]
+    assert json["statuses"] == ["DONE", "DONE", "DONE", "ERROR"]
+    print(len(json["steps"]))
+    assert len(json["steps"]) == 3
+
+def test_llm_20_q_errors_on_error_answer():
+    env = make("llm_20_questions", debug=True)
+    env.run([custom_questioner, custom_answerer, custom_questioner, error_agent])
+    json = env.toJSON()
+    assert json["name"] == "llm_20_questions"
+    assert json["rewards"] == [1, 1, 1, None]
+    assert json["statuses"] == ["DONE", "DONE", "DONE", "ERROR"]
+    print(len(json["steps"]))
+    assert len(json["steps"]) == 3
+
+def test_llm_20_q_errors_on_error_question():
+    env = make("llm_20_questions", debug=True)
+    env.run([custom_questioner, custom_answerer, error_agent, custom_answerer])
+    json = env.toJSON()
+    assert json["name"] == "llm_20_questions"
+    assert json["rewards"] == [1, 1, None, 1]
+    assert json["statuses"] == ["DONE", "DONE", "ERROR", "DONE"]
+    print(len(json["steps"]))
+    assert len(json["steps"]) == 2

{kaggle_environments-1.14.9.dist-info → kaggle_environments-1.14.11.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: kaggle-environments
-Version: 1.14.9
+Version: 1.14.11
 Summary: Kaggle Environments
 Home-page: https://github.com/Kaggle/kaggle-environments
 Author: Kaggle

{kaggle_environments-1.14.9.dist-info → kaggle_environments-1.14.11.dist-info}/RECORD

@@ -1,4 +1,4 @@
-kaggle_environments/__init__.py,sha256=-cMWlmk778J5lbZYSW-yQhB6Ai_NOhDZDiRJd0SL13k,1702
+kaggle_environments/__init__.py,sha256=jB8Zxnx8_IrtPX8ArsCgrfeRPXcejHOZvDx1-PhHiE0,1703
 kaggle_environments/agent.py,sha256=BpYrAC6EOhJF3Wzynq-CWg8l-z5uXW1Lq4qUsyOM0ik,6203
 kaggle_environments/api.py,sha256=eLBKqr11Ku4tdsMUdUqy74FIVEA_hdV3_QUpX84x3Z8,798
 kaggle_environments/core.py,sha256=IrEkN9cIA2djBAxI8Sz1GRpGNKjhqbnBdV6irAeTm8Q,27851
@@ -72,7 +72,8 @@ kaggle_environments/envs/kore_fleets/starter_bots/ts/test/observation.json,sha25
 kaggle_environments/envs/llm_20_questions/keywords.py,sha256=BnyVrV0_xbiDt_Td4YXSKae9hHndAo5YYfYVtyL4eS4,45282
 kaggle_environments/envs/llm_20_questions/llm_20_questions.js,sha256=o30Nx1Tt0eAxBsQukmyxYlr7GuvptTUADWUnEnYDHe8,6009
 kaggle_environments/envs/llm_20_questions/llm_20_questions.json,sha256=VYxgQEDCaJv910sQdE_FOtV2w1V83Y3C6nTjx4FFZyo,2101
-kaggle_environments/envs/llm_20_questions/llm_20_questions.py,sha256=vLNs0yKo86miW2TUYsmWyfOYn9f-jH-H7_krQ3neeHY,9873
+kaggle_environments/envs/llm_20_questions/llm_20_questions.py,sha256=Xj1slTw2WVwf94NNKV92qqJI52vuIY8quDwzuJq5Kgc,9563
+kaggle_environments/envs/llm_20_questions/test_llm_20_questions.py,sha256=sq04OePvY9VSR4Cjw4RI_i5FMwGni_ewpr5fcF2HldY,1831
 kaggle_environments/envs/lux_ai_2021/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 kaggle_environments/envs/lux_ai_2021/agents.py,sha256=bDqsY17Jxrk69a8HXb96ex2U1ipKkQPVJuXxmiA5hMA,421
 kaggle_environments/envs/lux_ai_2021/index.html,sha256=4Ec1FuHuuH1EvkvlLOpNUUFYLovnS_IA-rOOHEUjcGY,1550
@@ -176,9 +177,9 @@ kaggle_environments/envs/tictactoe/tictactoe.js,sha256=NZDT-oSG0a6a-rso9Ldh9qkJw
 kaggle_environments/envs/tictactoe/tictactoe.json,sha256=zMXZ8-fpT7FBhzz2FFBvRLn4XwtngjEqOieMvI6cCj8,1121
 kaggle_environments/envs/tictactoe/tictactoe.py,sha256=iLNU5V-lz7Xab-d1vpPMfU5jDM3QtgBUH63Y_SU7I9Y,3639
 kaggle_environments/static/player.html,sha256=HH8qvFfTIDw8eZvw5W88jcCu58Lo4_eUQ1ak46KWVGQ,22945
-kaggle_environments-1.14.9.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-kaggle_environments-1.14.9.dist-info/METADATA,sha256=9FBsEf3b8jV3jB8F8N6_bQtqPqQYTHQ6hjmfomTgzMo,10700
-kaggle_environments-1.14.9.dist-info/WHEEL,sha256=m9WAupmBd2JGDsXWQGJgMGXIWbQY3F5c2xBJbBhq0nY,110
-kaggle_environments-1.14.9.dist-info/entry_points.txt,sha256=HbVC-LKGQFV6lEEYBYyDTtrkHgdHJUWQ8_qt9KHGqz4,70
-kaggle_environments-1.14.9.dist-info/top_level.txt,sha256=v3MMWIPMQFcI-WuF_dJngHWe9Bb2yH_6p4wat1x4gAc,20
-kaggle_environments-1.14.9.dist-info/RECORD,,
+kaggle_environments-1.14.11.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+kaggle_environments-1.14.11.dist-info/METADATA,sha256=uL7fO1pEmq1BAPPyPUXv3mBa3JSlWnN8LbJcZb_kXfc,10701
+kaggle_environments-1.14.11.dist-info/WHEEL,sha256=m9WAupmBd2JGDsXWQGJgMGXIWbQY3F5c2xBJbBhq0nY,110
+kaggle_environments-1.14.11.dist-info/entry_points.txt,sha256=HbVC-LKGQFV6lEEYBYyDTtrkHgdHJUWQ8_qt9KHGqz4,70
+kaggle_environments-1.14.11.dist-info/top_level.txt,sha256=v3MMWIPMQFcI-WuF_dJngHWe9Bb2yH_6p4wat1x4gAc,20
+kaggle_environments-1.14.11.dist-info/RECORD,,