Import tests: Shayan1/Prod/tests/test_llm_player

Import the existing local project snapshot. This commit organizes existing files; it does not represent newly developed functionality.

Shayan1/Prod/tests/test_llm_player.py

@@ -0,0 +1,209 @@
+import pytest
+from unittest.mock import MagicMock, patch
+from nash.game import Game
+from nash.players import LLMPlayer, make_player, PLAYER_KINDS
+
+
+@pytest.fixture
+def mock_anthropic():
+    with patch("anthropic.Anthropic") as mock:
+        yield mock
+
+
+def test_llm_player_initialization(mock_anthropic):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    # Test PLAYER_KINDS
+    assert "llm" in PLAYER_KINDS
+
+    # Test make_player
+    player = make_player("llm", game, player_index=0, model="my-test-model")
+    assert isinstance(player, LLMPlayer)
+    assert player.game == game
+    assert player.player_index == 0
+    assert player.model == "my-test-model"
+    mock_anthropic.assert_called_once()
+
+
+def test_llm_player_choose_clean_json(mock_anthropic):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    logs = []
+
+    def debug_logger(player_index, message):
+        logs.append((player_index, message))
+
+    player = make_player("llm", game, player_index=0, debug_logger=debug_logger)
+
+    mock_client = mock_anthropic.return_value
+    mock_response = MagicMock()
+    mock_content = MagicMock()
+    mock_content.text = '{"reasoning": "I want to cooperate to maximize payoff.", "strategy_index": 0}'
+    mock_response.content = [mock_content]
+    mock_client.messages.create.return_value = mock_response
+
+    # Choose action with history
+    history = [(0, 1), (1, 0)]
+    action = player.choose(history)
+
+    # Assert return value is strategy_index
+    assert action == 0
+
+    # Assert logs captured reasoning
+    assert len(logs) == 1
+    assert logs[0] == (0, "I want to cooperate to maximize payoff.")
+
+    # Check that client.messages.create was called
+    mock_client.messages.create.assert_called_once()
+
+
+@pytest.mark.parametrize("model_output,expected_idx,expected_reasoning", [
+    (
+        '```json\n{"reasoning": "Markdown fence with json key", "strategy_index": 1}\n```',
+        1,
+        "Markdown fence with json key"
+    ),
+    (
+        '```\n{"reasoning": "Markdown fence without json key", "strategy_index": 0}\n```',
+        0,
+        "Markdown fence without json key"
+    ),
+    (
+        'Here is the response:\n{"reasoning": "Surrounding text", "strategy_index": 1}\nHope this helps!',
+        1,
+        "Surrounding text"
+    ),
+    (
+        'Some intro...\n```json\n{"reasoning": "Intro and fence and outro", "strategy_index": 0}\n```\nSome outro...',
+        0,
+        "Intro and fence and outro"
+    ),
+])
+def test_llm_player_resilient_json_parsing(mock_anthropic, model_output, expected_idx, expected_reasoning):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    player = LLMPlayer(game, player_index=0)
+
+    mock_client = mock_anthropic.return_value
+    mock_response = MagicMock()
+    mock_content = MagicMock()
+    mock_content.text = model_output
+    mock_response.content = [mock_content]
+    mock_client.messages.create.return_value = mock_response
+
+    action = player.choose([])
+    assert action == expected_idx
+
+
+def test_llm_player_invalid_json_raises(mock_anthropic):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    player = LLMPlayer(game, player_index=0)
+
+    mock_client = mock_anthropic.return_value
+    mock_response = MagicMock()
+    mock_content = MagicMock()
+    mock_content.text = "This is just text with no json braces at all."
+    mock_response.content = [mock_content]
+    mock_client.messages.create.return_value = mock_response
+
+    with pytest.raises(ValueError, match="No JSON object found in model response"):
+        player.choose([])
+
+
+def test_llm_player_prompt_contents(mock_anthropic):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    player = LLMPlayer(game, player_index=0)
+
+    mock_client = mock_anthropic.return_value
+    mock_response = MagicMock()
+    mock_content = MagicMock()
+    mock_content.text = '{"reasoning": "strategic choice", "strategy_index": 1}'
+    mock_response.content = [mock_content]
+    mock_client.messages.create.return_value = mock_response
+
+    history = [(0, 1), (1, 0)]
+    player.choose(history)
+
+    # Retrieve the arguments passed to create
+    mock_client.messages.create.assert_called_once()
+    kwargs = mock_client.messages.create.call_args[1]
+
+    # Check messages structure
+    assert "messages" in kwargs
+    messages = kwargs["messages"]
+    assert len(messages) == 1
+    prompt = messages[0]["content"]
+
+    # 1. Player name
+    assert "Player 'P1'" in prompt
+
+    # 2. Strategy options
+    assert "Cooperate" in prompt
+    assert "Defect" in prompt
+
+    # 3. Payoff matrix formatting
+    # Profile: (Cooperate | Cooperate) -> Payoffs: (2.00, 2.00)
+    assert "Profile: (Cooperate | Cooperate) -> Payoffs: (2.00, 2.00)" in prompt
+    assert "Profile: (Defect | Cooperate) -> Payoffs: (3.00, 0.00)" in prompt
+
+    # 4. History of recent rounds
+    # Round 1: P1: Cooperate, P2: Defect
+    assert "Round 1: P1: Cooperate, P2: Defect" in prompt
+    assert "Round 2: P1: Defect, P2: Cooperate" in prompt
+
+
+def test_llm_player_empty_history(mock_anthropic):
+    game = Game(
+        players=["P1", "P2"],
+        strategies=[["Cooperate", "Defect"], ["Cooperate", "Defect"]],
+        payoffs=[
+            [[2.0, 0.0], [3.0, 1.0]],
+            [[2.0, 3.0], [0.0, 1.0]]
+        ]
+    )
+    player = LLMPlayer(game, player_index=0)
+
+    mock_client = mock_anthropic.return_value
+    mock_response = MagicMock()
+    mock_content = MagicMock()
+    mock_content.text = '{"reasoning": "no history", "strategy_index": 0}'
+    mock_response.content = [mock_content]
+    mock_client.messages.create.return_value = mock_response
+
+    player.choose([])
+
+    kwargs = mock_client.messages.create.call_args[1]
+    prompt = kwargs["messages"][0]["content"]
+    assert "No rounds played yet." in prompt