mirror of
https://github.com/open-thought/reasoning-gym.git
synced 2026-04-19 12:58:07 +00:00
dced3bfc45
* init * fix tests * unify codeio * filtered for libraries not present in reasoning-gym * fix more bounds * puzzle24 * knight swap curriculum * fix number sorting * fix attributes * add validation of config in creation of dataset * dry run for instantiating and validating the datasets * remove unused imports * fix curriculum tests to reference newly updated attribute names
322 lines
12 KiB
Python
322 lines
12 KiB
Python
"""Tests for Ttsumego problem generation"""
|
|
|
|
import re
|
|
|
|
import pytest
|
|
|
|
from reasoning_gym.games.tsumego import TsumegoConfig, TsumegoDataset
|
|
|
|
|
|
def test_config_validation():
|
|
# Valid configuration
|
|
TsumegoConfig(min_board_size=9, max_board_size=13, max_stones=10, size=100, seed=42)
|
|
|
|
# Invalid configurations
|
|
with pytest.raises(ValueError):
|
|
TsumegoConfig(min_board_size=4, max_board_size=13, max_stones=10) # min_board_size too low
|
|
with pytest.raises(ValueError):
|
|
TsumegoConfig(min_board_size=9, max_board_size=20, max_stones=10) # max_board_size too high
|
|
with pytest.raises(ValueError):
|
|
TsumegoConfig(min_board_size=13, max_board_size=9, max_stones=10) # min_board_size > max_board_size
|
|
with pytest.raises(ValueError):
|
|
TsumegoConfig(min_board_size=9, max_board_size=13, max_stones=2) # max_stones too low
|
|
|
|
|
|
def test_dataset_item_properties():
|
|
config = TsumegoConfig(min_board_size=9, max_board_size=9, max_stones=15, size=100, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
item = dataset[0]
|
|
# Check that item has the required keys
|
|
for key in ["question", "answer", "metadata"]:
|
|
assert key in item
|
|
|
|
metadata = item["metadata"]
|
|
for key in ["difficulty", "board"]:
|
|
assert key in metadata
|
|
|
|
board = metadata["board"]
|
|
# Board size should be equal to the fixed min_board_size for this test
|
|
assert len(board) == config.min_board_size
|
|
assert all(len(row) == config.min_board_size for row in board)
|
|
# Check stone count does not exceed max_stones + 7 (to account for extra fill in capture formation)
|
|
stone_count = sum(cell in "XO" for row in board for cell in row)
|
|
assert stone_count <= config.max_stones + 7
|
|
|
|
|
|
def test_deterministic_generation():
|
|
config = TsumegoConfig(min_board_size=9, max_board_size=9, max_stones=10, seed=42)
|
|
dataset1 = TsumegoDataset(config)
|
|
dataset2 = TsumegoDataset(config)
|
|
for i in range(3):
|
|
item1 = dataset1[i]
|
|
item2 = dataset2[i]
|
|
assert item1["metadata"]["board"] == item2["metadata"]["board"]
|
|
assert item1["answer"] == item2["answer"]
|
|
|
|
|
|
def test_liberties_and_move():
|
|
# Use a small board for simplicity
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=10)
|
|
dataset = TsumegoDataset(config)
|
|
|
|
# Part 1: Liberty counting test
|
|
board_liberties = [
|
|
[".", "O", ".", ".", "."],
|
|
["O", "X", "O", ".", "."],
|
|
[".", "O", ".", ".", "."],
|
|
[".", ".", ".", ".", "."],
|
|
[".", ".", ".", ".", "."],
|
|
]
|
|
liberties = dataset._get_liberties(board_liberties, 1, 1)
|
|
assert len(liberties) == 0
|
|
liberties_edge = dataset._get_liberties(board_liberties, 0, 1)
|
|
assert len(liberties_edge) == 2
|
|
|
|
# Part 2: Test capturing move
|
|
# Construct a board where an enemy stone at (2,2) is surrounded on three sides,
|
|
# so that placing an "X" at (2,3) will remove its last liberty and capture it.
|
|
board_capture = [["." for _ in range(5)] for _ in range(5)]
|
|
board_capture[1][2] = "X"
|
|
board_capture[2][1] = "X"
|
|
board_capture[3][2] = "X"
|
|
board_capture[2][2] = "O"
|
|
# Now, (2,2) (enemy) has only one liberty at (2,3).
|
|
# Placing "X" at (2,3) should capture the enemy stone.
|
|
assert dataset._is_valid_move(board_capture, 2, 3, "X")
|
|
dataset._make_move(board_capture, 2, 3, "X")
|
|
# After move, captured_stones should be [(2,2)] and ko point set to (2,2).
|
|
assert not dataset._is_valid_move(board_capture, 2, 2, "O"), "Ko move should be invalid"
|
|
|
|
# Part 3: Test suicide move (without capture)
|
|
board_move = [
|
|
[".", "O", ".", ".", "."],
|
|
["O", ".", "O", ".", "."],
|
|
[".", "O", ".", ".", "."],
|
|
[".", ".", ".", ".", "."],
|
|
[".", ".", ".", ".", "."],
|
|
]
|
|
# Placing "X" at (1,1) would be suicide as all adjacent positions are occupied by "O".
|
|
assert not dataset._is_valid_move(board_move, 1, 1, "X")
|
|
|
|
|
|
def test_score_answer():
|
|
config = TsumegoConfig(min_board_size=9, max_board_size=9, max_stones=10, size=5, seed=123)
|
|
dataset = TsumegoDataset(config)
|
|
|
|
entry = dataset[0]
|
|
|
|
assert dataset.score_answer(entry["answer"], entry) == 1.0
|
|
|
|
# correct solution with additional text
|
|
assert 0 < dataset.score_answer("move: " + entry["answer"], entry) < 1.0
|
|
|
|
# Incorrect letter-number move, but correct letter-digit format
|
|
assert dataset.score_answer("D4", entry) == 0.05
|
|
|
|
# Invalid format
|
|
assert dataset.score_answer("invalid", entry) == 0.01
|
|
|
|
# Empty answer
|
|
assert dataset.score_answer("", entry) == 0.0
|
|
|
|
# None answer
|
|
assert dataset.score_answer(None, entry) == 0.0
|
|
|
|
# test optimal score for answers, patching each entry
|
|
for x in dataset:
|
|
assert len(x["metadata"]["board"]) == x["metadata"]["board_size"]
|
|
assert dataset.score_answer(x["answer"], entry=x) == 1.0
|
|
|
|
|
|
# Additional tests for game logic edge cases
|
|
|
|
|
|
def test_get_group():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [
|
|
["X", "X", "."],
|
|
[".", "X", "O"],
|
|
[".", ".", "O"],
|
|
]
|
|
group_X = dataset._get_group(board, 0, 0)
|
|
expected_group_X = {(0, 0), (0, 1), (1, 1)}
|
|
assert group_X == expected_group_X
|
|
|
|
group_O = dataset._get_group(board, 1, 2)
|
|
expected_group_O = {(1, 2), (2, 2)}
|
|
assert group_O == expected_group_O
|
|
|
|
|
|
def test_count_liberties():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [
|
|
["X", "X", "."],
|
|
[".", "X", "O"],
|
|
[".", ".", "O"],
|
|
]
|
|
group_X = {(0, 0), (0, 1), (1, 1)}
|
|
liberties_X = dataset._count_liberties(board, group_X)
|
|
# For (0,0): neighbor (1,0); (0,1): neighbor (0,2); (1,1): neighbors (1,0) and (2,1)
|
|
# Combined unique liberties: {(1,0), (0,2), (2,1)} so count should be 3
|
|
assert liberties_X == 3
|
|
|
|
|
|
def test_out_of_bounds_move():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [["." for _ in range(5)] for _ in range(5)]
|
|
# Test moves that are out of bounds
|
|
assert not dataset._is_valid_move(board, -1, 0, "X")
|
|
assert not dataset._is_valid_move(board, 0, -1, "X")
|
|
assert not dataset._is_valid_move(board, 5, 0, "X")
|
|
assert not dataset._is_valid_move(board, 0, 5, "X")
|
|
|
|
|
|
def test_move_on_occupied_intersection():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [["." for _ in range(5)] for _ in range(5)]
|
|
board[1][1] = "X"
|
|
# Attempting to play on an occupied spot should be invalid
|
|
assert not dataset._is_valid_move(board, 1, 1, "O")
|
|
assert not dataset._is_valid_move(board, 1, 1, "X")
|
|
|
|
|
|
def test_valid_non_capturing_move():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [["." for _ in range(5)] for _ in range(5)]
|
|
# A move on an empty board that doesn't result in capture or suicide should be valid
|
|
assert dataset._is_valid_move(board, 0, 0, "X")
|
|
move_result = dataset._make_move(board, 0, 0, "X")
|
|
assert move_result
|
|
assert board[0][0] == "X"
|
|
|
|
|
|
def test_multiple_capture():
|
|
# Set up a board where a move will capture multiple opponent stones,
|
|
# which should not trigger the ko rule (ko point remains None)
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
board = [
|
|
[".", ".", ".", ".", "."],
|
|
[".", "X", "X", "X", "."],
|
|
["X", "O", "O", ".", "."],
|
|
[".", "X", "X", "X", "."],
|
|
[".", ".", ".", ".", "."],
|
|
]
|
|
# Move at (2,3) with 'X' should capture the opponent stones at (2,1) and (2,2)
|
|
assert dataset._is_valid_move(board, 2, 3, "X")
|
|
move_result = dataset._make_move(board, 2, 3, "X")
|
|
assert move_result, "Move should be successfully made"
|
|
assert board[2][1] == ".", "Stone at (2,1) should be captured"
|
|
assert board[2][2] == ".", "Stone at (2,2) should be captured"
|
|
assert dataset._ko_point is None, "Ko point should not be set for multiple captures"
|
|
|
|
|
|
def test_would_capture():
|
|
config = TsumegoConfig(min_board_size=5, max_board_size=5, max_stones=10, size=1, seed=42)
|
|
dataset = TsumegoDataset(config)
|
|
# Create a scenario similar to the one in test_liberties_and_move for capturing
|
|
board_capture = [["." for _ in range(5)] for _ in range(5)]
|
|
board_capture[1][2] = "X"
|
|
board_capture[2][1] = "X"
|
|
board_capture[3][2] = "X"
|
|
board_capture[2][2] = "O"
|
|
# Placing 'X' at (2,3) should capture the stone at (2,2)
|
|
assert dataset._would_capture(board_capture, 2, 3, "X")
|
|
# In a scenario with no capture, the move should not be considered capturing
|
|
board_no_capture = [["." for _ in range(5)] for _ in range(5)]
|
|
board_no_capture[2][2] = "O"
|
|
assert not dataset._would_capture(board_no_capture, 0, 0, "X")
|
|
|
|
|
|
def test_capture_verification():
|
|
"""Verifies that the solution move in a generated puzzle captures at least one opponent stone."""
|
|
config = TsumegoConfig(min_board_size=9, max_board_size=9, max_stones=15, size=1, seed=10)
|
|
dataset = TsumegoDataset(config)
|
|
entry = dataset[0]
|
|
board = entry["metadata"]["board"]
|
|
|
|
# If solution is a letter+number string, convert it
|
|
|
|
def convert_solution(sol, board_size):
|
|
# sol is expected to be a string like 'E5'
|
|
letter = sol[0].upper()
|
|
number = int(sol[1:])
|
|
return (board_size - number, ord(letter) - ord("A"))
|
|
|
|
solution = entry["answer"]
|
|
assert isinstance(solution, str)
|
|
solution = convert_solution(solution, len(board))
|
|
initial_white = sum(row.count("O") for row in board)
|
|
|
|
# Make a deep copy of the board to simulate the move
|
|
board_after = [row[:] for row in board]
|
|
move_success = dataset._make_move(board_after, solution[0], solution[1], "X")
|
|
assert move_success, "The solution move should be legal."
|
|
|
|
final_white = sum(row.count("O") for row in board_after)
|
|
assert final_white < initial_white, "The solution move should capture at least one opponent stone."
|
|
|
|
|
|
def test_tsumego_curriculum():
|
|
"""Test the TsumegoCurriculum functionality"""
|
|
from reasoning_gym.games.tsumego import TsumegoCurriculum
|
|
|
|
curriculum = TsumegoCurriculum()
|
|
|
|
base_value = {"size": 150, "seed": 1}
|
|
|
|
# Test initial configuration
|
|
base_cfg = curriculum.generate_configuration(base_value)
|
|
assert base_cfg.seed == 1
|
|
assert base_cfg.size == 150
|
|
assert base_cfg.min_board_size == 5 and base_cfg.max_board_size == 5
|
|
assert base_cfg.max_stones == 5
|
|
|
|
# Test incrementing attribute level
|
|
curriculum.increment_attr_level("board_size")
|
|
increased_cfg = curriculum.generate_configuration(base_value)
|
|
assert increased_cfg.min_board_size == 5 and increased_cfg.max_board_size == 10
|
|
assert increased_cfg.max_stones == 5
|
|
|
|
# Test incrementing attribute level again, and increment max_stones
|
|
curriculum.increment_attr_level("board_size")
|
|
curriculum.increment_attr_level("max_stones")
|
|
increased_cfg_2 = curriculum.generate_configuration(base_value)
|
|
assert increased_cfg_2.min_board_size == 5 and increased_cfg_2.max_board_size == 15
|
|
assert increased_cfg_2.max_stones == 10
|
|
|
|
# Test decrementing attribute level
|
|
curriculum.decrement_attr_level("board_size")
|
|
decreased_cfg = curriculum.generate_configuration(base_value)
|
|
assert decreased_cfg.min_board_size == 5 and decreased_cfg.max_board_size == 10
|
|
assert decreased_cfg.max_stones == 10 # Unchanged
|
|
|
|
# Test global level adjustments
|
|
curriculum = TsumegoCurriculum() # Reset curriculum
|
|
assert curriculum.get_attr_level("board_size") == 0
|
|
|
|
# Increase global level
|
|
curriculum.increment_global_level()
|
|
assert curriculum.get_attr_level("board_size") == 1
|
|
|
|
global_level_cfg = curriculum.generate_configuration(base_value)
|
|
assert global_level_cfg.min_board_size == 5 and global_level_cfg.max_board_size == 10
|
|
|
|
# Increase global level again
|
|
curriculum.increment_global_level()
|
|
assert curriculum.get_attr_level("board_size") == 2
|
|
|
|
global_level_cfg_2 = curriculum.generate_configuration(base_value)
|
|
assert global_level_cfg_2.min_board_size == 5 and global_level_cfg_2.max_board_size == 15
|
|
|
|
# Decrease global level
|
|
curriculum.decrement_global_level()
|
|
assert curriculum.get_attr_level("board_size") == 1
|
|
|
|
global_level_cfg_3 = curriculum.generate_configuration(base_value)
|
|
assert global_level_cfg_3.min_board_size == 5 and global_level_cfg_3.max_board_size == 10
|