mirror of
https://github.com/open-thought/reasoning-gym.git
synced 2026-04-22 16:49:06 +00:00
Refactor LetterJumble
This commit is contained in:
EduardDurech
parent
b8ce5a8a5d
commit
18b6e71fa9
6 changed files with 550 additions and 190 deletions
|
|
@ -1,121 +1,289 @@
|
|||
"""Tests for letter jumbling task generation"""
|
||||
"""Unit tests for the letter jumble exercise."""
|
||||
|
||||
from random import Random
|
||||
from reasoning_gym.curricula.algorithmic.letter_jumble_curriculum import LetterJumbleCurriculum
|
||||
from reasoning_gym.exercises.algorithmic.letter_jumble import LetterJumbleExercise
|
||||
import unittest
|
||||
import random
|
||||
from collections import defaultdict
|
||||
|
||||
import pytest
|
||||
class TestLetterJumbleParsing(unittest.TestCase):
|
||||
"""Test parsing of letter jumble metadata"""
|
||||
|
||||
from reasoning_gym.algorithmic.letter_jumble import LetterJumbleConfig, LetterJumbleDataset
|
||||
def setUp(self):
|
||||
self.exercise = LetterJumbleExercise()
|
||||
|
||||
def test_parse_expression_basic(self):
|
||||
"""Test parsing of basic letter jumble metadata"""
|
||||
test_metadata = {
|
||||
"scrambled": {
|
||||
"scrambled_words": "EHLLO DLWOR",
|
||||
"original_words": ["HELLO", "WORLD"]
|
||||
}
|
||||
}
|
||||
parsed = self.exercise._parse_expression(test_metadata)
|
||||
self.assertEqual(parsed["scrambled_words"], ["EHLLO", "DLWOR"])
|
||||
self.assertEqual(parsed["original_words"], ["HELLO", "WORLD"])
|
||||
|
||||
def test_letter_jumble_config_validation():
|
||||
"""Test that invalid configs raise appropriate errors"""
|
||||
with pytest.raises(AssertionError):
|
||||
config = LetterJumbleConfig(min_word_len=0)
|
||||
config.validate()
|
||||
def test_parse_with_spaces(self):
|
||||
"""Test parsing with spaces and punctuation"""
|
||||
test_metadata = {
|
||||
"scrambled": {
|
||||
"scrambled_words": "EHLLO DLWOR!",
|
||||
"original_words": ["HELLO", "WORLD!"]
|
||||
}
|
||||
}
|
||||
parsed = self.exercise._parse_expression(test_metadata)
|
||||
self.assertEqual(parsed["scrambled_words"], ["EHLLO", "DLWOR!"])
|
||||
self.assertEqual(parsed["original_words"], ["HELLO", "WORLD!"])
|
||||
|
||||
with pytest.raises(AssertionError):
|
||||
config = LetterJumbleConfig(min_words=10, max_words=5)
|
||||
config.validate()
|
||||
def test_parse_mixed_case(self):
|
||||
"""Test parsing with mixed case text"""
|
||||
test_metadata = {
|
||||
"scrambled": {
|
||||
"scrambled_words": "HeLlO WoRlD",
|
||||
"original_words": ["hElLo", "wOrLd"]
|
||||
}
|
||||
}
|
||||
parsed = self.exercise._parse_expression(test_metadata)
|
||||
self.assertEqual(parsed["scrambled_words"], ["HeLlO", "WoRlD"])
|
||||
self.assertEqual(parsed["original_words"], ["hElLo", "wOrLd"])
|
||||
|
||||
with pytest.raises(AssertionError):
|
||||
config = LetterJumbleConfig(min_corruption_level=-0.1)
|
||||
config.validate()
|
||||
class TestLetterJumbleEvaluation(unittest.TestCase):
|
||||
"""Test evaluation of letter jumble problems"""
|
||||
|
||||
with pytest.raises(AssertionError):
|
||||
config = LetterJumbleConfig(max_corruption_level=1.1)
|
||||
config.validate()
|
||||
def setUp(self):
|
||||
self.exercise = LetterJumbleExercise()
|
||||
|
||||
def test_basic_unscrambling(self):
|
||||
"""Test basic unscrambling cases"""
|
||||
test_cases = [
|
||||
(["EHLLO"], "HELLO"), # Single word
|
||||
(["EHLLO", "DLWOR"], "HELLO WORLD"), # Two words
|
||||
(["AAAA"], "AAAA"), # Same letters
|
||||
(["ZBAC"], "ABCZ"), # Sorted order
|
||||
(["HELLO"], "HELLO") # Already unscrambled
|
||||
]
|
||||
for scrambled, expected in test_cases:
|
||||
parsed = {
|
||||
"scrambled_words": scrambled,
|
||||
"original_words": expected.split()
|
||||
}
|
||||
result = self.exercise._evaluate_expression(parsed)
|
||||
self.assertEqual(result, expected)
|
||||
|
||||
def test_letter_jumble_deterministic():
|
||||
"""Test that dataset generates same items with same seed"""
|
||||
config = LetterJumbleConfig(seed=42, size=10)
|
||||
dataset1 = LetterJumbleDataset(config)
|
||||
dataset2 = LetterJumbleDataset(config)
|
||||
def test_mixed_case_unscrambling(self):
|
||||
"""Test unscrambling with mixed case"""
|
||||
test_cases = [
|
||||
(["HeLlO"], "hElLo"), # Mixed case, single word
|
||||
(["WoRlD", "HeLlO"], "wOrLd hElLo"), # Mixed case, multiple words
|
||||
(["AbCdE"], "aBcDe") # Mixed case, alternating
|
||||
]
|
||||
for scrambled, expected in test_cases:
|
||||
parsed = {
|
||||
"scrambled_words": scrambled,
|
||||
"original_words": expected.split()
|
||||
}
|
||||
result = self.exercise._evaluate_expression(parsed)
|
||||
self.assertEqual(result, expected)
|
||||
|
||||
for i in range(len(dataset1)):
|
||||
assert dataset1[i] == dataset2[i]
|
||||
def test_with_spaces_and_punctuation(self):
|
||||
"""Test unscrambling with spaces and punctuation"""
|
||||
test_cases = [
|
||||
(["EHLLO!", "DLWOR?"], "HELLO! WORLD?"),
|
||||
(["EHLLO.", "DLWOR."], "HELLO. WORLD."),
|
||||
(["EHLLO,", "DLWOR,"], "HELLO, WORLD,")
|
||||
]
|
||||
for scrambled, expected in test_cases:
|
||||
parsed = {
|
||||
"scrambled_words": scrambled,
|
||||
"original_words": expected.split()
|
||||
}
|
||||
result = self.exercise._evaluate_expression(parsed)
|
||||
self.assertEqual(result, expected)
|
||||
|
||||
class TestLetterJumbleGeneration(unittest.TestCase):
|
||||
"""Test problem generation"""
|
||||
|
||||
def test_letter_jumble_scrambling():
|
||||
"""Test the word scrambling logic"""
|
||||
config = LetterJumbleConfig(
|
||||
min_word_len=4,
|
||||
max_word_len=8,
|
||||
min_words=1,
|
||||
max_words=1,
|
||||
min_corruption_level=0.5,
|
||||
max_corruption_level=0.5,
|
||||
size=1,
|
||||
seed=42,
|
||||
)
|
||||
dataset = LetterJumbleDataset(config)
|
||||
def setUp(self):
|
||||
self.curriculum = LetterJumbleCurriculum()
|
||||
self.exercise = LetterJumbleExercise()
|
||||
self.rng = random.Random(42)
|
||||
self.curriculum.rng = self.rng
|
||||
|
||||
# Test with known word
|
||||
word = "testing"
|
||||
rng = Random(42)
|
||||
scrambled = dataset._scramble_word(word, 0.5, rng)
|
||||
def test_problem_structure(self):
|
||||
"""Test that generated problems have the correct structure"""
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
|
||||
# Verify scrambled word:
|
||||
# - Has same length as original
|
||||
assert len(scrambled) == len(word)
|
||||
# - Contains same characters
|
||||
assert sorted(scrambled) == sorted(word)
|
||||
# - Is different from original (with high probability given 0.5 corruption)
|
||||
assert scrambled != word
|
||||
# Check basic structure
|
||||
self.assertIn("question", problem)
|
||||
self.assertIn("answer", problem)
|
||||
self.assertIn("metadata", problem)
|
||||
|
||||
# Check metadata structure
|
||||
metadata = problem["metadata"]
|
||||
self.assertEqual(metadata["type"], "direct")
|
||||
self.assertIn("executed_parts", metadata)
|
||||
executed_parts = metadata["executed_parts"]
|
||||
self.assertIn("scrambled_words", executed_parts)
|
||||
self.assertIn("original_words", executed_parts)
|
||||
|
||||
def test_letter_jumble_dataset_items():
|
||||
"""Test basic properties of generated items"""
|
||||
config = LetterJumbleConfig(
|
||||
min_word_len=4,
|
||||
max_word_len=8,
|
||||
min_words=3,
|
||||
max_words=5,
|
||||
min_corruption_level=0.1,
|
||||
max_corruption_level=0.3,
|
||||
size=50,
|
||||
seed=42,
|
||||
)
|
||||
dataset = LetterJumbleDataset(config)
|
||||
def test_word_length_ranges(self):
|
||||
"""Test that word lengths are within expected ranges"""
|
||||
# Test all word length levels
|
||||
level_max_lengths = {0: 5, 1: 8, 2: 64}
|
||||
|
||||
for i in range(len(dataset)):
|
||||
item = dataset[i]
|
||||
for level, max_length in level_max_lengths.items():
|
||||
self.curriculum.set_attr_level("word_length", level)
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
words = problem["metadata"]["executed_parts"]["original_words"]
|
||||
for word in words:
|
||||
self.assertLessEqual(len(word), max_length)
|
||||
self.assertGreaterEqual(len(word), 2) # Min length is 2
|
||||
|
||||
# Check item structure
|
||||
assert isinstance(item, dict)
|
||||
assert "question" in item
|
||||
assert "answer" in item
|
||||
assert "metadata" in item
|
||||
def test_word_count_ranges(self):
|
||||
"""Test that word counts are within expected ranges"""
|
||||
# Test all word count levels
|
||||
level_word_counts = {0: 3, 1: 5, 2: 20}
|
||||
|
||||
# Check metadata
|
||||
metadata = item["metadata"]
|
||||
assert "num_words" in metadata
|
||||
assert "corruption_level" in metadata
|
||||
assert "scrambled_words" in metadata
|
||||
assert "original_words" in metadata
|
||||
for level, max_words in level_word_counts.items():
|
||||
self.curriculum.set_attr_level("num_words", level)
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
words = problem["metadata"]["executed_parts"]["original_words"]
|
||||
self.assertLessEqual(len(words), max_words)
|
||||
self.assertGreaterEqual(len(words), 1) # Min words is 1
|
||||
|
||||
# Verify word counts
|
||||
num_words = metadata["num_words"]
|
||||
assert config.min_words <= num_words <= config.max_words
|
||||
assert len(metadata["scrambled_words"]) == num_words
|
||||
assert len(metadata["original_words"]) == num_words
|
||||
class TestLetterJumbleComprehensive(unittest.TestCase):
|
||||
"""Comprehensive tests for letter jumble"""
|
||||
|
||||
# Verify corruption level
|
||||
assert config.min_corruption_level <= metadata["corruption_level"] <= config.max_corruption_level
|
||||
def setUp(self):
|
||||
self.curriculum = LetterJumbleCurriculum()
|
||||
self.exercise = LetterJumbleExercise()
|
||||
self.rng = random.Random(42)
|
||||
self.curriculum.rng = self.rng
|
||||
|
||||
# Verify word properties
|
||||
for word in metadata["original_words"]:
|
||||
assert config.min_word_len <= len(word) <= config.max_word_len
|
||||
assert word.isalpha()
|
||||
def test_corruption_levels(self):
|
||||
"""Test different corruption levels"""
|
||||
corruption_levels = [0.1, 0.3, 0.9]
|
||||
num_samples = 100 # Test with multiple samples
|
||||
|
||||
# Test each level
|
||||
for level, expected_corruption in enumerate(corruption_levels):
|
||||
self.curriculum.set_attr_level("corruption_level", level)
|
||||
differences = []
|
||||
|
||||
def test_letter_jumble_iteration():
|
||||
"""Test that iteration respects dataset size"""
|
||||
config = LetterJumbleConfig(size=5, seed=42)
|
||||
dataset = LetterJumbleDataset(config)
|
||||
# Generate multiple problems to measure average corruption
|
||||
for _ in range(num_samples):
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
metadata = problem["metadata"]["executed_parts"]
|
||||
# Calculate character differences
|
||||
preserve_len = self.curriculum.attributes["preserve_length"].levels[self.curriculum.get_attr_level("preserve_length")]
|
||||
for orig, scrambled in zip(metadata["original_words"], metadata["scrambled_words"]):
|
||||
if len(orig) > preserve_len:
|
||||
diff_count = sum(1 for a, b in zip(orig, scrambled) if a != b)
|
||||
differences.append(diff_count / len(orig))
|
||||
|
||||
items = list(dataset)
|
||||
assert len(items) == config.size
|
||||
# Check average corruption level is reasonable
|
||||
# It's okay if actual corruption is lower than target due to:
|
||||
# 1. Some swaps might cancel out previous swaps
|
||||
# 2. The same characters might be swapped multiple times
|
||||
# 3. The preserve_length attribute prevents some characters from being swapped
|
||||
# 4. For short words, even a few swaps can make them readable
|
||||
if differences:
|
||||
avg_corruption = sum(differences) / len(differences)
|
||||
# Only check that we don't exceed target by too much
|
||||
self.assertLess(avg_corruption, expected_corruption + 0.1,
|
||||
f"Corruption level {avg_corruption:.2f} too high (target: {expected_corruption:.2f})")
|
||||
# And ensure we have some corruption
|
||||
self.assertGreater(avg_corruption, 0.02,
|
||||
f"Corruption level {avg_corruption:.2f} too low (should be above 0.02)")
|
||||
|
||||
# Test multiple iterations yield same items
|
||||
assert items == list(dataset)
|
||||
def test_template_variation(self):
|
||||
"""Test that different templates are used"""
|
||||
templates_seen = set()
|
||||
num_samples = 100
|
||||
|
||||
for _ in range(num_samples):
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
templates_seen.add(problem["question"].split(":")[0])
|
||||
|
||||
self.assertGreater(len(templates_seen), 1, "Not enough template variation")
|
||||
|
||||
def test_comprehensive_random_evaluation(self):
|
||||
"""Test random evaluation with various configurations and track statistics."""
|
||||
self.rng = random.Random(42) # Fixed seed for reproducibility
|
||||
self.curriculum.rng = self.rng
|
||||
|
||||
# Track statistics
|
||||
word_lengths = defaultdict(int)
|
||||
word_counts = defaultdict(int)
|
||||
corruption_levels = defaultdict(list)
|
||||
consecutive_words_count = 0
|
||||
total_samples = 1000
|
||||
|
||||
# Generate test cases
|
||||
for _ in range(total_samples):
|
||||
# Set random attribute levels
|
||||
for attr in self.curriculum.attributes:
|
||||
max_level = len(self.curriculum.attributes[attr].levels) - 1
|
||||
self.curriculum.set_attr_level(attr, self.rng.randint(0, max_level))
|
||||
|
||||
# Generate and evaluate a random problem
|
||||
problem = self.exercise.generate(self.curriculum)
|
||||
metadata = problem["metadata"]["executed_parts"]
|
||||
original_words = metadata["original_words"]
|
||||
scrambled_words = metadata["scrambled_words"]
|
||||
|
||||
# Track statistics
|
||||
word_counts[len(original_words)] += 1
|
||||
for word in original_words:
|
||||
word_lengths[len(word)] += 1
|
||||
|
||||
# Calculate corruption levels
|
||||
for orig, scrambled in zip(original_words, scrambled_words):
|
||||
preserve_len = self.curriculum.attributes["preserve_length"].levels[self.curriculum.get_attr_level("preserve_length")]
|
||||
if len(orig) > preserve_len:
|
||||
diff_count = sum(1 for a, b in zip(orig, scrambled) if a != b)
|
||||
corruption_levels[len(orig)].append(diff_count / len(orig))
|
||||
|
||||
# Check if words are consecutive in source text
|
||||
if len(original_words) > 1:
|
||||
text = " ".join(self.curriculum.words)
|
||||
phrase = " ".join(original_words)
|
||||
if phrase in text:
|
||||
consecutive_words_count += 1
|
||||
|
||||
# Verify scrambling is valid
|
||||
for orig, scrambled in zip(original_words, scrambled_words):
|
||||
# Check lengths match
|
||||
self.assertEqual(len(orig), len(scrambled))
|
||||
# Check same letters are used
|
||||
self.assertEqual(sorted(orig), sorted(scrambled))
|
||||
|
||||
# Print statistics
|
||||
print("\nWord length distribution:")
|
||||
for length, count in sorted(word_lengths.items()):
|
||||
print(f" Length {length}: {count}")
|
||||
|
||||
print("\nWord count distribution:")
|
||||
for count, freq in sorted(word_counts.items()):
|
||||
print(f" {count} words: {freq}")
|
||||
|
||||
print("\nAverage corruption levels by word length:")
|
||||
for length, levels in sorted(corruption_levels.items()):
|
||||
avg = sum(levels) / len(levels) if levels else 0
|
||||
print(f" Length {length}: {avg:.2f}")
|
||||
|
||||
print(f"\nConsecutive words: {consecutive_words_count}/{total_samples}")
|
||||
|
||||
# Verify statistical properties
|
||||
self.assertTrue(any(length >= 8 for length in word_lengths),
|
||||
"No long words generated")
|
||||
self.assertTrue(any(count >= 3 for count in word_counts.values()),
|
||||
"Not enough variation in word counts")
|
||||
self.assertTrue(consecutive_words_count > 0,
|
||||
"No consecutive words generated")
|
||||
self.assertTrue(consecutive_words_count < total_samples,
|
||||
"Too many consecutive words")
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
Loading…
Add table
Add a link
Reference in a new issue