Data collisions notebooks and data (#406)

* Add collisions data

* Fix logic issues in basic_arithmetic and gsm_symbolic data

notebooks/collisions/collisions.txt

@@ -0,0 +1,86 @@
+complex_arithmetic, 0
+intermediate_integration, 12
+polynomial_equations, 0
+polynomial_multiplication, 0
+simple_equations, 0
+simple_integration, 0
+ab, 0
+base_conversion, 1
+binary_alternation, 0
+binary_matrix, 0
+caesar_cipher, 2
+count_primes, 0
+cryptarithm, 0
+game_of_life, 0
+game_of_life_halting, 17
+graph_color, 0
+group_anagrams, 0
+isomorphic_strings, 0
+jugs, 11
+letter_counting, 0
+letter_jumble, 0
+manipulate_matrix, 0
+number_filtering, 0
+number_sorting, 0
+palindrome_generation, 0
+palindrome_partitioning, 0
+pool_matrix, 0
+ransom_note, 0
+rotate_matrix, 0
+rotten_oranges, 0
+sentence_reordering, 6
+spell_backward, 98
+spiral_matrix, 0
+string_insertion, 0
+string_manipulation, 0
+string_splitting, 36
+string_synthesis, 36
+word_ladder, 0
+word_sequence_reversal, 0
+word_sorting, 0
+arc_1d, 0
+arc_agi, 0
+rearc, 0
+basic_arithmetic, 0
+bitwise_arithmetic, 0
+calendar_arithmetic, 0
+chain_sum, 0
+count_bits, 0
+decimal_arithmetic, 0
+decimal_chain_sum, 0
+dice, 0
+fraction_simplification, 0
+gcd, 0
+gsm_symbolic, 0
+lcm, 3
+leg_counting, 0
+number_format, 0
+power_function, 0
+prime_factorization, 16
+products, 3
+time_intervals, 0
+bf, 5
+codeio, 2
+color_cube_rotation, 0
+figlet_font, 0
+modulo_grid, 0
+needle_haystack, 0
+number_sequence, 8
+rectangle_count, 0
+rubiks_cube, 1
+countdown, 0
+emoji_mystery, 0
+advanced_geometry, 0
+simple_geometry, 0
+course_schedule, 0
+family_relationships, 0
+largest_island, 5
+quantum_lock, 3
+shortest_path, 0
+list_functions, 0
+aiw, 0
+circuit_logic, 0
+knights_knaves, 0
+propositional_logic, 2
+self_reference, 31
+syllogism, 0

reasoning_gym/arithmetic/basic_arithmetic.py

@@ -161,10 +161,12 @@ class BasicArithmeticDataset(ProceduralDataset):
                         right_parts.append(")")
 
                     else:
-                        divisor = rng.choice(find_common_divisors(dividend, 0))
+                        if dividend != 0:
+                            divisor = rng.choice(find_common_divisors(dividend, 0))
+                        else:
+                            divisor = rng.randint(1, 10**num_digits - 1)
                         left_parts.append(str(divisor))
                         left_parts.append("+")
-
                     left_parts.extend(right_parts)
                 else:
                     if dividend != 0:

reasoning_gym/arithmetic/gsm_symbolic/generators_00_49.py

@@ -2049,7 +2049,7 @@ def generate_27(rng: Random, difficulty: float = 1.0) -> dict[str, Any]:
         third_complex = int(first_two * percent_bigger / 100)
         total_apartments = first_two + third_complex + first_two
         weekly_visits = total_apartments * freq
-        weekly_earnings = weekly_visits * rate
+        weekly_earnings = round(weekly_visits * rate, 2)
 
         question = f"{name} collects garbage from {n} different apartment complexes. The first {n_first} have {apartments_each} apartments each and the last one is {percent_bigger}% bigger than the other {n_first} combined. {name} collects garbage {freq} times a week from each place and he gets paid {currency}{rate:.2f} per collection for each apartment. How much money does he make in a week?"