|
| 1 | +import numpy |
| 2 | +import pygad |
| 3 | +import random |
| 4 | + |
| 5 | +# Global constants for testing |
| 6 | +num_generations = 100 |
| 7 | +num_parents_mating = 5 |
| 8 | +sol_per_pop = 10 |
| 9 | +num_genes = 3 |
| 10 | +random_seed = 42 |
| 11 | + |
| 12 | +def fitness_func(ga_instance, solution, solution_idx): |
| 13 | + """Single-objective fitness function.""" |
| 14 | + return numpy.sum(solution**2) |
| 15 | + |
| 16 | +def fitness_func_multi(ga_instance, solution, solution_idx): |
| 17 | + """Multi-objective fitness function.""" |
| 18 | + return [numpy.sum(solution**2), numpy.sum(solution)] |
| 19 | + |
| 20 | +def test_best_solution_consistency_single_objective(): |
| 21 | + """ |
| 22 | + Test best_solution() consistency for single-objective optimization. |
| 23 | + """ |
| 24 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 25 | + num_parents_mating=num_parents_mating, |
| 26 | + fitness_func=fitness_func, |
| 27 | + sol_per_pop=sol_per_pop, |
| 28 | + num_genes=num_genes, |
| 29 | + random_seed=random_seed, |
| 30 | + suppress_warnings=True |
| 31 | + ) |
| 32 | + ga_instance.run() |
| 33 | + |
| 34 | + # Call with last_generation_fitness |
| 35 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 36 | + |
| 37 | + # Call without pop_fitness |
| 38 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 39 | + |
| 40 | + assert numpy.array_equal(sol1, sol2) |
| 41 | + assert fitness1 == fitness2 |
| 42 | + assert idx1 == idx2 |
| 43 | + print("test_best_solution_consistency_single_objective passed.") |
| 44 | + |
| 45 | +def test_best_solution_consistency_multi_objective(): |
| 46 | + """ |
| 47 | + Test best_solution() consistency for multi-objective optimization. |
| 48 | + """ |
| 49 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 50 | + num_parents_mating=num_parents_mating, |
| 51 | + fitness_func=fitness_func_multi, |
| 52 | + sol_per_pop=sol_per_pop, |
| 53 | + num_genes=num_genes, |
| 54 | + random_seed=random_seed, |
| 55 | + parent_selection_type="nsga2", |
| 56 | + suppress_warnings=True |
| 57 | + ) |
| 58 | + ga_instance.run() |
| 59 | + |
| 60 | + # Call with last_generation_fitness |
| 61 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 62 | + |
| 63 | + # Call without pop_fitness |
| 64 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 65 | + |
| 66 | + assert numpy.array_equal(sol1, sol2) |
| 67 | + assert numpy.array_equal(fitness1, fitness2) |
| 68 | + assert idx1 == idx2 |
| 69 | + print("test_best_solution_consistency_multi_objective passed.") |
| 70 | + |
| 71 | +def test_best_solution_before_run(): |
| 72 | + """ |
| 73 | + Test best_solution() consistency before run() is called. |
| 74 | + """ |
| 75 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 76 | + num_parents_mating=num_parents_mating, |
| 77 | + fitness_func=fitness_func, |
| 78 | + sol_per_pop=sol_per_pop, |
| 79 | + num_genes=num_genes, |
| 80 | + random_seed=random_seed, |
| 81 | + suppress_warnings=True |
| 82 | + ) |
| 83 | + |
| 84 | + # Before run(), last_generation_fitness is None |
| 85 | + # We can still call best_solution(), it should call cal_pop_fitness() |
| 86 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 87 | + |
| 88 | + # Now cal_pop_fitness() should match ga_instance.best_solution() output if we pass it |
| 89 | + pop_fitness = ga_instance.cal_pop_fitness() |
| 90 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=pop_fitness) |
| 91 | + |
| 92 | + assert numpy.array_equal(sol1, sol2) |
| 93 | + assert fitness1 == fitness2 |
| 94 | + assert idx1 == idx2 |
| 95 | + print("test_best_solution_before_run passed.") |
| 96 | + |
| 97 | +def test_best_solution_with_save_solutions(): |
| 98 | + """ |
| 99 | + Test best_solution() consistency when save_solutions=True. |
| 100 | + This tests the caching mechanism in cal_pop_fitness(). |
| 101 | + """ |
| 102 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 103 | + num_parents_mating=num_parents_mating, |
| 104 | + fitness_func=fitness_func, |
| 105 | + sol_per_pop=sol_per_pop, |
| 106 | + num_genes=num_genes, |
| 107 | + random_seed=random_seed, |
| 108 | + save_solutions=True, |
| 109 | + suppress_warnings=True |
| 110 | + ) |
| 111 | + ga_instance.run() |
| 112 | + |
| 113 | + # Call with last_generation_fitness |
| 114 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 115 | + |
| 116 | + # Call without pop_fitness (this will trigger cal_pop_fitness which uses saved solutions) |
| 117 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 118 | + |
| 119 | + assert numpy.array_equal(sol1, sol2) |
| 120 | + assert fitness1 == fitness2 |
| 121 | + assert idx1 == idx2 |
| 122 | + print("test_best_solution_with_save_solutions passed.") |
| 123 | + |
| 124 | +def test_best_solution_with_save_best_solutions(): |
| 125 | + """ |
| 126 | + Test best_solution() consistency when save_best_solutions=True. |
| 127 | + """ |
| 128 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 129 | + num_parents_mating=num_parents_mating, |
| 130 | + fitness_func=fitness_func, |
| 131 | + sol_per_pop=sol_per_pop, |
| 132 | + num_genes=num_genes, |
| 133 | + random_seed=random_seed, |
| 134 | + save_best_solutions=True, |
| 135 | + suppress_warnings=True |
| 136 | + ) |
| 137 | + ga_instance.run() |
| 138 | + |
| 139 | + # Call with last_generation_fitness |
| 140 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 141 | + |
| 142 | + # Call without pop_fitness |
| 143 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 144 | + |
| 145 | + assert numpy.array_equal(sol1, sol2) |
| 146 | + assert fitness1 == fitness2 |
| 147 | + assert idx1 == idx2 |
| 148 | + print("test_best_solution_with_save_best_solutions passed.") |
| 149 | + |
| 150 | +def test_best_solution_with_keep_elitism(): |
| 151 | + """ |
| 152 | + Test best_solution() consistency when keep_elitism > 0. |
| 153 | + """ |
| 154 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 155 | + num_parents_mating=num_parents_mating, |
| 156 | + fitness_func=fitness_func, |
| 157 | + sol_per_pop=sol_per_pop, |
| 158 | + num_genes=num_genes, |
| 159 | + random_seed=random_seed, |
| 160 | + keep_elitism=2, |
| 161 | + suppress_warnings=True |
| 162 | + ) |
| 163 | + ga_instance.run() |
| 164 | + |
| 165 | + # Call with last_generation_fitness |
| 166 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 167 | + |
| 168 | + # Call without pop_fitness |
| 169 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 170 | + |
| 171 | + assert numpy.array_equal(sol1, sol2) |
| 172 | + assert fitness1 == fitness2 |
| 173 | + assert idx1 == idx2 |
| 174 | + print("test_best_solution_with_keep_elitism passed.") |
| 175 | + |
| 176 | +def test_best_solution_with_keep_parents(): |
| 177 | + """ |
| 178 | + Test best_solution() consistency when keep_parents > 0. |
| 179 | + Note: keep_parents is ignored if keep_elitism > 0 (default is 1). |
| 180 | + So this tests the case where keep_parents is passed but effectively ignored by population update, |
| 181 | + yet we check if best_solution() still works consistently. |
| 182 | + """ |
| 183 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 184 | + num_parents_mating=num_parents_mating, |
| 185 | + fitness_func=fitness_func, |
| 186 | + sol_per_pop=sol_per_pop, |
| 187 | + num_genes=num_genes, |
| 188 | + random_seed=random_seed, |
| 189 | + keep_parents=2, |
| 190 | + suppress_warnings=True |
| 191 | + ) |
| 192 | + ga_instance.run() |
| 193 | + |
| 194 | + # Call with last_generation_fitness |
| 195 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 196 | + |
| 197 | + # Call without pop_fitness |
| 198 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 199 | + |
| 200 | + assert numpy.array_equal(sol1, sol2) |
| 201 | + assert fitness1 == fitness2 |
| 202 | + assert idx1 == idx2 |
| 203 | + print("test_best_solution_with_keep_parents passed.") |
| 204 | + |
| 205 | +def test_best_solution_with_keep_parents_elitism_0(): |
| 206 | + """ |
| 207 | + Test best_solution() consistency when keep_parents > 0 and keep_elitism = 0. |
| 208 | + This ensures the 'keep_parents' logic in cal_pop_fitness is exercised. |
| 209 | + """ |
| 210 | + ga_instance = pygad.GA(num_generations=num_generations, |
| 211 | + num_parents_mating=num_parents_mating, |
| 212 | + fitness_func=fitness_func, |
| 213 | + sol_per_pop=sol_per_pop, |
| 214 | + num_genes=num_genes, |
| 215 | + random_seed=random_seed, |
| 216 | + keep_elitism=0, |
| 217 | + keep_parents=2, |
| 218 | + suppress_warnings=True |
| 219 | + ) |
| 220 | + ga_instance.run() |
| 221 | + |
| 222 | + # Call with last_generation_fitness |
| 223 | + sol1, fitness1, idx1 = ga_instance.best_solution(pop_fitness=ga_instance.last_generation_fitness) |
| 224 | + |
| 225 | + # Call without pop_fitness |
| 226 | + sol2, fitness2, idx2 = ga_instance.best_solution() |
| 227 | + |
| 228 | + assert numpy.array_equal(sol1, sol2) |
| 229 | + assert fitness1 == fitness2 |
| 230 | + assert idx1 == idx2 |
| 231 | + print("test_best_solution_with_keep_parents_elitism_0 passed.") |
| 232 | + |
| 233 | +def test_best_solution_pop_fitness_validation(): |
| 234 | + """ |
| 235 | + Test validation of the pop_fitness parameter in best_solution(). |
| 236 | + |
| 237 | + Note: num_generations=1 is used for speed as evolution is not needed. |
| 238 | + sol_per_pop=5 is used to provide a small population for testing invalid lengths. |
| 239 | + """ |
| 240 | + ga_instance = pygad.GA(num_generations=1, |
| 241 | + num_parents_mating=1, |
| 242 | + fitness_func=fitness_func, |
| 243 | + sol_per_pop=5, |
| 244 | + num_genes=3, |
| 245 | + suppress_warnings=True |
| 246 | + ) |
| 247 | + |
| 248 | + # Test invalid type |
| 249 | + try: |
| 250 | + ga_instance.best_solution(pop_fitness="invalid") |
| 251 | + except ValueError as e: |
| 252 | + assert "expected to be list, tuple, or numpy.ndarray" in str(e) |
| 253 | + print("Validation: Invalid type caught.") |
| 254 | + |
| 255 | + # Test invalid length |
| 256 | + try: |
| 257 | + ga_instance.best_solution(pop_fitness=[1, 2, 3]) # Length 3, but sol_per_pop is 5 |
| 258 | + except ValueError as e: |
| 259 | + assert "must match the length of the 'self.population' attribute" in str(e) |
| 260 | + print("Validation: Invalid length caught.") |
| 261 | + |
| 262 | +def test_best_solution_single_objective_tie(): |
| 263 | + """ |
| 264 | + Test best_solution() when there is a tie in fitness values. |
| 265 | + It should return the first solution with the maximum fitness. |
| 266 | +
|
| 267 | + Note: sol_per_pop=5 must match the length of the manual pop_fitness array below. |
| 268 | + num_generations=1 is sufficient for testing selection logic. |
| 269 | + """ |
| 270 | + ga_instance = pygad.GA(num_generations=1, |
| 271 | + num_parents_mating=1, |
| 272 | + fitness_func=fitness_func, |
| 273 | + sol_per_pop=5, |
| 274 | + num_genes=3, |
| 275 | + suppress_warnings=True |
| 276 | + ) |
| 277 | + |
| 278 | + # Mock fitness with a tie at index 1 and 3 |
| 279 | + pop_fitness = numpy.array([10, 50, 20, 50, 5]) |
| 280 | + |
| 281 | + sol, fitness, idx = ga_instance.best_solution(pop_fitness=pop_fitness) |
| 282 | + |
| 283 | + assert fitness == 50 |
| 284 | + assert idx == 1 # First occurrence |
| 285 | + print("test_best_solution_single_objective_tie passed.") |
| 286 | + |
| 287 | +def test_best_solution_with_parallel_processing(): |
| 288 | + """ |
| 289 | + Test best_solution() with parallel_processing enabled. |
| 290 | +
|
| 291 | + Note: num_generations=5 is used to ensure the initial population and first generation |
| 292 | + trigger parallel fitness calculation. |
| 293 | + """ |
| 294 | + ga_instance = pygad.GA(num_generations=5, |
| 295 | + num_parents_mating=2, |
| 296 | + fitness_func=fitness_func, |
| 297 | + sol_per_pop=10, |
| 298 | + num_genes=3, |
| 299 | + random_seed=random_seed, |
| 300 | + parallel_processing=["thread", 2], |
| 301 | + suppress_warnings=True |
| 302 | + ) |
| 303 | + # best_solution() should work and trigger cal_pop_fitness() internally |
| 304 | + sol, fitness, idx = ga_instance.best_solution() |
| 305 | + assert sol is not None |
| 306 | + assert fitness is not None |
| 307 | + print("test_best_solution_with_parallel_processing passed.") |
| 308 | + |
| 309 | +def test_best_solution_with_fitness_batch_size(): |
| 310 | + """ |
| 311 | + Test best_solution() with fitness_batch_size > 1. |
| 312 | +
|
| 313 | + Note: num_generations=5 and sol_per_pop=10 provide enough work for batch processing. |
| 314 | + """ |
| 315 | + def fitness_func_batch(ga_instance, solutions, indices): |
| 316 | + return [numpy.sum(s**2) for s in solutions] |
| 317 | + |
| 318 | + ga_instance = pygad.GA(num_generations=5, |
| 319 | + num_parents_mating=2, |
| 320 | + fitness_func=fitness_func_batch, |
| 321 | + sol_per_pop=10, |
| 322 | + num_genes=3, |
| 323 | + random_seed=random_seed, |
| 324 | + fitness_batch_size=2, |
| 325 | + suppress_warnings=True |
| 326 | + ) |
| 327 | + |
| 328 | + sol, fitness, idx = ga_instance.best_solution() |
| 329 | + assert sol is not None |
| 330 | + assert fitness is not None |
| 331 | + print("test_best_solution_with_fitness_batch_size passed.") |
| 332 | + |
| 333 | +def test_best_solution_pop_fitness_types(): |
| 334 | + """ |
| 335 | + Test best_solution() with different types for the pop_fitness parameter. |
| 336 | +
|
| 337 | + Note: sol_per_pop=3 must match the length of fitness_vals below. |
| 338 | + num_generations=1 is sufficient for this type-check test. |
| 339 | + """ |
| 340 | + ga_instance = pygad.GA(num_generations=1, |
| 341 | + num_parents_mating=1, |
| 342 | + fitness_func=fitness_func, |
| 343 | + sol_per_pop=3, |
| 344 | + num_genes=3, |
| 345 | + suppress_warnings=True |
| 346 | + ) |
| 347 | + |
| 348 | + fitness_vals = [1.0, 5.0, 2.0] |
| 349 | + |
| 350 | + # Test list |
| 351 | + _, _, idx_list = ga_instance.best_solution(pop_fitness=fitness_vals) |
| 352 | + # Test tuple |
| 353 | + _, _, idx_tuple = ga_instance.best_solution(pop_fitness=tuple(fitness_vals)) |
| 354 | + # Test numpy array |
| 355 | + _, _, idx_ndarray = ga_instance.best_solution(pop_fitness=numpy.array(fitness_vals)) |
| 356 | + |
| 357 | + assert idx_list == idx_tuple == idx_ndarray == 1 |
| 358 | + print("test_best_solution_pop_fitness_types passed.") |
| 359 | + |
| 360 | +if __name__ == "__main__": |
| 361 | + test_best_solution_consistency_single_objective() |
| 362 | + test_best_solution_consistency_multi_objective() |
| 363 | + test_best_solution_before_run() |
| 364 | + test_best_solution_with_save_solutions() |
| 365 | + test_best_solution_with_save_best_solutions() |
| 366 | + test_best_solution_with_keep_elitism() |
| 367 | + test_best_solution_with_keep_parents() |
| 368 | + test_best_solution_with_keep_parents_elitism_0() |
| 369 | + test_best_solution_pop_fitness_validation() |
| 370 | + test_best_solution_single_objective_tie() |
| 371 | + test_best_solution_with_parallel_processing() |
| 372 | + test_best_solution_with_fitness_batch_size() |
| 373 | + test_best_solution_pop_fitness_types() |
| 374 | + print("\nAll tests passed!") |
0 commit comments