File size: 14,994 Bytes
e2f8b29 45eee48 e2f8b29 02e58fe 45eee48 e2f8b29 56b42b0 e2f8b29 56b42b0 e2f8b29 56b42b0 e2f8b29 56b42b0 e2f8b29 56b42b0 e2f8b29 56b42b0 e2f8b29 45eee48 02e58fe 45eee48 02e58fe 45eee48 02e58fe 45eee48 02e58fe 45eee48 02e58fe 45eee48 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 | """Test grader functions — each returns 0.0-1.0."""
from __future__ import annotations
import pytest
from ml_training_debugger.graders import (
_submitted_diagnosis,
grade_episode,
grade_task_001,
grade_task_003,
grade_task_005,
grade_task_006,
grade_task_007,
)
from ml_training_debugger.models import EpisodeState
from ml_training_debugger.scenarios import sample_scenario
@pytest.fixture
def scenario_001():
return sample_scenario("task_001", seed=42)
@pytest.fixture
def scenario_003():
return sample_scenario("task_003", seed=42)
@pytest.fixture
def scenario_005():
return sample_scenario("task_005", seed=42)
class TestGradeTask001:
def test_perfect_score(self, scenario_001):
state = EpisodeState(
gradients_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"modify_config",
"restart_run",
"mark_diagnosed:lr_too_high",
],
)
score = grade_task_001(state, scenario_001)
assert score == 1.0
def test_wrong_diagnosis(self, scenario_001):
state = EpisodeState(
gradients_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"modify_config",
"restart_run",
"mark_diagnosed:data_leakage",
],
)
score = grade_task_001(state, scenario_001)
assert score < 0.7 # Missing diagnosis credit
def test_no_investigation(self, scenario_001):
state = EpisodeState(
diagnosis_submitted=True,
actions_taken=["mark_diagnosed:lr_too_high"],
)
score = grade_task_001(state, scenario_001)
assert 0.0 < score < 1.0
def test_score_in_range(self, scenario_001):
state = EpisodeState()
score = grade_task_001(state, scenario_001)
assert 0.0 <= score <= 1.0
class TestGradeTask003:
def test_perfect_score(self, scenario_003):
state = EpisodeState(
data_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_data_batch",
"patch_data_loader",
"restart_run",
"mark_diagnosed:data_leakage",
],
)
score = grade_task_003(state, scenario_003)
assert score == pytest.approx(1.0)
def test_wrong_diagnosis(self, scenario_003):
state = EpisodeState(
data_inspected=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_data_batch",
"mark_diagnosed:overfitting",
],
)
score = grade_task_003(state, scenario_003)
assert score < 0.5
class TestGradeTask005:
def test_perfect_score_thorough(self, scenario_005):
"""Thorough agent inspects weights AND data — gets perfect score."""
state = EpisodeState(
gradients_inspected=True,
gradients_were_normal=True,
model_modes_inspected=True,
model_weights_inspected=True,
data_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"inspect_model_weights",
"inspect_model_modes",
"fix_model_mode",
"restart_run",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_005(state, scenario_005)
assert score == pytest.approx(1.0)
def test_quick_fix_partial_credit(self, scenario_005):
"""Agent that skips weight inspection gets partial credit."""
state = EpisodeState(
gradients_inspected=True,
gradients_were_normal=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_model_modes",
"fix_model_mode",
"restart_run",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_005(state, scenario_005)
# No weights, no data → 0.05+0.05+0.08+0.10+0.40 = 0.68
assert score == pytest.approx(0.68)
def test_red_herring_chaser(self, scenario_005):
"""Agent that chases gradient red herring loses 0.20."""
state = EpisodeState(
gradients_inspected=True,
gradients_were_normal=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"add_callback", # Wrong: chases red herring
"inspect_model_modes",
"fix_model_mode",
"restart_run",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_005(state, scenario_005)
# -0.20 penalty for add_callback after normal gradients
assert score == pytest.approx(0.48)
def test_wrong_fix_penalty(self, scenario_005):
"""Agent that tries modify_config (wrong fix) gets penalized."""
state = EpisodeState(
gradients_inspected=True,
gradients_were_normal=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"modify_config", # Wrong: LR isn't the problem
"inspect_model_modes",
"fix_model_mode",
"restart_run",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_005(state, scenario_005)
# -0.10 penalty for modify_config on task 5
assert score == pytest.approx(0.58)
def test_double_trap_devastates_score(self, scenario_005):
"""Agent that falls for BOTH traps (add_callback + modify_config) scores poorly."""
state = EpisodeState(
gradients_inspected=True,
gradients_were_normal=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"add_callback",
"modify_config",
"inspect_model_modes",
"fix_model_mode",
"restart_run",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_005(state, scenario_005)
# -0.20 (add_callback) + -0.10 (modify_config) = -0.30 penalties
assert score == pytest.approx(0.38)
class TestGradeEpisode:
def test_dispatch_to_correct_grader(self, scenario_001):
state = EpisodeState(
gradients_inspected=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"mark_diagnosed:lr_too_high",
],
)
score = grade_episode("task_001", state, scenario_001)
assert 0.0 <= score <= 1.0
def test_unknown_task_returns_zero(self, scenario_001):
state = EpisodeState()
score = grade_episode("task_999", state, scenario_001)
assert score == 0.0
class TestGradeTask006:
@pytest.fixture
def scenario_006(self):
return sample_scenario("task_006", seed=42)
def test_perfect_score_thorough(self, scenario_006):
"""Thorough agent inspects ALL systems including weights — gets perfect score."""
state = EpisodeState(
code_inspected=True,
gradients_inspected=True,
model_modes_inspected=True,
model_weights_inspected=True,
data_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"inspect_model_weights",
"inspect_model_modes",
"inspect_code",
"fix_code",
"restart_run",
"mark_diagnosed:code_bug",
],
)
score = grade_task_006(state, scenario_006)
assert score == pytest.approx(1.0)
def test_no_weights_inspection_partial(self, scenario_006):
"""Agent that skips weight inspection gets reduced fix/restart credit."""
state = EpisodeState(
code_inspected=True,
gradients_inspected=True,
model_modes_inspected=True,
data_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"inspect_model_modes",
"inspect_code",
"fix_code",
"restart_run",
"mark_diagnosed:code_bug",
],
)
score = grade_task_006(state, scenario_006)
# 0.05*4 + 0.08 + 0.08 + 0.45 = 0.81
assert score == pytest.approx(0.81)
assert score < 1.0 # Must not be perfect without weights
def test_minimal_investigation(self, scenario_006):
"""Agent that only inspects code, fixes, and diagnoses."""
state = EpisodeState(
code_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_code",
"fix_code",
"restart_run",
"mark_diagnosed:code_bug",
],
)
score = grade_task_006(state, scenario_006)
# 0.05 + 0.08 + 0.08 + 0.45 = 0.66
assert score == pytest.approx(0.66)
def test_wrong_diagnosis(self, scenario_006):
"""Submitting batchnorm_eval_mode on a code_bug task fails."""
state = EpisodeState(
code_inspected=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_code",
"mark_diagnosed:batchnorm_eval_mode",
],
)
score = grade_task_006(state, scenario_006)
assert score < 0.2 # Only gets code_inspected bonus
def test_score_in_range(self, scenario_006):
state = EpisodeState()
score = grade_task_006(state, scenario_006)
assert 0.0 <= score <= 1.0
class TestGradeTask007:
def test_perfect_score_thorough(self):
"""Thorough agent inspects weights — gets perfect score."""
scenario = sample_scenario("task_007", seed=42)
state = EpisodeState(
gradients_inspected=True,
data_inspected=True,
model_weights_inspected=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"inspect_model_weights",
"inspect_model_modes",
"modify_config",
"restart_run",
"mark_diagnosed:scheduler_misconfigured",
],
)
score = grade_task_007(state, scenario)
assert score == pytest.approx(1.0)
def test_no_weights_partial(self):
"""Agent without weight inspection gets reduced fix/restart credit."""
scenario = sample_scenario("task_007", seed=42)
state = EpisodeState(
gradients_inspected=True,
data_inspected=True,
model_modes_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"inspect_model_modes",
"modify_config",
"restart_run",
"mark_diagnosed:scheduler_misconfigured",
],
)
score = grade_task_007(state, scenario)
# 0.05*3 + 0.12 + 0.12 + 0.40 = 0.79
assert score == pytest.approx(0.79)
assert score < 1.0
def test_wrong_fix_penalty(self):
"""Agent that patches data loader (wrong fix) gets penalized."""
scenario = sample_scenario("task_007", seed=42)
state = EpisodeState(
gradients_inspected=True,
data_inspected=True,
fix_action_taken=True,
restart_after_fix=True,
diagnosis_submitted=True,
actions_taken=[
"inspect_gradients",
"inspect_data_batch",
"patch_data_loader",
"modify_config",
"restart_run",
"mark_diagnosed:scheduler_misconfigured",
],
)
score = grade_task_007(state, scenario)
# Normal partial score minus 0.10 penalty
assert score < 0.75
def test_wrong_diagnosis(self):
scenario = sample_scenario("task_007", seed=42)
state = EpisodeState(
diagnosis_submitted=True,
actions_taken=["mark_diagnosed:overfitting"],
)
score = grade_task_007(state, scenario)
assert score < 0.5
def test_score_in_range(self):
scenario = sample_scenario("task_007", seed=42)
state = EpisodeState()
score = grade_task_007(state, scenario)
assert 0.0 <= score <= 1.0
class TestSubmittedDiagnosis:
def test_finds_diagnosis(self):
state = EpisodeState(
actions_taken=["inspect_gradients", "mark_diagnosed:lr_too_high"],
)
assert _submitted_diagnosis(state) == "lr_too_high"
def test_no_diagnosis(self):
state = EpisodeState(actions_taken=["inspect_gradients"])
assert _submitted_diagnosis(state) is None
def test_latest_diagnosis(self):
state = EpisodeState(
actions_taken=[
"mark_diagnosed:overfitting",
"mark_diagnosed:lr_too_high",
],
)
assert _submitted_diagnosis(state) == "lr_too_high"
|