| from __future__ import annotations |
|
|
| import sys |
| import unittest |
| from pathlib import Path |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| sys.path.insert(0, str(ROOT / "src")) |
|
|
| from physics_model.dynamic_sampling import ( |
| ALL_MAX, |
| OUTCOME_VARIATION, |
| PROCESS_ONLY, |
| SILENT, |
| apply_dynamic_sampling, |
| classify_group, |
| ) |
| from physics_sft.grpo_reward import RewardConfig |
|
|
|
|
| class DynamicSamplingTests(unittest.TestCase): |
| def test_process_only_variation_is_kept(self) -> None: |
| totals = [1.10, 1.25, 1.10, 1.10] |
| outcomes = [1.00, 1.00, 1.00, 1.00] |
| processes = [0.10, 0.25, 0.10, 0.10] |
| label = classify_group( |
| totals, |
| outcome_rewards=outcomes, |
| process_rewards=processes, |
| max_total=1.75, |
| use_process=True, |
| ) |
| self.assertEqual(label, PROCESS_ONLY) |
| result = apply_dynamic_sampling( |
| [totals], |
| outcome_groups=[outcomes], |
| process_groups=[processes], |
| max_total=1.75, |
| use_process=True, |
| ) |
| self.assertEqual(result.keep_indices, [0]) |
| self.assertEqual(result.labels[0], PROCESS_ONLY) |
| self.assertFalse(result.resample) |
|
|
| def test_all_equal_total_is_filtered(self) -> None: |
| totals = [0.25, 0.25, 0.25, 0.25] |
| result = apply_dynamic_sampling( |
| [totals], |
| outcome_groups=[totals], |
| process_groups=[[0, 0, 0, 0]], |
| max_total=1.75, |
| use_process=True, |
| resample_cap=3, |
| ) |
| self.assertEqual(result.labels[0], SILENT) |
| self.assertEqual(result.keep_indices, []) |
| self.assertTrue(result.resample) |
|
|
| def test_all_max_uses_configured_maximum(self) -> None: |
| cfg = RewardConfig(use_cot_process_reward=True) |
| totals = [cfg.max_total_reward] * 4 |
| label = classify_group(totals, max_total=cfg.max_total_reward, use_process=True) |
| self.assertEqual(label, ALL_MAX) |
| other = classify_group(totals, max_total=9.99, use_process=True) |
| self.assertEqual(other, SILENT) |
|
|
| def test_disable_process_uses_outcome_only(self) -> None: |
| totals = [1.10, 1.25, 1.10, 1.10] |
| outcomes = [1.00, 1.00, 1.00, 1.00] |
| processes = [0.10, 0.25, 0.10, 0.10] |
| result = apply_dynamic_sampling( |
| [totals], |
| outcome_groups=[outcomes], |
| process_groups=[processes], |
| max_total=1.25, |
| use_process=False, |
| ) |
| self.assertEqual(result.labels[0], SILENT) |
| self.assertEqual(result.keep_indices, []) |
|
|
| mixed_outcome = apply_dynamic_sampling( |
| [[1.0, 1.25, 1.0, 1.0]], |
| outcome_groups=[[1.0, 1.25, 1.0, 1.0]], |
| process_groups=[[0, 0, 0, 0]], |
| max_total=1.25, |
| use_process=False, |
| ) |
| self.assertEqual(mixed_outcome.labels[0], OUTCOME_VARIATION) |
| self.assertEqual(mixed_outcome.keep_indices, [0]) |
|
|
| def test_resample_cap_exhausts(self) -> None: |
| silent = [[0.0, 0.0, 0.0, 0.0]] |
| mid = apply_dynamic_sampling(silent, max_total=1.25, resample_attempt=1, resample_cap=3) |
| self.assertTrue(mid.resample) |
| self.assertFalse(mid.cap_exhausted) |
| done = apply_dynamic_sampling(silent, max_total=1.25, resample_attempt=3, resample_cap=3) |
| self.assertFalse(done.resample) |
| self.assertTrue(done.cap_exhausted) |
| self.assertEqual(done.keep_indices, []) |
|
|
| def test_short_batch_does_not_raise(self) -> None: |
| short = [[0.2, 0.8]] |
| result = apply_dynamic_sampling(short, max_total=1.25, use_process=False) |
| self.assertEqual(result.keep_indices, [0]) |
| leftover = apply_dynamic_sampling([[0.1]], max_total=1.25) |
| self.assertEqual(leftover.labels[0], SILENT) |
|
|
|
|
| if __name__ == "__main__": |
| unittest.main() |
|
|
