fix(training): make policy evaluation portable

This commit is contained in:
Josh Creek
2026-09-01 18:56:31 +01:00
parent 066aee96cc
commit e56850a236
3 changed files with 25 additions and 0 deletions
+8
View File
@@ -139,6 +139,14 @@ appends to `training/eval_history.json` — the long-term progress record.
Evaluate each new candidate against the previous promoted bot and a fixed
early reference to see absolute progress over time.
The evaluator launches Godot with the explicit headless display driver,
Compatibility renderer, dummy audio driver, and a temporary writable log path
so evaluation is reproducible on machines where the default renderer or
`user://` log location is unavailable. A two-seed smoke check on 2026-09-01
completed 10 paired episodes per seed for the current Stage 6 export versus
`hard.json` (both runs finished 35 with 2 draws); this is only a runtime
smoke and is not promotion evidence for the Stage 6 gate.
If a model was trained with the locomotion mask on (curriculum stages 1, 2,
and 5 — see below), pass `--grounded-a`/`--grounded-b` for whichever side it's on.
The eval otherwise runs `AIShipController` fully unmasked regardless of how a
+9
View File
@@ -17,6 +17,7 @@ import json
import os
import pathlib
import subprocess
import tempfile
TRAINING_DIR = pathlib.Path(__file__).resolve().parent
GAME_DIR = TRAINING_DIR.parent / "Game"
@@ -37,6 +38,14 @@ def run_half(
) -> dict:
cmd = [
godot_bin,
"--display-driver",
"headless",
"--rendering-method",
"gl_compatibility",
"--audio-driver",
"Dummy",
"--log-file",
str(pathlib.Path(tempfile.gettempdir()) / "cosmic-clash-evaluate-godot.log"),
"--path",
str(GAME_DIR),
TRAINING_SCENE,
+8
View File
@@ -55,6 +55,14 @@ class EvaluatePairTests(unittest.TestCase):
command = run_process.call_args.args[0]
self.assertIn("--eval_team_size=2", command)
@patch("evaluate.subprocess.run")
def test_run_uses_portable_headless_renderer_and_writable_log(self, run_process) -> None:
run_process.return_value.stdout = 'EVAL_RESULT {"episodes": 2, "goals_a": 1, "goals_b": 0, "draws": 1}\n'
evaluate.run_half("godot", "a", "b", 2, 16, 9)
command = run_process.call_args.args[0]
for option in ("--display-driver", "headless", "--rendering-method", "gl_compatibility", "--audio-driver", "Dummy", "--log-file"):
self.assertIn(option, command)
@patch("evaluate.run_half")
def test_2v2_evaluation_preserves_side_swap_and_team_size(self, run_half) -> None:
run_half.side_effect = [