File size: 5,310 Bytes
b68f9f3 5ee4f7e b68f9f3 5ee4f7e b68f9f3 5ee4f7e b68f9f3 5ee4f7e b68f9f3 5ee4f7e b68f9f3 5ee4f7e b68f9f3 | 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 | from __future__ import annotations
import io
import json
import os
import subprocess
import sys
from deberta_ime.evaluation_cli import run
def test_evaluation_cli_import_does_not_require_model_runtime() -> None:
script = """
import builtins
real_import = builtins.__import__
def guarded_import(name, *args, **kwargs):
if name == "torch" or name.startswith("transformers"):
raise AssertionError(f"unexpected model runtime import: {name}")
return real_import(name, *args, **kwargs)
builtins.__import__ = guarded_import
from deberta_ime.evaluation_cli import run
assert callable(run)
"""
environment = {**os.environ, "PYTHONPATH": "src"}
completed = subprocess.run(
[sys.executable, "-c", script],
cwd=os.getcwd(),
env=environment,
capture_output=True,
text=True,
check=False,
)
assert completed.returncode == 0, completed.stderr
def test_evaluation_cli_scores_finite_predictions_and_writes_receipts(tmp_path) -> None:
items_path = tmp_path / "items.json"
predictions_path = tmp_path / "predictions.json"
items_path.write_text(
json.dumps(
[
{
"id": "clean",
"input": "猫です",
"references": ["猫です"],
"label": "clean",
},
{
"id": "typo",
"input": "犬でし",
"references": ["犬です"],
"label": "typo",
},
],
ensure_ascii=False,
),
encoding="utf-8",
)
predictions_path.write_text(
json.dumps(
[
{
"id": "clean",
"candidates": [],
"provenance": "provider",
"reason": "baseline_best",
"margin": 0.0,
},
{
"id": "typo",
"candidates": ["犬です"],
"provenance": "deberta",
"reason": "accepted",
"margin": 1.25,
},
],
ensure_ascii=False,
),
encoding="utf-8",
)
stdout = io.StringIO()
exit_code = run(
[
"--items",
str(items_path),
"--predictions",
str(predictions_path),
"--dataset-name",
"fixture-clean-typo",
"--dataset-revision",
"fixture-v1",
"--dataset-license",
"test-only",
"--output-dir",
str(tmp_path / "outputs"),
"--stem",
"fixture",
],
stdout=stdout,
)
summary = json.loads(stdout.getvalue())
report = json.loads((tmp_path / "outputs" / "fixture.json").read_text("utf-8"))
assert exit_code == 0
assert summary["ok"] is True
assert report["schema_version"] == 2
assert report["status"] == "LOCAL_FINITE_CANDIDATE_EVALUATION"
assert report["evaluation"]["metrics"]["effective_acc_at_1"] == 1.0
assert report["evaluation"]["metrics"]["overcorrection_rate"] == 0.0
assert report["evaluation"]["metrics"]["declared_provenance_counts"] == {
"deberta": 1,
"provider": 1,
}
assert report["evaluation"]["metrics"]["mean_reported_margin"] == 0.625
assert report["dataset"] == {
"name": "fixture-clean-typo",
"revision": "fixture-v1",
"license": "test-only",
}
assert len(report["artifacts"]["items"]["sha256"]) == 64
markdown = (tmp_path / "outputs" / "fixture.md").read_text("utf-8")
assert "fixture-clean-typo" in markdown
assert "Accepted candidate misses: 0" in markdown
assert "Selection errors: 0" in markdown
def test_evaluation_cli_adapts_ajimee_without_inventing_clean_labels(tmp_path) -> None:
items_path = tmp_path / "ajimee.json"
predictions_path = tmp_path / "predictions.json"
items_path.write_text(
json.dumps(
[
{
"index": "7",
"input": "セイネンシ",
"expected_output": ["青年誌"],
}
],
ensure_ascii=False,
),
encoding="utf-8",
)
predictions_path.write_text(
json.dumps([{"index": "7", "candidates": ["青年誌"]}], ensure_ascii=False),
encoding="utf-8",
)
run(
[
"--items",
str(items_path),
"--predictions",
str(predictions_path),
"--format",
"ajimee",
"--output-dir",
str(tmp_path / "outputs"),
],
stdout=io.StringIO(),
)
report = json.loads(
(tmp_path / "outputs" / "finite_candidate_evaluation.json").read_text("utf-8")
)
assert report["evaluation"]["metrics"]["effective_acc_at_1"] == 1.0
assert report["evaluation"]["metrics"]["overcorrection_rate"] is None
assert report["evaluation"]["metrics"]["clean_rows"] == 0
assert report["dataset"] == {
"name": "AJIMEE-compatible input",
"revision": "unverified-by-sha256",
"license": "unspecified",
}
|