Blame view

scripts/evaluation/eval_framework/__init__.py 1.48 KB
c81b0fc1   tangwang   scripts/evaluatio...
1
2
3
4
5
6
7
8
9
10
11
12
13
14
  """
  Search evaluation framework: pooled relevance annotation, live metrics, batch reports.
  
  Importing this package ensures the project root is on ``sys.path`` (for ``api.*`` imports).
  """
  
  from __future__ import annotations
  
  from .utils import ensure_project_on_path
  
  ensure_project_on_path()
  
  from .constants import (  # noqa: E402
      DEFAULT_ARTIFACT_ROOT,
c81b0fc1   tangwang   scripts/evaluatio...
15
      DEFAULT_QUERY_FILE,
c81b0fc1   tangwang   scripts/evaluatio...
16
      PROJECT_ROOT,
d73ca84a   tangwang   refine eval case ...
17
18
19
20
      RELEVANCE_LV0,
      RELEVANCE_LV1,
      RELEVANCE_LV2,
      RELEVANCE_LV3,
a345b01f   tangwang   eval framework
21
      RELEVANCE_NON_IRRELEVANT,
c81b0fc1   tangwang   scripts/evaluatio...
22
      VALID_LABELS,
c81b0fc1   tangwang   scripts/evaluatio...
23
24
25
26
  )
  from .framework import SearchEvaluationFramework  # noqa: E402
  from .store import EvalStore, QueryBuildResult  # noqa: E402
  from .cli import build_cli_parser, main  # noqa: E402
2059d959   tangwang   feat(eval): 多评估集统...
27
  from .datasets import EvalDatasetSnapshot, resolve_dataset  # noqa: E402
c81b0fc1   tangwang   scripts/evaluatio...
28
29
30
31
32
33
34
35
36
37
38
  from .web_app import create_web_app  # noqa: E402
  from .reports import render_batch_report_markdown  # noqa: E402
  from .utils import (  # noqa: E402
      ensure_dir,
      sha1_text,
      utc_now_iso,
      utc_timestamp,
  )
  
  __all__ = [
      "DEFAULT_ARTIFACT_ROOT",
c81b0fc1   tangwang   scripts/evaluatio...
39
      "DEFAULT_QUERY_FILE",
2059d959   tangwang   feat(eval): 多评估集统...
40
      "EvalDatasetSnapshot",
c81b0fc1   tangwang   scripts/evaluatio...
41
      "EvalStore",
c81b0fc1   tangwang   scripts/evaluatio...
42
43
      "PROJECT_ROOT",
      "QueryBuildResult",
d73ca84a   tangwang   refine eval case ...
44
45
46
47
      "RELEVANCE_LV0",
      "RELEVANCE_LV1",
      "RELEVANCE_LV2",
      "RELEVANCE_LV3",
a345b01f   tangwang   eval framework
48
      "RELEVANCE_NON_IRRELEVANT",
c81b0fc1   tangwang   scripts/evaluatio...
49
50
51
52
53
54
      "SearchEvaluationFramework",
      "VALID_LABELS",
      "build_cli_parser",
      "create_web_app",
      "ensure_dir",
      "main",
c81b0fc1   tangwang   scripts/evaluatio...
55
      "render_batch_report_markdown",
2059d959   tangwang   feat(eval): 多评估集统...
56
      "resolve_dataset",
c81b0fc1   tangwang   scripts/evaluatio...
57
58
59
60
      "sha1_text",
      "utc_now_iso",
      "utc_timestamp",
  ]