Implement #334: Integrate cross-project experiment into pipeline #346
Conversation
|
|
||
| class CrossProjectPrepareTask: | ||
| MAX_SUBTYPES_SAMPLE_SIZE = 25 | ||
| MAX_PROJECT_SAMPLE_SIZE = 50 |
There was a problem hiding this comment.
Move these two constants to config parameters. Use same names.
| @@ -238,6 +249,32 @@ def __add_run_ex3_subprocess(available_detectors: List[str], available_datasets: | |||
| __setup_run_arguments(experiment_parser, available_detectors) | |||
|
|
|||
|
|
|||
| def __add_run_cross_project_create_index(available_datasets: List[str], subparsers) -> None: | |||
There was a problem hiding this comment.
Implicitly run create index with prepare.
| __setup_filter_arguments(parser, available_datasets) | ||
|
|
||
|
|
||
| def __add_run_cross_project_prepare(subparsers) -> None: |
There was a problem hiding this comment.
Provide as top-level task ./mubench checkout-xp.
M8is
force-pushed
the
cross-project-experiment
branch
from
e447a83
to
4cd544b
Compare
M8is
force-pushed
the
cross-project-experiment
branch
from
24009e2
to
bd16a52
Compare
|
Try collecting results from AsterskTasks for passing prepared xp data on to subsequent tasks. If this does not work (easily), we persist indexes version-wise and load these after preparation. |
|
Turns out we really didn't need anything more since we can write one index per version. Woops. Well, now that it's implemented we might as well keep it. It isn't that complicated and seems to be a useful feature. Essentially, we can now split something up into arbitrarily small tasks and just accumulate the results. At this point, I'm wondering if Anyway, we can now run ex2 and ex3 |
|
@salsolatragus I just remembered that the solution I've implemented now is exactly what you suggested. There's definitely a lot of space for improvements, but it should work for now. |
| return { | ||
| key_detector_mode: DetectAllFindingsTask.__DETECTOR_MODE, | ||
| key_target_src_paths: version_compile.original_sources_paths, | ||
| key_target_classes_paths: version_compile.original_classes_paths, | ||
| key_dependency_classpath: version_compile.get_full_classpath() | ||
| key_dependency_classpath: version_compile.get_full_classpath(), | ||
| key_training_src_path: xp_sources_paths |
There was a problem hiding this comment.
What happens if there is no xp paths.
| @@ -40,7 +47,7 @@ def _get_findings_path(self, detector: Detector, version: ProjectVersion, misuse | |||
| def _get_detector_arguments(version_compile: VersionCompile, misuse_compile: MisuseCompile): | |||
| return { | |||
| key_detector_mode: DetectProvidedCorrectUsagesTask.__DETECTOR_MODE, | |||
| key_training_src_path: misuse_compile.correct_usage_sources_path, | |||
| key_training_src_path: [misuse_compile.correct_usage_sources_path], | |||
|
This should be ready now. |
salsolatragus
force-pushed
the
master
branch
3 times, most recently
from
8f1f1a5
to
edd7d88
Compare
Implement #334: Integrate cross-project experiment into pipeline
@salsolatragus I thought you may want to check the current state. The tasks are now integrated into the pipeline and should work the same way as before (plus pipeline features like the default config). I've set them up to be called separately, but that is easy to reconfigure.