Replacing Monkey-Patching by PavelMarian · Pull Request #1421 · aimclub/FEDOT

PavelMarian · 2026-03-05T09:29:10Z

Summary

Introduced ExecutionContext and IndustrialExtension to switch base and industrial context
Provided serialization through __getstate__ and __setstate__ within ExecutionContext
Modified initializer_idustrial_models.py to apply changes

Next Steps

Test implemetations
Fix imports
Ensure there is no global side effects

Context

github-actions · 2026-03-05T09:30:36Z

Code in this pull request still contains PEP8 errors, please write the /fix-pep8 command in the comments below to create commit with automatic fixes.

Comment last updated at Fri, 29 May 2026 08:52:30

Lopa10ko · 2026-03-05T15:24:03Z

/fix-pep8

Lopa10ko

Main concern: The new context is not yet driving behavior – most of FEDOT/Industrial still relies on global singletons and direct imports, so the extension system is more of a configuration holder than a true execution context.

We want to create “no global side effects; context is passed explicitly” and (optional) protocols per extension point; those parts are currently incomplete.

No unit tests for the context yet :(

Lopa10ko · 2026-03-10T13:54:16Z

+import golem.core.tuning.optuna_tuner as OptunaImpl
+
+
+class ExecutionContext:


ExecutionContext is created and mutated (via IndustrialExtension.apply), but it is not passed to the places that actually perform evaluation, splitting, tuning, etc.

all existing usages of IndustrialModels still look like:

self.repo = IndustrialModels().setup_repository() with IndustrialModels(): ...

these mutate OperationTypesRepository.__repository_dict__ and then continue to use the global/static APIs instead of going through the context attributes.

Lopa10ko · 2026-03-10T13:54:51Z

+
+    def __getstate__(self):
+        return {
+            "industrial": True,


where does the context and IndustrialModels actually used in FEDOT API (not in FedotIndustrial API)?

context != industrial context, context might be non-industrial also.

Lopa10ko · 2026-03-10T14:00:49Z

+
+class ExecutionContext:
+    def __init__(self) -> None:
+        self.operation_registry = OperationTypesRepository()


(nit) here design is very concrete and tightly coupled

maybe you can try to introduce protocols or small interfaces for major extension points, for example:
OperationRegistryProtocol
MergerProtocol
SplitterProtocol
TunerFactoryProtocol / TunerProtocol

(nit) where possible, avoid eager instantiation of heavy components in __init__, instead use lazy properties or factories to avoid performance penalties and to make constructing a context cheaper

still an issue

Lopa10ko · 2026-03-10T14:03:12Z

+
+    def __getstate__(self):
+        return {
+            "industrial": True,


the industrial flag is stored but never used or set in __setstate__.

IMPORTANT: also any backend‑specific overrides made by IndustrialExtension.apply are lost upon unpickling, because __init__ recreates a default (non‑industrial) context.

Lopa10ko · 2026-03-10T14:05:39Z

+        context.topo_features_fit = fit_topo_extractor_industrial
+        context.topo_features_transform = transform_topo_extractor_industrial
+        context.ts_smoothing_transform = transform_smoothing_industrial
+        context.optuna_optuna_tuner = DaskOptunaTuner


remove this duplicated line

see the 33rd line here:

context.optuna_optuna_tuner = DaskOptunaTuner if "dask" in self.backend else OptunaTuner

Suggested change

context.optuna_optuna_tuner = DaskOptunaTuner

Lopa10ko · 2026-03-10T14:07:06Z

-                get_industrial_search_space)
-        setattr(ApiComposer, "_get_default_mutations",
-                _get_default_industrial_mutations)
+    def __enter__(self) -> ExecutionContext:


this now returns an ExecutionContext, but previously the context manager was used only for side effects (switching repos and monkey patching).

many call sites still use with IndustrialModels(): and ignore the returned value

Lopa10ko · 2026-03-10T14:08:32Z

-        self._replace_operation(to_industrial=False, backend=backend)
-        common_rules.append(has_no_resample)
+        self.context = ExecutionContext()
+        self.context.common_rules.append(has_no_resample)


this creates a new ExecutionContext every time and sets self.context.common_rules.append(has_no_resample)

this might not match expectations where callers think they are "restoring" a default state

Lopa10ko · 2026-03-10T14:12:07Z

-                       'default_tags': []}})
-        OperationTypesRepository.assign_repo('model', self.base_model_path)
+        self.setup_default_repository()
+        self.context = None


shouldn't it restore the context to FEDOT models along with context?
in setup_default_repository you create a fresh context, but after that call there is self.context = None

Lopa10ko · 2026-04-01T10:37:31Z

+
+class ExecutionContext:
+    def __init__(self) -> None:
+        self.operation_registry = OperationTypesRepository()


still an issue

Lopa10ko · 2026-04-01T10:40:29Z

                 eval_n_jobs: int = 1,
-                 do_unfit: bool = True):
+                 do_unfit: bool = True,
+                 context: Optional[ExecutionContext] = None):


why pass context here?
it isn't used in this constructor or the pipeline evaluation methods

Lopa10ko · 2026-04-01T10:46:22Z


 class LaggedImplementation(DataOperationImplementation):
-    def __init__(self, params: Optional[OperationParameters]):
+    def __init__(self, params: Optional[OperationParameters], context: Optional[ExecutionContext] = None):


why pass context into this operation alone?

the context requires refactoring into smaller, more manageable parts (for example, the lagged context or the data operation context would be more appropriate for specific operations that need certain overrides - not all overrides are necessary in these cases)

Lopa10ko · 2026-04-01T10:47:25Z

+        split_func_dict = {DataTypesEnum.multi_ts: context.data_split__split_time_series,
+                           DataTypesEnum.ts: context.data_split__split_time_series,
+                           DataTypesEnum.table: context.data_split__split_any,
+                           DataTypesEnum.image: context.data_split__split_any,
+                           DataTypesEnum.text: context.data_split__split_any}


data split context - should be separate object within context or within context extension

Lopa10ko · 2026-04-01T10:48:11Z

@@ -35,7 +38,8 @@ def __init__(self, outputs: List['OutputData'], data_type: DataTypesEnum = None)
            raise ValueError('There are no common indices for outputs')

        # Find first output with the main target & resulting task
-        self.main_output = DataMerger.find_main_output(outputs)
+        # self.main_output = DataMerger.find_main_output(outputs)
+        self.main_output = self.context.merger_find_main_output(outputs)

    @staticmethod
    def get(outputs: List['OutputData']) -> 'DataMerger':
@@ -71,7 +75,8 @@ def merge(self) -> 'InputData':

        common_predicts = self.find_common_predicts()
        mergeable_predicts = self.preprocess_predicts(common_predicts)
-        merged_features = self.merge_predicts(mergeable_predicts)
+        # merged_features = self.merge_predicts(mergeable_predicts)
+        merged_features = self.context.merger_merge_predicts(mergeable_predicts)


data merger context within context or extension

Lopa10ko · 2026-04-23T14:53:04Z

+
+
+def industrial_context_factory(backend: str = "default"):
+    return IndustrialContext(backend=backend)


в импорте этого нет, появляется ошибка IndustrialContext is not defined.

Используется ли этот метод вообще? Почему этот и другие методы здесь являются фабриками (если хочется использовать паттерн фабрики - пожалуйста, но здесь он реализован не как фабрика) и зачем они здесь, если они просто создают инстансы классов (причем с параметрами по умолчанию в конструкторах)?

манифест регистрирует эти методы как протоколы, протоколы отдают конкретные имлементации (напр. IndustrialSplitter), эти классы переопределяют нужные методы (внутри вызывают логику из индастриала) вместо их патчинга, так?

зачем эти ненужные добавочные слои в виде классов вида IndustrialSplitter, как и где это подгружается в контекст? есть контекст, в нем должно быть назначение этих методов (дефолтное FEDOT назначение работает), а как работает для Industrial?

см. дальше замечания по контексту в context.py

опиши, пожалуйста, полный цикл жизни extension

Lopa10ko · 2026-04-23T15:17:54Z

+                                           topo_features_factory, ts_smoothing_factory, api_composer_tune_factory)
+
+
+FEDOT_INDUSTRIAL_MANIFEST = ExtensionManifest(


верно понимаю, что это общая входная точка для патченых методов?
как планируется использовать это?

Lopa10ko · 2026-04-23T15:20:58Z

@@ -0,0 +1,109 @@
+import Protocol


Import "Protocol" could not be resolved....

Lopa10ko · 2026-04-23T15:31:21Z

+
+    def _apply_protocols(self):
+        # Splitters
+        splitters = resolve_protocol_instance("splitters", backend=self.backend)


где этот метод?

Lopa10ko · 2026-04-23T15:33:57Z

+    if context_name == "core":
+        return ExecutionContext(backend=backend)
+
+    from fedot.extensions.registry import get_registered_extension, get_registered_extensions


а Industrial extension где-то регистрируется?

added context execution

a60b550

PavelMarian requested a review from Lopa10ko March 5, 2026 09:29

udate context

ed7ac9b

Lopa10ko assigned Lopa10ko and PavelMarian and unassigned Lopa10ko Mar 5, 2026

Automated autopep8 fixes

bf06bdf

Lopa10ko requested changes Mar 10, 2026

View reviewed changes

PavelMarian changed the title ~~Replacing Mokey-Patching~~ Replacing Monkey-Patching Mar 12, 2026

PavelMarian added 2 commits March 19, 2026 17:35

forwared context flow

60c10d3

merged trees

6fb6cdd

Lopa10ko requested changes Apr 1, 2026

View reviewed changes

PavelMarian added 3 commits April 21, 2026 15:52

introduced industrial context manifest & protocols

9fcb7ac

Merge branch 'refactor/industrial-synergy' into execution-context

b76f2f9

Delete fedot/core/context.py

7b565f5

Lopa10ko requested changes Apr 23, 2026

View reviewed changes

PavelMarian added 3 commits May 15, 2026 03:07

fixed execution context

2350269

fixed

8036b2a

removed default_backend

2a4018b

PavelMarian force-pushed the execution-context branch from b4791f7 to 2a4018b Compare May 28, 2026 22:45

PavelMarian added 6 commits May 29, 2026 01:46

Delete test/unit/context/test_initialization.py

751f9e5

Delete test/unit/data/test_supplementary_data.py

394eabf

Delete test/integration/pipelines/tuning/test_pipeline_tuning.py

87e2a52

Delete fedot/industrial/industrial_extension.py

56e8a2b

Delete fedot/core/context/default_backend.py

5988ce6

updated main.py

f492f8b

		import golem.core.tuning.optuna_tuner as OptunaImpl


		class ExecutionContext:



		def industrial_context_factory(backend: str = "default"):
		return IndustrialContext(backend=backend)

		topo_features_factory, ts_smoothing_factory, api_composer_tune_factory)


		FEDOT_INDUSTRIAL_MANIFEST = ExtensionManifest(

Uh oh!

Conversation

PavelMarian commented Mar 5, 2026

Summary

Next Steps

Context

Uh oh!

github-actions Bot commented Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Comment last updated at Fri, 29 May 2026 08:52:30

Uh oh!

Lopa10ko commented Mar 5, 2026

Uh oh!

Lopa10ko left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Mar 5, 2026 •

edited

Loading