[Improvement] Support for candidate multiple dimensional search constraints. by aptsunny · Pull Request #316 · open-mmlab/mmrazor

aptsunny · 2022-10-11T07:13:39Z

Motivation

Continued improvements of EvolutionSearchLoop throught multi-resources constraints.

Use cases (Optional)

Modify the search configs file as follow.

flops_range=(0., 330.) ->  constraints_range=dict(flops=(0., 330.)),

train_cfg = dict(
    _delete_=True,
    type='mmrazor.EvolutionSearchLoop',
    dataloader=_base_.val_dataloader,
    evaluator=_base_.val_evaluator,
    max_epochs=20,
    num_candidates=20,
    top_k=10,
    num_mutation=5,
    num_crossover=5,
    mutate_prob=0.2,
    constraints_range=dict(params=(0, 55), flops=(0., 330.)),
    dump_derived_mutable=True,
    score_key='accuracy/top1')

Checklist

After PR:

The search configuration file for the search algorithm needs to be changed. Please fix it

gaoyang07 · 2022-10-11T07:20:38Z

mmrazor/engine/runner/evolution_search_loop.py


        self.top_k_candidates = Candidates()
-        if resource_estimator_cfg is None:
+        if resource_estimator_cfg:


Should it be if resource_estimator_cfg is None?

default resource_estimator_cfg is dict.

Then directly use self.estimator = ResourceEstimator(**resource_estimator_cfg)

gaoyang07 · 2022-10-11T07:24:17Z

mmrazor/engine/runner/evolution_search_loop.py

            crossover_candidate = self._crossover()

-            if self._check_constraints(random_subnet=crossover_candidate):
+            flag = False


reify the flag here as it suits the particular judgment of model constraints.

gaoyang07 · 2022-10-11T07:30:06Z

mmrazor/structures/subnet/candidate.py

+        any to tuple(any, float, float)."""

        def _format_item(item: Any):
            """Transform any to tuple(any, float)."""


modify docstring here.

gaoyang07 · 2022-10-13T08:02:05Z

mmrazor/engine/runner/utils/check.py

    assert hasattr(model, 'set_subnet') and hasattr(model, 'architecture')
    model.set_subnet(subnet)
-    fix_mutable = export_fix_subnet(model)
+    fix_mutable = export_fix_subnet(model, dump_derived)


remove dump_derived

gaoyang07 · 2022-10-13T08:02:22Z

mmrazor/engine/runner/evolution_search_loop.py

            crossover_candidate = self._crossover()

-            if self._check_constraints(random_subnet=crossover_candidate):
+            flag = False


rename flag

sunnyxiaohu · 2022-10-13T09:46:38Z

update type-hint for Candidate::append, extend, insert. It seems that they must accept Dict as input.

sunnyxiaohu · 2022-10-12T07:03:23Z

mmrazor/engine/runner/subnet_sampler_loop.py

        return subnet

-    def _check_constraints(self, random_subnet: SupportRandomSubnet) -> bool:
+    def _check_constraints(self,


update docstring

sunnyxiaohu · 2022-10-13T08:54:05Z

mmrazor/engine/runner/evolution_search_loop.py

+        constraints_range (Dict[str, Any]): Constraints to be used for
+            screening candidates. Defaults to dict(flops=(0, 330)).
+        resource_estimator_cfg (Dict[str, Any]): Used for building a
+            resource estimator. Default to dict().


type: Default -> Defaults

sunnyxiaohu · 2022-10-13T09:18:35Z

mmrazor/structures/subnet/candidate.py

-    """The data structure of sampled candidate. The format is [(any, float),
-    (any, float), ...].
+    """The data structure of sampled candidate. The format is Union[Dict[Any,
+    Dict], List[Dict[Any, Dict]]].


The format is List[Dict[Any, Dict]]

mmrazor/structures/subnet/candidate.py

codecov · 2022-10-13T13:24:38Z

Codecov Report

❗ No coverage uploaded for pull request base (autoformer_algo_dev@b4b7e24). Click here to learn what that means.
Patch has no changes to coverable lines.

Additional details and impacted files

@@                  Coverage Diff                   @@
##             autoformer_algo_dev     #316   +/-   ##
======================================================
  Coverage                       ?   82.84%           
======================================================
  Files                          ?      190           
  Lines                          ?     8375           
  Branches                       ?     1307           
======================================================
  Hits                           ?     6938           
  Misses                         ?     1175           
  Partials                       ?      262

Flag	Coverage Δ
unittests	`82.84% <0.00%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

* Update mmcls pplnn benchmark * Update supported model list

* support for litehrnet * support ncnn vulkan * update supported models * add docstring * add one ut and fix wrapper * add test_shuffleunit * add test_stem_forward * add last ut * fix flake8 * fix yapf * fix lint * fix yapf * fix comments * unified chunk * mv adaptive_avg_pool * fix lint * move adaptive_avg_pool_ncnn * fix lint * symbolic rewriter of adaptive_avg_pool2d * fix lint * fix yapf * fix ci

openmmlab-bot · 2023-03-28T09:29:58Z

aptsunny，您好！您在MMRazor项目中给我们提的PR非常重要，感谢您付出私人时间帮助改进开源项目，相信很多开发者会从你的PR中受益。
我们非常期待与您继续合作，OpenMMLab专门成立了贡献者组织MMSIG，为贡献者们提供开源证书、荣誉体系和专享好礼，可通过添加微信：openmmlabwx 联系我们（请备注mmsig+GitHub id），由衷希望您能加入！
Dear aptsunny,
First of all, we want to express our gratitude for your significant PR in the MMXXX project. Your contribution is highly appreciated, and we are grateful for your efforts in helping improve this open-source project during your personal time. We believe that many developers will benefit from your PR.
We are looking forward to continuing our collaboration with you. OpenMMLab has established a special contributors' organization called MMSIG, which provides contributors with open-source certificates, a recognition system, and exclusive rewards. You can contact us by adding our WeChat（if you have WeChat): openmmlab_wx, or join in our discord： https://discord.gg/raweFPmdzG. We sincerely hope you will join us!
Best regards！ @aptsunny

aptsunny added 4 commits October 9, 2022 14:59

update candidates

965279a

update subnet_sampler_loop

ab6f1f4

update candidate

f5f6a7f

Merge branch 'autoformer_algo_dev' into autoformer_algo_candidate

cdd6eae

aptsunny requested review from gaoyang07 and sunnyxiaohu October 11, 2022 07:14

gaoyang07 reviewed Oct 11, 2022

View reviewed changes

aptsunny added 2 commits October 13, 2022 11:08

rename variable

4ec1444

rename variable

a1fa920

aptsunny force-pushed the autoformer_algo_candidate branch from 4fdfa3a to a1fa920 Compare October 13, 2022 03:09

clean

2293ddd

gaoyang07 reviewed Oct 13, 2022

View reviewed changes

sunnyxiaohu reviewed Oct 13, 2022

View reviewed changes

update

540e696

add doc string

9038e35

aptsunny merged commit 325dd32 into open-mmlab:autoformer_algo_dev Oct 14, 2022

aptsunny mentioned this pull request Oct 14, 2022

Revert "[Improvement] Support for candidate multiple dimensional search constraints." #321

Merged

humu789 pushed a commit to humu789/mmrazor that referenced this pull request Feb 13, 2023

[Docs] Update mmcls pplnn benchmark (open-mmlab#316)

543777f

* Update mmcls pplnn benchmark * Update supported model list

Conversation

aptsunny commented Oct 11, 2022

Motivation

Use cases (Optional)

Checklist

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!

sunnyxiaohu commented Oct 13, 2022

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!

Uh oh!

codecov bot commented Oct 13, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

openmmlab-bot commented Mar 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

codecov bot commented Oct 13, 2022 •

edited

Loading

openmmlab-bot commented Mar 28, 2023 •

edited

Loading