[Refactor] Task and Tune API

[KKIEEK]

Motivation

To build runnable task from cfg

TODO

• Enhance work_dir (experiment name is determined by work_dir)
• Refactor log_analysis
• Add test code for Tuner class
• Pass test code
• Final test

Modification

Please briefly describe what modification is made in this PR.

Expected behavior

work_dirs
  |- mmdet_asynchb_nevergrad_pso (config name)
    |- 221221_113600 (experiment name)
      |- best_trial
        |- work_dirs (log of best trial)
        |- best_trial.log
      |- DataParallelTrainer_738f915b_1_data_samples_per_gpu=5,model=yolo_x_8x8,optimizer=adam ... (trial_name)
        |- rank_0 (log dir for task process)
          |- work_dirs
            |- faster_rcnn_r50_fpn_1x_coco
              |- faster_rcnn_r50_fpn_1x_coco.py
              |- latest.pth
        |- rank_1
        |- params.json
        |- params.pkl 
        |- progress.csv
        |- result.json
      |- DataParallelTrainer_7caa5acf_2_data_samples_per_gpu=6,model=faster_rcnn_swin_s_p4_w7_fpn,optimizer=adamw ...
        |- (trial directory name is determined by trial_name_creator and trial_dirname_creator)
      |- some.json
      |- some.pkl

Result

work_dirs/mmdet_asynchb_nevergrad_pso/20221223_070902
├── 1_data_samples_per_gpu=7,model=yolo_x_8x8,optimizer=adamw
│   ├── error.pkl
│   ├── error.txt
│   ├── events.out.tfevents.1671779346.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 2005c6e6
│   │           ├── 2005c6e6.py
│   │           ├── 20221223_070919.log
│   │           └── 20221223_070919.log.json
│   └── result.json
├── 2_data_samples_per_gpu=3,model=tood_r101_dcnv2,optimizer=adamw
│   ├── checkpoint_-00001
│   ├── events.out.tfevents.1671779388.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── progress.csv
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 60d57f42
│   │           ├── 20221223_071001.log
│   │           ├── 20221223_071001.log.json
│   │           └── 60d57f42.py
│   └── result.json
├── 3_data_samples_per_gpu=6,model=yolo_x_8x8,optimizer=rms
│   ├── error.pkl
│   ├── error.txt
│   ├── events.out.tfevents.1671779444.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 777781b3
│   │           ├── 20221223_071057.log
│   │           ├── 20221223_071057.log.json
│   │           └── 777781b3.py
│   └── result.json
├── 4_data_samples_per_gpu=3,model=tood_r101_dcnv2,optimizer=sgd
│   ├── checkpoint_-00001
│   ├── events.out.tfevents.1671779486.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── progress.csv
│   ├── rank_0
│   │   └── work_dirs
│   │       └── b42854ef
│   │           ├── 20221223_071139.log
│   │           ├── 20221223_071139.log.json
│   │           └── b42854ef.py
│   └── result.json
├── 5_data_samples_per_gpu=2,model=faster_rcnn_x101_64x4d_fpn,optimizer=rms
│   ├── checkpoint_-00001
│   ├── events.out.tfevents.1671779532.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── progress.csv
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 463aa231
│   │           ├── 20221223_071225.log
│   │           ├── 20221223_071225.log.json
│   │           └── 463aa231.py
│   └── result.json
├── 6_data_samples_per_gpu=3,model=faster_rcnn_swin_s_p4_w7_fpn,optimizer=sgd
│   ├── checkpoint_-00001
│   ├── events.out.tfevents.1671779574.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── progress.csv
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 037ff5be
│   │           ├── 037ff5be.py
│   │           ├── 20221223_071308.log
│   │           └── 20221223_071308.log.json
│   └── result.json
├── 7_data_samples_per_gpu=7,model=yolo_x_8x8,optimizer=adam
│   ├── error.pkl
│   ├── error.txt
│   ├── events.out.tfevents.1671779618.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 7b5d2242
│   │           ├── 20221223_071351.log
│   │           ├── 20221223_071351.log.json
│   │           └── 7b5d2242.py
│   └── result.json
├── 8_data_samples_per_gpu=5,model=faster_rcnn_x101_64x4d_fpn,optimizer=adam
│   ├── checkpoint_-00001
│   ├── events.out.tfevents.1671779660.test-hpo-6667d579dc-nvhjg
│   ├── params.json
│   ├── params.pkl
│   ├── progress.csv
│   ├── rank_0
│   │   └── work_dirs
│   │       └── 61b2f6a0
│   │           ├── 20221223_071433.log
│   │           ├── 20221223_071433.log.json
│   │           └── 61b2f6a0.py
│   └── result.json
├── best_trial
│   ├── best_trial.log
│   └── log
│       ├── checkpoint_-00001
│       ├── events.out.tfevents.1671779388.test-hpo-6667d579dc-nvhjg
│       ├── params.json
│       ├── params.pkl
│       ├── progress.csv
│       ├── rank_0
│       │   └── work_dirs
│       │       └── 60d57f42
│       │           ├── 20221223_071001.log
│       │           ├── 20221223_071001.log.json
│       │           └── 60d57f42.py
│       └── result.json
├── experiment_state-2022-12-23_07-09-06.json
├── mmdet_asynchb_nevergrad_pso.py
├── search_gen_state-2022-12-23_07-09-06.json
├── searcher-state-2022-12-23_07-09-06.pkl
├── trainable.pkl
└── tuner.pkl

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[yhna940]

GOOD 💡

[KKIEEK]

Suggestion

• To support a relative path, follow works must be preceded.

# at tune/config.py
class _CustomTorchBackend(_TorchBackend):
    ...

    def on_start(self, worker_group: WorkerGroup,
        ...

        def set_env_vars(addr, port, rank, world_size):
            # fix here
            os.environ['TUNE_ORIG_WORKING_DIR'] = os.getcwd()

            os.environ['MASTER_ADDR'] = addr
            os.environ['MASTER_PORT'] = str(port)
            os.environ['RANK'] = str(rank)
            os.environ['LOCAL_RANK'] = str(rank)
            os.environ['WORLD_SIZE'] = str(world_size)

# at core/rewriters/rel2abs.py
import os

from .base import BaseRewriter
from .builder import REWRITERS

@REWRITERS.register_module()
class Rel2Abs(BaseRewriter):

    def __call__(self, context: dict) -> dict:
        os.chdir(os.environ['TUNE_ORIG_WORKING_DIR'])  # or some task
        return context

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (main@9475dc2). Click here to learn what that means.
Patch has no changes to coverable lines.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #118   +/-   ##
=======================================
  Coverage        ?   73.09%           
=======================================
  Files           ?       58           
  Lines           ?     1617           
  Branches        ?      238           
=======================================
  Hits            ?     1182           
  Misses          ?      334           
  Partials        ?      101           

Flag	Coverage Δ
unittests	73.09% <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.