aiauto-client 0.1.14__py3-none-any.whl → 0.1.16__py3-none-any.whl

aiauto/core.py

@@ -188,11 +188,24 @@ class StudyWrapper:
         parallelism: int = 2,
         requirements_file: Optional[str] = None,
         requirements_list: Optional[List[str]] = None,
-        resources_requests: Optional[Dict[str, str]] = {"cpu": "256m", "memory": "256Mi"},
-        resources_limits: Optional[Dict[str, str]] = {"cpu": "256m", "memory": "256Mi"},
+        resources_requests: Optional[Dict[str, str]] = None,
+        resources_limits: Optional[Dict[str, str]] = None,
         runtime_image: Optional[str] = 'ghcr.io/astral-sh/uv:python3.8-bookworm-slim',
         use_gpu: bool = False
     ) -> None:
+        # 리소스 기본값 설정
+        if resources_requests is None:
+            if use_gpu:
+                resources_requests = {"cpu": "2", "memory": "4Gi"}
+            else:
+                resources_requests = {"cpu": "1", "memory": "1Gi"}
+        
+        if resources_limits is None:
+            if use_gpu:
+                resources_limits = {"cpu": "2", "memory": "4Gi"}
+            else:
+                resources_limits = {"cpu": "1", "memory": "1Gi"}
+        
         if runtime_image is None or runtime_image == "":
             if use_gpu:
                 runtime_image = "pytorch/pytorch:2.1.0-cuda12.1-cudnn8-runtime"

aiauto/serializer.py

@@ -59,9 +59,12 @@ def object_to_json(obj: Union[object, dict, None]) -> str:
             
             # __init__의 실제 파라미터인지 확인
             if param_name in valid_params:
-                # PatientPruner의 wrapped_pruner만 특별 처리
+                # PatientPruner의 wrapped_pruner 특별 처리
                 if class_name == "PatientPruner" and param_name == "wrapped_pruner" and value is not None:
                     kwargs[param_name] = json.loads(object_to_json(value))
+                # CmaEsSampler와 QMCSampler의 independent_sampler 특별 처리
+                elif param_name == "independent_sampler" and value is not None and class_name in ["CmaEsSampler", "QMCSampler"]:
+                    kwargs[param_name] = json.loads(object_to_json(value))
                 # Callable 타입은 제외 (gamma, weights 등)
                 elif not callable(value):
                     kwargs[param_name] = value

{aiauto_client-0.1.14.dist-info → aiauto_client-0.1.16.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: aiauto-client
-Version: 0.1.14
+Version: 0.1.16
 Summary: AI Auto HPO (Hyperparameter Optimization) Client Library
 Author-email: AIAuto Team <ainode@zeroone.ai>
 Project-URL: Homepage, https://dashboard.common.aiauto.pangyo.ainode.ai
@@ -121,7 +121,7 @@ Jupyter Notebook이나 Python REPL에서 정의한 함수는 Serialize 할 수
 import aiauto
 import optuna
 
-def objective(trial: optuna.trial.Trial):
+def objective(trial):
     """
     이 함수는 외부 서버에서 실행됩니다.
     모든 import는 함수 내부에 작성하세요.
@@ -170,7 +170,7 @@ time.sleep(5)
 # `https://dashboard.common.aiauto.pangyo.ainode.ai/study` 에서 생성된 study 확인 가능
 
 # objective 함수 정의
-def objective(trial: optuna.trial.Trial):
+def objective(trial):
     """실제 실행은 사용자 로컬 컴퓨터가 아닌 서버에서 실행 될 함수"""
     x = trial.suggest_float('x', -10, 10)
     y = trial.suggest_float('y', -10, 10)
@@ -214,7 +214,7 @@ time.sleep(5)
 
 # objective 함수 정의
 # https://docs.pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html 참고
-def objective(trial: optuna.trial.Trial):
+def objective(trial):
     """
     실제 실행은 사용자 로컬 컴퓨터가 아닌 서버에서 실행 될 함수
     모든 import는 함수 내부에 존재해야 함
@@ -224,6 +224,8 @@ def objective(trial: optuna.trial.Trial):
     from torch.utils.data import DataLoader, random_split, Subset
     from torchvision import transforms, datasets
     import torch.nn.functional as F
+    
+    import optuna
 
     # 하이퍼파라미터 샘플링
     lr = trial.suggest_float('learning_rate', 1e-5, 1e-1, log=True)
@@ -315,7 +317,12 @@ study_wrapper.optimize(
     n_trials=100,
     parallelism=4,
     use_gpu=True,  # GPU 사용
-    requirements_list=['torch', 'torchvision']  # Pod에서 자동 설치
+    runtime_image='pytorch/pytorch:2.1.0-cuda11.8-cudnn8-runtime',  # default image for use_gpu True
+    # requirements_list=['torch', 'torchvision']  # Pod에서 자동 설치  # pip list 명시는 다운로드 받는데 느림, runtime_image 를 torch 로 명시하는게 나음
+    resources_requests={
+        "cpu": "2",
+        "memory": "4Gi",
+    },
 )
 time.sleep(5)
 ```
@@ -344,7 +351,7 @@ time.sleep(5)
 
 # objective 함수 정의
 # https://docs.pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html 참고
-def objective(trial: optuna.trial.Trial):
+def objective(trial):
     """
     실제 실행은 사용자 로컬 컴퓨터가 아닌 서버에서 실행 될 함수
     모든 import는 함수 내부에 존재해야 함
@@ -446,7 +453,16 @@ study_wrapper.optimize(
     n_trials=100,
     parallelism=4,
     use_gpu=True,  # GPU 사용
-    requirements_list=['torch', 'torchvision', 'fvcore']  # Pod에서 자동 설치
+    runtime_image='pytorch/pytorch:2.1.0-cuda11.8-cudnn8-runtime',  # default image for use_gpu True
+    requirements_list=[  # pip list 명시는 다운로드 받는데 느림, runtime_image 를 torch 로 명시하는게 나음
+        # 'torch', 
+        # 'torchvision', 
+        'fvcore',
+    ],  # Pod에서 자동 설치
+    resources_requests={
+        "cpu": "2",
+        "memory": "4Gi",
+    },
 )
 time.sleep(5)
 ```
@@ -545,15 +561,15 @@ study = controller.create_study('exp1', direction='minimize')
 time.sleep(5)
 
 def objective(trial):
-import numpy as np
-x = trial.suggest_float('x', -10, 10)
-return (x - 1.23) ** 2
+    import numpy as np
+    x = trial.suggest_float('x', -10, 10)
+    return (x - 1.23) ** 2
 
 study.optimize(
-objective,
-n_trials=64,
-parallelism=8,
-requirements_list=['numpy'],
+    objective,
+    n_trials=64,
+    parallelism=8,
+    requirements_list=['numpy'],
 )
 time.sleep(5)
 ```
@@ -600,17 +616,17 @@ import optuna, aiauto, time
 
 controller = aiauto.AIAutoController('aiauto_xxx')
 study = controller.create_study(
-study_name='cnn',
-direction='minimize',
-sampler=optuna.samplers.TPESampler(seed=42),
-pruner=optuna.pruners.MedianPruner(n_startup_trials=5),
+    study_name='cnn',
+    direction='minimize',
+    sampler=optuna.samplers.TPESampler(seed=42),
+    pruner=optuna.pruners.MedianPruner(n_startup_trials=5),
 )
 time.sleep(5)
 
 def objective(trial):
-import numpy as np
-lr = trial.suggest_float('lr', 1e-5, 1e-1, log=True)
-return (np.log10(lr) + 2) ** 2
+    import numpy as np
+    lr = trial.suggest_float('lr', 1e-5, 1e-1, log=True)
+    return (np.log10(lr) + 2) ** 2
 
 study.optimize(objective, n_trials=50, parallelism=4)
 time.sleep(5)

aiauto_client-0.1.16.dist-info/RECORD

@@ -0,0 +1,10 @@
+aiauto/__init__.py,sha256=sF7sJaXg7-MqolSYLxsaXAir1dBzARhXLrHo7zLsupg,345
+aiauto/_config.py,sha256=hTFh2bH9m-HuX6QCpNtBC0j6rEB0S97hhPKjbEjv4Tg,89
+aiauto/constants.py,sha256=rBibGOQHHrdkwaai92-3I8-N0cu-B4CoCoQbG9-Cl8k,821
+aiauto/core.py,sha256=Hz3HP6xJGey6LbF04cYQtPPqerf4LhDgt0BVBPPf0j4,10364
+aiauto/http_client.py,sha256=v_nPdb-2tIeH1XrOYqzMGvFfXLKEDbQoSaQYPsB0Hik,2587
+aiauto/serializer.py,sha256=BJmeq6uCD9D2_6bXu_sMBQLSsXCUMIMM10iX923DTXE,2749
+aiauto_client-0.1.16.dist-info/METADATA,sha256=Lc0WSJ4UFunxbOLJWm-0K9thPPJl98ijomAJpvG2knI,25625
+aiauto_client-0.1.16.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+aiauto_client-0.1.16.dist-info/top_level.txt,sha256=Sk2ctO9_Bf_tAPwq1x6Vfl6OuL29XzwMTO4F_KG6oJE,7
+aiauto_client-0.1.16.dist-info/RECORD,,

aiauto_client-0.1.14.dist-info/RECORD

@@ -1,10 +0,0 @@
-aiauto/__init__.py,sha256=sF7sJaXg7-MqolSYLxsaXAir1dBzARhXLrHo7zLsupg,345
-aiauto/_config.py,sha256=hTFh2bH9m-HuX6QCpNtBC0j6rEB0S97hhPKjbEjv4Tg,89
-aiauto/constants.py,sha256=rBibGOQHHrdkwaai92-3I8-N0cu-B4CoCoQbG9-Cl8k,821
-aiauto/core.py,sha256=eEwit5oL8DIfglOVe2km_7MAtuZquEd5Xvkbq6EaW9o,9945
-aiauto/http_client.py,sha256=v_nPdb-2tIeH1XrOYqzMGvFfXLKEDbQoSaQYPsB0Hik,2587
-aiauto/serializer.py,sha256=cedmfJaoy-axG2jyFvUIVbyu9Fqen0hot6qDHnwSlGM,2466
-aiauto_client-0.1.14.dist-info/METADATA,sha256=nybQSHynZV0YCkHsqe4IZRIYPrBi3zKgHutlZXsJg8c,25022
-aiauto_client-0.1.14.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-aiauto_client-0.1.14.dist-info/top_level.txt,sha256=Sk2ctO9_Bf_tAPwq1x6Vfl6OuL29XzwMTO4F_KG6oJE,7
-aiauto_client-0.1.14.dist-info/RECORD,,