如何将使用 GridSearchCV 创建的模型部署到 sagemaker

Question

我已经使用 GridSearchCv 构建了一个模型，现在我想将其部署到 sagemaker 并创建端点，我该怎么做。

代码示例

_data=pd.read_csv('s3://ml-titanic-dataset/train_data/train.csv')
_data.head()

#handling age NA
_data['Age'].fillna(_data.groupby('Sex')['Age'].transform(lambda x:x.mean()),inplace=True)
_data['Cabin']=_data['Cabin'].fillna('missing')

# selecting columns
x=_data.loc[:,['Pclass','Sex','Age','SibSp','Parch','Fare','Embarked']]
y=_data['Survived']

x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.20,train_size=0.80,random_state=42)

cat_col=x.select_dtypes('O').columns.values
num_col=x.select_dtypes(['int64','float64']).columns.values

cat_pipe=Pipeline([
    ('impute',SimpleImputer(strategy='most_frequent')),
    ('encoding',OneHotEncoder(drop='first',handle_unknown='ignore'))
])

num_pipe=Pipeline([
    ('scaling',StandardScaler())
])

ct=ColumnTransformer([
     ('categorical',cat_pipe,cat_col),
     ('numerical',num_pipe,num_col)    
],remainder='passthrough')


final_pipe=Pipeline([
    ('data_processing',ct),
    ('mdl','passthrough')
])
# final_pipe.fit_transform(x_train,y_train)
parameter=[
    {
        'mdl':[RandomForestClassifier()]
    },
    {
        'mdl':[LogisticRegression()],
        'mdl__C':[0.4,0.6,0.8,0.9]
    },
    {
        'mdl':[AdaBoostClassifier()],
        #'mdl__estimator':[LogisticRegression(),DecisionTreeClassifier()]
    }
]


grid_search=GridSearchCV(final_pipe,parameter,cv=5,scoring='f1',verbose=5,error_score='raise')
grid_search.fit(x_train,y_train)
accuracy_score(y_test,grid_search.predict(x_test))

我是否可以将自行创建的模型部署到 sagemaker，或者我必须使用内置 sagemaker 算法来部署和创建端点，请帮助我理解

Answer 1

您不需要使用SageMaker内置算法，您可以自带算法，也可以自带图像。如果这是基于 sklearn 的模型，您可以使用预构建的 SKLearn 图像 并带上您自己的推理代码，详细信息请参见此处 - https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-推理代码.html

这是一个示例笔记本 - https://github.com/aws/amazon-sagemaker-examples/blob/main/advanced_functionity/scikit_learn_bring_your_own_model/scikit_learn_bring_your_own_model.ipynb