keras开发成sklearn接口

Posted 2021-01-06 wzdly

我们可以通过包装器将Sequential模型（仅有一个输入）作为Scikit-Learn工作流的一部分，相关的包装器定义在keras.wrappers.scikit_learn.py中：

这里有两个包装器可用：

分类器接口：keras.wrappers.scikit_learn.KerasClassifier(build_fn=None, **sk_params)

回归器接口：keras.wrappers.scikit_learn.KerasRegressor(build_fn=None, **sk_params)

参考文献：https://keras-cn.readthedocs.io/en/latest/scikit-learn_API/

"""
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasRegressor
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
def model(optimizer="adam"):
    #create model
    model = Sequential()
    model.add(Dense(input_dim=4,units=12,activation="relu"))
    model.add(Dense(units=8,activation="relu"))
    model.add(Dense(units=1,activation="sigmoid"))
    #compile model
    model.compile(loss="mse",optimizer=optimizer,metrics=["accuracy"],)
    return model
#######################################################################################
#create data
np.random.seed(seed=10)
X = np.random.randn(100,4)
y = np.random.randn(100)

#split data in train dataset and test dataset
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

#using wrappers to create sklearn interface

model = KerasRegressor(build_fn=model,epochs=10,batch_size=5)

#training
model.fit(X_train,y_train)
#predicting
y_pred = model.predict(X_test)
#evalution
print("mse:"+str(mean_squared_error(y_test,y_pred)))

#cross_validation
from sklearn.model_selection import cross_val_score
mse = cross_val_score(estimator=model,X=X,y=y,cv=5,n_jobs=1,scoring="neg_mean_squared_error")
print("average value of mse:"+str(mse))
#########################################################################################
#adjust parameters of model
#gridSearchCV
from sklearn.model_selection import GridSearchCV
params = {"optimizer":[‘rmsprop‘,‘adam‘],
          "epochs": [5,10],
          "batch_size":[5,10],
        }

gridSearchCV = GridSearchCV(estimator=model,param_grid=params,cv=5)
result = gridSearchCV.fit(X,y)

result.best_params_
result.best_score_
#########################################################################################
#skopt
from skopt.space import Real,Integer,Categorical
from skopt.utils import use_named_args
from skopt import gp_minimize

space = [Categorical(categories=[‘rmsprop‘,‘adam‘],name="optimizer"),
         Categorical(categories=[1,2,3],name="epochs")]

@use_named_args(space)
def objective(**params):
    model.set_params(**params)
    return -np.mean(cross_val_score(model,X,y,cv=5,n_jobs=1,scoring="neg_mean_squared_error"))

result = gp_minimize(objective, space, n_calls=50, random_state=0)
print("best score:%.4f"%(result.fun))
print("best parameters:",result.x)