从我发现的内容来看，还有另外1个这样的问题（Speed-up nested cross-validation），但是在尝试了该站点和microsoft上建议的几种修复之后，安装MPI对我也不起作用，所以我希望还有另一个问题包装或回答该问题。
我正在寻找比较多种算法和进行网格搜索的各种参数（也许参数太多？）的方法，除了mpi4py之外还有什么方法可以加快我的代码的运行速度？据我了解，我不能使用n_jobs = -1，因为那是不嵌套的？
还要注意，我无法在下面尝试查看的许多参数上运行它（运行时间超过了我的时间）。如果我给每个模型仅两个参数进行比较，则只有2小时后才会有结果。此外，我在252行和25个特征列以及4个类别变量的数据集上运行此代码，以预测（“确定”，“可能”，“可能”或“未知”）某个基因（具有252个基因）是否影响疾病。使用SMOTE会将样本大小增加到420，这样就可以使用了。
dataset= pd.read_csv('data.csv')
data = dataset.drop(["gene"],1)
df = data.iloc[:,0:24]
df = df.fillna(0)
X = MinMaxScaler().fit_transform(df)
le = preprocessing.LabelEncoder()
encoded_value = le.fit_transform(["certain", "likely", "possible", "unlikely"])
Y = le.fit_transform(data["category"])
sm = SMOTE(random_state=100)
X_res, y_res = sm.fit_resample(X, Y)
seed = 7
logreg = LogisticRegression(penalty='l1', solver='liblinear',multi_class='auto')
LR_par= {'penalty':['l1'], 'C': [0.5, 1, 5, 10], 'max_iter':[500, 1000, 5000]}
rfc =RandomForestClassifier()
param_grid = {'bootstrap': [True, False],
'max_depth': [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, None],
'max_features': ['auto', 'sqrt'],
'min_samples_leaf': [1, 2, 4,25],
'min_samples_split': [2, 5, 10, 25],
'n_estimators': [200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000]}
mlp = MLPClassifier(random_state=seed)
parameter_space = {'hidden_layer_sizes': [(10,20), (10,20,10), (50,)],
'activation': ['tanh', 'relu'],
'solver': ['adam', 'sgd'],
'max_iter': [10000],
'alpha': [0.1, 0.01, 0.001],
'learning_rate': ['constant','adaptive']}
gbm = GradientBoostingClassifier(min_samples_split=25, min_samples_leaf=25)
param = {"loss":["deviance"],
"learning_rate": [0.15,0.1,0.05,0.01,0.005,0.001],
"min_samples_split": [2, 5, 10, 25],
"min_samples_leaf": [1, 2, 4,25],
"max_depth":[10, 20, 30, 40, 50, 60, 70, 80, 90, 100, None],
"max_features":['auto', 'sqrt'],
"criterion": ["friedman_mse"],
"n_estimators":[200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000]
}
svm = SVC(gamma="scale", probability=True)
tuned_parameters = {'kernel':('linear', 'rbf'), 'C':(1,0.25,0.5,0.75)}
def baseline_model(optimizer='adam', learn_rate=0.01):
model = Sequential()
model.add(Dense(100, input_dim=X_res.shape[1], activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(50, activation='relu')) #8 is the dim/ the number of hidden units (units are the kernel)
model.add(Dense(4, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])
return model
keras = KerasClassifier(build_fn=baseline_model, batch_size=32, epochs=100, verbose=0)
learn_rate = [0.001, 0.01, 0.1, 0.2, 0.3]
optimizer = ['SGD', 'RMSprop', 'Adagrad', 'Adadelta', 'Adam', 'Adamax', 'Nadam']
kerasparams = dict(optimizer=optimizer, learn_rate=learn_rate)
inner_cv = KFold(n_splits=10, shuffle=True, random_state=seed)
outer_cv = KFold(n_splits=10, shuffle=True, random_state=seed)
models = []
models.append(('GBM', GridSearchCV(gbm, param, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('RFC', GridSearchCV(rfc, param_grid, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('LR', GridSearchCV(logreg, LR_par, cv=inner_cv, iid=False, n_jobs=1)))
models.append(('SVM', GridSearchCV(svm, tuned_parameters, cv=inner_cv, iid=False, n_jobs=1)))
models.append(('MLP', GridSearchCV(mlp, parameter_space, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('Keras', GridSearchCV(estimator=keras, param_grid=kerasparams, cv=inner_cv,iid=False, n_jobs=1)))
results = []
names = []
scoring = 'accuracy'
X_train, X_test, Y_train, Y_test = train_test_split(X_res, y_res, test_size=0.2, random_state=0)
for name, model in models:
nested_cv_results = model_selection.cross_val_score(model, X_res, y_res, cv=outer_cv, scoring=scoring)
results.append(nested_cv_results)
names.append(name)
msg = "Nested CV Accuracy %s: %f (+/- %f )" % (name, nested_cv_results.mean()*100, nested_cv_results.std()*100)
print(msg)
model.fit(X_train, Y_train)
print('Test set accuracy: {:.2f}'.format(model.score(X_test, Y_test)*100), '%')
print("Best Parameters: \n{}\n".format(model.best_params_))
print("Best CV Score: \n{}\n".format(model.best_score_))
举个例子，大多数数据集都是二进制的，看起来像这样：
gene Tissue Druggable Eigenvalue CADDvalue Catalogpresence Category
ACE 1 1 1 0 1 Certain
ABO 1 0 0 0 0 Likely
TP53 1 1 0 0 0 Possible
任何有关如何加快速度的指南都将受到赞赏。
编辑：我也尝试过使用dask进行并行处理，但是我不确定这样做是否正确，而且运行似乎没有更快：
for name, model in models:
with joblib.parallel_backend('dask'):
nested_cv_results = model_selection.cross_val_score(model, X_res, y_res, cv=outer_cv, scoring=scoring)
results.append(nested_cv_results)
names.append(name)
msg = "Nested CV Accuracy %s: %f (+/- %f )" % (name, nested_cv_results.mean()*100, nested_cv_results.std()*100)
print(msg)
model.fit(X_train, Y_train)
print('Test set accuracy: {:.2f}'.format(model.score(X_test, Y_test)*100), '%')
#print("Best Estimator: \n{}\n".format(model.best_estimator_))
print("Best Parameters: \n{}\n".format(model.best_params_))
print("Best CV Score: \n{}\n".format(model.best_score_)) #average of all cv folds for a single combination of the parameters you specify
编辑：还要注意减少网格搜索，例如，我尝试使用每个模型5个参数，但是这仍然需要几个小时才能完成，因此，如果有效率方面的任何建议，则在减少数量的同时会有所帮助我将不胜感激。
答案 0 :(得分：3)
Dask-ML具有可扩展的实现GridSearchCV和RandomSearchCV，我相信它们会替代Scikit-Learn。它们与Scikit-Learn开发人员一起开发。
它们可以更快的原因有两个：
答案 1 :(得分：2)
两件事：
代替admin.auth().setCustomUserClaims(uid, {user_type: "driver"}).then(() => {
// The new custom claims will propagate to the user's ID token the
// next time a new one is issued.
});
尝试使用HyperOpt-它是用于串行和并行优化的Python库。
应用GridSearch后：
SMOTE然后您可以使用import umap
dim_reduced = umap.UMAP(
min_dist=min_dist,
n_neighbors=neighbours,
random_state=1234,
).fit_transform(smote_output)
进行列车测试。
减小维数将有助于消除数据中的噪声，您无需处理25个功能，而是将它们降低到2个（使用UMAP）或选择的组件数量（使用PCA）。这应该对性能产生重大影响。
答案 2 :(得分：1)
在您的情况下有个轻松的胜利，那就是....开始使用并行处理:)。如果您有群集，dask将为您提供帮助（它可以在一台计算机上运行，但是与sklearn中的默认计划相比，改进并不明显），但是如果您打算在一个群集上运行它，一台机器（但具有多个内核/线程和“足够的”内存），则可以并行运行嵌套的CV。唯一的技巧是sklearn不允许您在多个进程中运行外部CV 循环。但是，它将允许您在多个线程中运行内部循环。
目前，您在外部CV循环中有n_jobs=None（这是cross_val_score中的默认值），这意味着n_jobs=1，这是您可以与{{1 }}放在嵌套的简历中。
但是，通过在所有使用的sklearn中设置n_jobs=some_reasonable_number，可以轻松获得收益。 GridSearchCV不一定是some_reasonable_number（但这是一个很好的起点）。有些算法要么停留在-1上，而不是n_jobs=n_cores（例如，n_threads），要么已经具有内置的多处理功能（例如，xgboost），并且可能如果产生太多进程，则会发生冲突。
答案 3 :(得分：1)
IIUC，您正在尝试从sklearn文档中并行化this example。如果是这样，那么这里是一种可能的解决方法
为什么达斯克不起作用
和
关于此问题的任何建设性指导或进一步知识
一般进口
import numpy as np
import pandas as pd
from sklearn.preprocessing import MinMaxScaler
from sklearn import preprocessing
from imblearn.over_sampling import SMOTE
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.svm import SVC
from sklearn.model_selection import cross_val_score, GridSearchCV, KFold, train_test_split
from sklearn.neural_network import MLPClassifier
import dask_ml.model_selection as dcv
import time
数据
dask_ml
dask_ml #### Dataset 1 - longer version of data in the question
d = """gene Tissue Druggable Eigenvalue CADDvalue Catalogpresence Category
ACE 1 1 1 0 1 Certain
ABO 1 0 0 0 0 Likely
TP53 1 1 0 0 0 Possible"""
data = pd.DataFrame([x.split(' ') for x in d.split('\n')])
data.columns = data.loc[0,:]
data.drop(0, axis=0, inplace=True)
data = pd.concat([data]*15)
data = data.drop(["gene"],1)
df = data.iloc[:,0:5]
X = MinMaxScaler().fit_transform(df)
le = preprocessing.LabelEncoder()
encoded_value = le.fit_transform(["Certain", "Likely", "Possible"])
Y = le.fit_transform(data["Category"])
sm = SMOTE(random_state=100)
X_res, y_res = sm.fit_resample(X, Y)
#### Dataset 2 - iris dataset from example in sklearn nested cross validation docs
# Load the dataset
from sklearn.datasets import load_iris
iris = load_iris()
X_res = iris.data
y_res = iris.target
#### Dataset 3 - size (#rows, #columns) is adjustable (I used this to time code execution)
X_res = pd.DataFrame(np.random.rand(300,50), columns=['col_'+str(c+1) for c in list(range(50))])
from random import shuffle
cats = ["paris", "barcelona", "kolkata", "new york", 'sydney']
y_values = cats*int(len(X_res)/len(cats))
shuffle(y_values)
y_res = pd.Series(y_values)
实例化分类器-问题代码没有变化
seed = 7
logreg = LogisticRegression(penalty='l1', solver='liblinear',multi_class='auto')
LR_par= {'penalty':['l1'], 'C': [0.5, 1, 5, 10], 'max_iter':[500, 1000, 5000]}
mlp = MLPClassifier(random_state=seed)
parameter_space = {'hidden_layer_sizes': [(10,20), (10,20,10), (50,)],
'activation': ['tanh', 'relu'],
'solver': ['adam', 'sgd'],
'max_iter': [10000],
'alpha': [0.1, 0.01, 0.001],
'learning_rate': ['constant','adaptive']}
rfc =RandomForestClassifier()
param_grid = {'bootstrap': [True, False],
'max_depth': [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, None],
'max_features': ['auto', 'sqrt'],
'min_samples_leaf': [1, 2, 4,25],
'min_samples_split': [2, 5, 10, 25],
'n_estimators': [200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000]}
gbm = GradientBoostingClassifier(min_samples_split=25, min_samples_leaf=25)
param = {"loss":["deviance"],
"learning_rate": [0.15,0.1,0.05,0.01,0.005,0.001],
"min_samples_split": [2, 5, 10, 25],
"min_samples_leaf": [1, 2, 4,25],
"max_depth":[10, 20, 30, 40, 50, 60, 70, 80, 90, 100, None],
"max_features":['auto', 'sqrt'],
"criterion": ["friedman_mse"],
"n_estimators":[200, 400, 600, 800, 1000, 1200, 1400, 1600, 1800, 2000]
}
svm = SVC(gamma="scale", probability=True)
tuned_parameters = {'kernel':('linear', 'rbf'), 'C':(1,0.25,0.5,0.75)}
inner_cv = KFold(n_splits=10, shuffle=True, random_state=seed)
outer_cv = KFold(n_splits=10, shuffle=True, random_state=seed)
使用GridSearchCV实现的dask_ml（最初由@MRocklin here提出）-请参阅dask_ml文档以获取dask_ml.model_selection.GridSearchCV
KerasClassifier和辅助函数baseline_model()，但我处理前者的方法与其他方法相同models = []
models.append(('MLP', dcv.GridSearchCV(mlp, parameter_space, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('GBM', dcv.GridSearchCV(gbm, param, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('RFC', dcv.GridSearchCV(rfc, param_grid, cv=inner_cv,iid=False, n_jobs=1)))
models.append(('LR', dcv.GridSearchCV(logreg, LR_par, cv=inner_cv, iid=False, n_jobs=1)))
models.append(('SVM', dcv.GridSearchCV(svm, tuned_parameters, cv=inner_cv, iid=False, n_jobs=1)))
初始化一个额外的空白列表以保存非嵌套的简历结果
non_nested_results = []
nested_results = []
names = []
scoring = 'accuracy'
X_train, X_test, Y_train, Y_test = train_test_split(X_res, y_res, test_size=0.2, random_state=0)
Joblib和dask客户端设置
# Create a local cluster
from dask.distributed import Client
client = Client(processes=False, threads_per_worker=4,
n_workers=1, memory_limit='6GB')
from sklearn.externals import joblib
根据sklearn docs example执行嵌套的简历
GridSearchCV cross_val_score sklearn模型（SVC）start = time.time()
for name, model in [models[-1]]:
# Non_nested parameter search and scoring
with joblib.parallel_backend('dask'):
model.fit(X_train, Y_train)
non_nested_results.append(model.best_score_)
# Nested CV with parameter optimization
nested_score = cross_val_score(model, X=X_train, y=Y_train, cv=outer_cv)
nested_results.append(nested_score.mean())
names.append(name)
msg = "Nested CV Accuracy %s: %f (+/- %f )" %\
(name, np.mean(nested_results)*100, np.std(nested_results)*100)
print(msg)
print('Test set accuracy: {:.2f}'.format(model.score(X_test, Y_test)*100), '%')
print("Best Estimator: \n{}\n".format(model.best_estimator_))
print("Best Parameters: \n{}\n".format(model.best_params_))
print("Best CV Score: \n{}\n".format(model.best_score_))
score_difference = [a_i - b_i for a_i, b_i in zip(non_nested_results, nested_results)]
print("Average difference of {0:6f} with std. dev. of {1:6f}."
.format(np.mean(score_difference), np.std(score_difference)))
print('Total running time of the script: {:.2f} seconds' .format(time.time()-start))
client.close()
下面是使用数据集3的输出（具有脚本执行时间）
不带dask 1
Nested CV Accuracy SVM: 20.416667 (+/- 0.000000 )
Test set accuracy: 16.67 %
Best Estimator:
SVC(C=0.75, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape='ovr', degree=3, gamma='scale', kernel='linear',
max_iter=-1, probability=True, random_state=None, shrinking=True,
tol=0.001, verbose=False)
Best Parameters:
{'C': 0.75, 'kernel': 'linear'}
Best CV Score:
0.2375
Average difference of 0.033333 with std. dev. of 0.000000.
Total running time of the script: 23.96 seconds
使用dask（使用n_workers=1和threads_per_worker=4的Output + Timing） 2
Nested CV Accuracy SVM: 18.750000 (+/- 0.000000 )
Test set accuracy: 13.33 %
Best Estimator:
SVC(C=0.5, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape='ovr', degree=3, gamma='scale', kernel='rbf',
max_iter=-1, probability=True, random_state=None, shrinking=True,
tol=0.001, verbose=False)
Best Parameters:
{'C': 0.5, 'kernel': 'rbf'}
Best CV Score:
0.1916666666666667
Average difference of 0.004167 with std. dev. of 0.000000.
Total running time of the script: 8.84 seconds
使用dask（使用n_workers=4和threads_per_worker=4的Output + Timing） 2
Nested CV Accuracy SVM: 23.333333 (+/- 0.000000 )
Test set accuracy: 21.67 %
Best Estimator:
SVC(C=0.25, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape='ovr', degree=3, gamma='scale', kernel='linear',
max_iter=-1, probability=True, random_state=None, shrinking=True,
tol=0.001, verbose=False)
Best Parameters:
{'C': 0.25, 'kernel': 'linear'}
Best CV Score:
0.25
Average difference of 0.016667 with std. dev. of 0.000000.
Total running time of the script: 7.52 seconds
使用dask（使用n_workers=1和threads_per_worker=8的Output + Timing） 2
Nested CV Accuracy SVM: 20.416667 (+/- 0.000000 )
Test set accuracy: 18.33 %
Best Estimator:
SVC(C=1, cache_size=200, class_weight=None, coef0=0.0,
decision_function_shape='ovr', degree=3, gamma='scale', kernel='rbf',
max_iter=-1, probability=True, random_state=None, shrinking=True,
tol=0.001, verbose=False)
Best Parameters:
{'C': 1, 'kernel': 'rbf'}
Best CV Score:
0.23333333333333334
Average difference of 0.029167 with std. dev. of 0.000000.
Total running time of the script: 7.06 seconds
1 使用sklearn.model_selection.GridSearchCV()而不使用joblib()
2 使用dask_ml.model_selection.GridSearchCV()替换sklearn.model_selection.GridSearchCV()并使用joblib()
此答案中有关代码和输出的注释
sklearn.model_selection.GridSearchCV()和cross_val_score的顺序颠倒了
Client(..., n_workers=n, threads_per_worker=m)和n>1和/或m=4 or m=8是否可接受/不正确 关于dask_ml 使用的一般评论（据我了解）
ParallelPostFit：如果您想使用joblib在集群上训练大型scikit-learn模型（但训练/测试数据适合内存）-aka分布式scikit-learn -那么您可以使用集群进行训练，并且骨架代码（根据dask_ml文档）如下所示
系统详细信息（用于执行代码）
dask==1.2.0
dask-ml==0.12.0
numpy==1.16.2+mkl
pandas==0.24.0
scikit-learn==0.20.3
sklearn==0.0
OS==Windows 8 (64-bit)
Python version (import platform; print(platform.python_version()))==3.7.2