7.10 参数调优

7.10.1 grid_search.GridSearchCV类

　　sklearn库提供了grid_search模块，用来对模型参数进行调优。grid search采用基于网格搜索的交叉验证法来选择模型参数，避免了参数选择的盲目性和随意性。
　　其中grid_search类的GridSearchCV方法实现了fit，predict，predict_proba等方法，并通过交叉验证对参数空间进行求解，寻找最佳的参数。使用方法如下：

sklearn.grid_search.GridSearchCV(estimator,
    param_grid, scoring=None, fit_params=None,
    n_jobs=1, iid=True, refit=True, cv=None,
    verbose=0, pre_dispatch='2*n_jobs', error_score='raise')

7.10.2 常用参数解读

　　estimator：所使用的分类器，如estimator=RandomForestClassifier(min_samples_split=100,min_samples_leaf=20,max_depth=8,max_features='sqrt',random_state=10), 并且传入除需要确定最佳的参数之外的其他参数。每一个分类器都需要一个scoring参数，或者score方法。
　　param_grid：值为字典或者列表，即需要最优化的参数的取值，param_grid =param_test1，param_test1 = {'n_estimators':range(10,71,10)}。
　　scoring :准确度评价标准，默认None,这时需要使用score函数；或者如scoring='roc_auc'，根据所选模型不同，评价准则不同。字符串（函数名），或是可调用对象，需要其函数签名形如：scorer(estimator, X, y)；如果是None，则使用estimator的误差估计函数。scoring参数选择如下：

Scoring	Function	Comment
accuracy	metrics.accuracy_score
average_precision	metrics.average_precision_score
f1	metrics.f1_score	for binary targets
f1_micro	metrics.f1_score	micro-averaged
f1_macro	metrics.f1_score	macro-averaged
f1_weighted	metrics.f1_score	weighted average
f1_samples	metrics.f1_score	by multilabel sample
neg_log_loss	metrics.log_loss	requires predict_proba support
precision	metrics.precision_score	suffixes apply as with ‘f1’
recall	metrics.recall_score	suffixes apply as with f1
roc_auc	metrics.roc_auc_score
adjusted_rand_score	metrics.adjusted_rand_score
neg_mean_absolute_error	metrics.mean_absolute_error
neg_mean_squared_error	metrics.mean_squared_error
neg_median_absolute_error	metrics.median_absolute_error
r2	metrics.r2_score

　　cv :交叉验证参数，默认None，使用三折交叉验证。指定fold数量，默认为3，也可以是yield训练/测试数据的生成器。
　　refit :默认为True,程序将会以交叉验证训练集得到的最佳参数，重新对所有可用的训练集与开发集进行，作为最终用于性能评估的最佳模型参数。即在搜索参数结束后，用最佳参数结果再次fit一遍全部数据集。
　　iid:默认True,为True时，默认为各个样本fold概率分布一致，误差估计为所有样本之和，而非各个fold的平均。
　　verbose：日志冗长度，int：冗长度，0：不输出训练过程，1：偶尔输出，>1：对每个子模型都输出。
　　n_jobs: 并行数，int：个数,-1：跟CPU核数一致, 1:默认值。
　　pre_dispatch：指定总共分发的并行任务数。当n_jobs大于1时，数据将在每个运行点进行复制，这可能导致OOM，而设置pre_dispatch参数，则可以预先划分总共的job数量，使数据最多被复制pre_dispatch次

7.10.3 常用方法

　　grid.fit()：运行网格搜索
　　grid_scores_：给出不同参数情况下的评价结果
　　best_params_：描述了已取得最佳结果的参数的组合
　　best_score_：成员提供优化过程期间观察到的最好的评分

7.10.4 代码实例

示例代码一：

#-*- coding:utf-8 -*-
import numpy as np
import pandas as pd
import scipy as sp
import copy,os,sys,psutil
import lightgbm as lgb
from lightgbm.sklearn import LGBMRegressor
from sklearn.model_selection import GridSearchCV
from sklearn.datasets import dump_svmlight_file
from svmutil import svm_read_problem

from sklearn import  metrics   #Additional scklearn functions
from sklearn.grid_search import GridSearchCV   #Perforing grid search

from featureProject.ly_features import make_train_set
from featureProject.my_import import split_data
# from featureProject.features import TencentReport
from featureProject.my_import import feature_importance2file


def print_best_score(gsearch,param_test):
     # 输出best score
    print("Best score: %0.3f" % gsearch.best_score_)
    print("Best parameters set:")
    # 输出最佳的分类器到底使用了怎样的参数
    best_parameters = gsearch.best_estimator_.get_params()
    for param_name in sorted(param_test.keys()):
        print("\t%s: %r" % (param_name, best_parameters[param_name]))

def lightGBM_CV():
    print ('获取内存占用率： '+(str)(psutil.virtual_memory().percent)+'%')
    data, labels = make_train_set(24000000,25000000)
    values = data.values;
    param_test = {
        'max_depth': range(5,15,2),
        'num_leaves': range(10,40,5),
    }
    estimator = LGBMRegressor(
        num_leaves = 50, # cv调节50是最优值
        max_depth = 13,
        learning_rate =0.1, 
        n_estimators = 1000, 
        objective = 'regression', 
        min_child_weight = 1, 
        subsample = 0.8,
        colsample_bytree=0.8,
        nthread = 7,
    )
    gsearch = GridSearchCV( estimator , param_grid = param_test, scoring='roc_auc', cv=5 )
    gsearch.fit( values, labels )
    gsearch.grid_scores_, gsearch.best_params_, gsearch.best_score_
    print_best_score(gsearch,param_test)


if __name__ == '__main__':
    lightGBM_CV()

示例代码二：

from __future__ import print_function
from pprint import pprint
from time import time
import logging

from sklearn.datasets import fetch_20newsgroups
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.linear_model import SGDClassifier
from sklearn.grid_search import GridSearchCV
from sklearn.pipeline import Pipeline

print(__doc__)

# Display progress logs on stdout
logging.basicConfig(level=logging.INFO,
                    format='%(asctime)s %(levelname)s %(message)s')


###############################################################################
# Load some categories from the training set
categories = [
    'alt.atheism',
    'talk.religion.misc',
]
# Uncomment the following to do the analysis on all the categories
#categories = None

print("Loading 20 newsgroups dataset for categories:")
print(categories)

data = fetch_20newsgroups(subset='train', categories=categories)
print("%d documents" % len(data.filenames))
print("%d categories" % len(data.target_names))
print()

###############################################################################
# 使用pipeline定义文本分类问题常见的工作流，包含向量化和一个简单的分类器
pipeline = Pipeline([
    ('vect', CountVectorizer()),
    ('tfidf', TfidfTransformer()),
    ('clf', SGDClassifier()),
])

# 参数空间：
# 定义了pipeline中各个模型的需要穷尽求解的参数空间，比如：clf__penalty': ('l2', 'elasticnet')
# 表示SGDClassifier分类器的正则化选项为L2和elasticnet，训练时模型会分别使用这两个正则化方法来寻求最佳的方式
parameters = {
    'vect__max_df': (0.5, 0.75, 1.0),
    #'vect__max_features': (None, 5000, 10000, 50000),
    'vect__ngram_range': ((1, 1), (1, 2)),  # unigrams or bigrams
    #'tfidf__use_idf': (True, False),
    #'tfidf__norm': ('l1', 'l2'),
    'clf__alpha': (0.00001, 0.000001),
    'clf__penalty': ('l2', 'elasticnet'),
    #'clf__n_iter': (10, 50, 80),
}

if __name__ == "__main__":

    # 通过GridSearchCV来寻求最佳参数空间
    grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, verbose=1)

    print("Performing grid search...")
    print("pipeline:", [name for name, _ in pipeline.steps])
    print("parameters:")
    pprint(parameters)
    t0 = time()

    # 这里只需调用一次fit函数就可以了
    grid_search.fit(data.data, data.target)
    print("done in %0.3fs" % (time() - t0))
    print()

    # 输出best score
    print("Best score: %0.3f" % grid_search.best_score_)
    print("Best parameters set:")
    # 输出最佳的分类器到底使用了怎样的参数
    best_parameters = grid_search.best_estimator_.get_params()
    for param_name in sorted(parameters.keys()):
        print("\t%s: %r" % (param_name, best_parameters[param_name]))

运行结果：

Loading 20 newsgroups dataset for categories:
['alt.atheism', 'talk.religion.misc']
1427 documents
2 categories

Performing grid search...
pipeline: ['vect', 'tfidf', 'clf']
parameters:
{'clf__alpha': (1.0000000000000001e-05, 9.9999999999999995e-07),
 'clf__n_iter': (10, 50, 80),
 'clf__penalty': ('l2', 'elasticnet'),
 'tfidf__use_idf': (True, False),
 'vect__max_n': (1, 2),
 'vect__max_df': (0.5, 0.75, 1.0),
 'vect__max_features': (None, 5000, 10000, 50000)}
done in 1737.030s

Best score: 0.940
Best parameters set:
    clf__alpha: 9.9999999999999995e-07
    clf__n_iter: 50
    clf__penalty: 'elasticnet'
    tfidf__use_idf: True
    vect__max_n: 2
    vect__max_df: 0.75
    vect__max_features: 50000