python︱sklearn一些小技巧的记录(训练集划分/pipelline/交叉验证等)

机器学习︱R+python 专栏收录该内容
75 篇文章 6 订阅

sklearn里面包含内容太多,所以一些实用小技巧还是挺好用的。
sklearn.cross_validation 如果没有了,则需要使用 sklearn.model_selection



1、LabelEncoder

简单来说 LabelEncoder 是对不连续的数字或者文本进行编号

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
le.fit([1,5,67,100])
le.transform([1,1,100,67,5])

输出: array([0,0,3,2,1])


2、OneHotEncoder

OneHotEncoder 用于将表示分类的数据扩维:

from sklearn.preprocessing import OneHotEncoder
ohe = OneHotEncoder()
ohe.fit([[1],[2],[3],[4]])
ohe.transform([2],[3],[1],[4]).toarray()

输出:[ [0,1,0,0] , [0,0,1,0] , [1,0,0,0] ,[0,0,0,1] ]
正如keras中的keras.utils.to_categorical(y_train, num_classes)

来看一个笔者自己写的关于:dataframe onehot的小函数:

from sklearn.preprocessing import OneHotEncoder
import pandas as pd

#原始数据集
data = [['自有房',40,50000],
        ['无自有房',22,13000],
        ['自有房',30,30000]]
data = pd.DataFrame(data,columns=['house','age','income'])

def dataframe_one_hot(data,select_col = 'house'):

    word2id = {ds:n for n,ds in enumerate(set(data[select_col]))}
    id2word = {n:ds for n,ds in enumerate(set(data[select_col]))}
    data[select_col] = data[select_col].map(lambda x : word2id[x] )
    
    
    onehot_data = data[[select_col]]
    enc = OneHotEncoder(categories='auto')
    enc.fit(onehot_data)
    onehot_data = enc.transform(onehot_data).toarray()
    onehot_data = pd.DataFrame(onehot_data )
    onehot_data.columns = [id2word[od] for od in onehot_data.columns]
    
    #for od in onehot_data.columns:
    #    data[od] = onehot_data[od]
    return onehot_data

onehot_data = dataframe_one_hot(data,select_col = 'house')

最后的结果如下:
在这里插入图片描述


.

3、sklearn.model_selection.train_test_split随机划分训练集和测试集

一般形式:
train_test_split是交叉验证中常用的函数,功能是从样本中随机的按比例选取train data和testdata,形式为:

X_train,X_test, y_train, y_test =
cross_validation.train_test_split(train_data,train_target,test_size=0.4, random_state=0)

参数解释:

  • train_data:所要划分的样本特征集
  • train_target:所要划分的样本结果
  • test_size:样本占比,如果是整数的话就是样本的数量
  • random_state:是随机数的种子。
  • 随机数种子:其实就是该组随机数的编号,在需要重复试验的时候,保证得到一组一样的随机数。比如你每次都填1,其他参数一样的情况下你得到的随机数组是一样的。但填0或不填,每次都会不一样。
    随机数的产生取决于种子,随机数和种子之间的关系遵从以下两个规则:
  • 种子不同,产生不同的随机数;种子相同,即使实例不同也产生相同的随机数。
from sklearn.cross_validation import train_test_split  
train= loan_data.iloc[0: 55596, :]  
test= loan_data.iloc[55596:, :]  
# 避免过拟合,采用交叉验证,验证集占训练集20%,固定随机种子(random_state)  
train_X,test_X, train_y, test_y = train_test_split(train,  
                                                   target,  
                                                   test_size = 0.2,  
                                                   random_state = 0)  
train_y= train_y['label']  
test_y= test_y['label']  

附加:shuffle 一键随机打乱:

from sklearn.utils import shuffle
shuffle([1,2,3])
>>>[1, 3, 2]

shuffle随机打乱

.


4、pipeline

本节参考与文章:用 Pipeline 将训练集参数重复应用到测试集
pipeline 实现了对全部步骤的流式化封装和管理,可以很方便地使参数集在新数据集上被重复使用。

pipeline 可以用于下面几处:

  • 模块化 Feature Transform,只需写很少的代码就能将新的 Feature 更新到训练集中。
  • 自动化 Grid Search,只要预先设定好使用的 Model 和参数的候选,就能自动搜索并记录最佳的 Model。
  • 自动化 Ensemble Generation,每隔一段时间将现有最好的 K 个 Model 拿来做 Ensemble。

问题是要对数据集 Breast Cancer Wisconsin 进行分类,
它包含 569 个样本,第一列 ID,第二列类别(M=恶性肿瘤,B=良性肿瘤),
第 3-32 列是实数值的特征。

from pandas as pd
from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import LabelEncoder

df = pd.read_csv('https://archive.ics.uci.edu/ml/machine-learning-databases/'
                 'breast-cancer-wisconsin/wdbc.data', header=None)
                                 # Breast Cancer Wisconsin dataset

X, y = df.values[:, 2:], df.values[:, 1]

encoder = LabelEncoder()
y = encoder.fit_transform(y)
                    >>> encoder.transform(['M', 'B'])
                    array([1, 0])
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=0)

我们要用 Pipeline 对训练集和测试集进行如下操作:

  • 先用 StandardScaler 对数据集每一列做标准化处理,(是 transformer)
  • 再用 PCA 将原始的 30 维度特征压缩的 2 维度,(是 transformer)
  • 最后再用模型 LogisticRegression。(是 Estimator)
  • 调用 Pipeline 时,输入由元组构成的列表,每个元组第一个值为变量名,元组第二个元素是 sklearn 中的 transformer
    或 Estimator。

注意中间每一步是 transformer,即它们必须包含 fit 和 transform 方法,或者 fit_transform。
最后一步是一个 Estimator,即最后一步模型要有 fit 方法,可以没有 transform 方法。

然后用 Pipeline.fit对训练集进行训练,pipe_lr.fit(X_train, y_train)
再直接用 Pipeline.score 对测试集进行预测并评分 pipe_lr.score(X_test, y_test)

from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.linear_model import LogisticRegression

from sklearn.pipeline import Pipeline

pipe_lr = Pipeline([('sc', StandardScaler()),
                    ('pca', PCA(n_components=2)),
                    ('clf', LogisticRegression(random_state=1))
                    ])
pipe_lr.fit(X_train, y_train)
print('Test accuracy: %.3f' % pipe_lr.score(X_test, y_test))

                # Test accuracy: 0.947

还可以用来选择特征:

例如用 SelectKBest 选择特征,
分类器为 SVM,

anova_filter = SelectKBest(f_regression, k=5)
clf = svm.SVC(kernel='linear')
anova_svm = Pipeline([('anova', anova_filter), ('svc', clf)])

当然也可以应用 K-fold cross validation:

model = Pipeline(estimators)
seed = 7
kfold = KFold(n_splits=10, random_state=seed)
results = cross_val_score(model, X, Y, cv=kfold)
print(results.mean())

Pipeline 的工作方式

当管道 Pipeline 执行 fit 方法时,
首先 StandardScaler 执行 fit 和 transform 方法,
然后将转换后的数据输入给 PCA,
PCA 同样执行 fit 和 transform 方法,
再将数据输入给 LogisticRegression,进行训练。
这里写图片描述

参考:
python 数据处理中的 LabelEncoder 和 OneHotEncoder
sklearn 中的 Pipeline 机制
用 Pipeline 将训练集参数重复应用到测试集


5 稀疏矩阵合并

from scipy.sparse import hstack
x_train_spar = hstack([x_train_ens_s, x_train_tfidf])
hstack([X, csr_matrix(char_embed), csr_matrix(word_embed)], format='csr')

多项式PolynomialFeatures,有点类似FFM模型,可以组合特征

poly = PolynomialFeatures(degree=2, interaction_only=True, include_bias=False)
#degree控制多项式最高次数
x_train_new = poly.fit_transform(x_train)

6 sklearn中的交叉验证


X = np.array([[1, 2,5,6,6,6], [3, 4], [1, 2], [3, 4]])
y = np.array([1, 2, 3, 4])
kf = KFold(n_splits=2, random_state=None, shuffle=False)
kf.get_n_splits(X)

print(kf)  
for train_index, test_index in kf.split(X):
    print("TRAIN:", train_index, "TEST:", test_index)
    X_train, X_test = X[train_index], X[test_index]
    y_train, y_test = y[train_index], y[test_index]


>>> KFold(n_splits=2, random_state=None, shuffle=False)
>>> TRAIN: [2 3] TEST: [0 1]
>>> TRAIN: [0 1] TEST: [2 3]

KFold是将X,Y分成两份,x样本个数为4,则随机抽2次,每次2个样本。Y为长度为4的List,也对应X进行划分。
kf.split(X) 就等于[(array([2, 3]), array([0, 1])), (array([0, 1]), array([2, 3]))],这个是index,所以需要从X之中根据Index取出数据。

如果用在文本分类之中,在此举两个例子:

来源于达观杯的实践


# daguan-classify-2018-master\src\model.py
def stacking_first(train, train_y, test):
    savepath = './stack_op{}_dt{}_tfidf{}/'.format(args.option, args.data_type, args.tfidf)
    os.makedirs(savepath, exist_ok=True)

    count_kflod = 0
    num_folds = 6
    kf = KFold(n_splits=num_folds, shuffle=True, random_state=10)
    predict = np.zeros((test.shape[0], config.n_class))
    oof_predict = np.zeros((train.shape[0], config.n_class))
    scores = []
    f1s = []

    for train_index, test_index in kf.split(train):

        kfold_X_train = {}
        kfold_X_valid = {}

        y_train, y_test = train_y[train_index], train_y[test_index]

        kfold_X_train, kfold_X_valid = train[train_index], train[test_index]

        model_prefix = savepath + 'DNN' + str(count_kflod)
        if not os.path.exists(model_prefix):
            os.mkdir(model_prefix)

        M = 4  # number of snapshots
        alpha_zero = 1e-3  # initial learning rate
        snap_epoch = 16
        snapshot = SnapshotCallbackBuilder(snap_epoch, M, alpha_zero)

        res_model = get_model(train)
        res_model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

        # res_model.fit(train_x, train_y, batch_size=BATCH_SIZE, epochs=EPOCH, verbose=1,  class_weight=class_weight)
        res_model.fit(kfold_X_train, y_train, batch_size=BATCH_SIZE, epochs=snap_epoch, verbose=1,
                      validation_data=(kfold_X_valid, y_test),
                      callbacks=snapshot.get_callbacks(model_save_place=model_prefix))

        evaluations = []
        for i in os.listdir(model_prefix):
            if '.h5' in i:
                evaluations.append(i)

        preds1 = np.zeros((test.shape[0], config.n_class))
        preds2 = np.zeros((len(kfold_X_valid), config.n_class))
        for run, i in enumerate(evaluations):
            res_model.load_weights(os.path.join(model_prefix, i))
            preds1 += res_model.predict(test, verbose=1) / len(evaluations)
            preds2 += res_model.predict(kfold_X_valid, batch_size=128) / len(evaluations)

        predict += preds1 / num_folds
        oof_predict[test_index] = preds2

        accuracy = mb.cal_acc(oof_predict[test_index], np.argmax(y_test, axis=1))
        f1 = mb.cal_f_alpha(oof_predict[test_index], np.argmax(y_test, axis=1), n_out=config.n_class)
        print('the kflod cv is : ', str(accuracy))
        print('the kflod f1 is : ', str(f1))
        count_kflod += 1
        scores.append(accuracy)
        f1s.append(f1)
    print('total scores is ', np.mean(scores))
    print('total f1 is ', np.mean(f1s))
    return predict

来源于:kaggle恶意评价比赛的实践


# https://github.com/sermakarevich/jigsaw-toxic-comment-classification-challenge/blob/master/bin/models.py
def predict(self):
    test_number = 1
    for train_i, valid_i in self.cv.split(self.x_train, self.y_train):
        model = self.get_model(**self.model_kwargs)
        x_train = self.x_train[train_i]
        y_train = self.y_train[train_i]
        x_valid = self.x_train[valid_i]
        y_valid = self.y_train[valid_i]
        for i in self.epochs:
            model.fit(x_train, y_train, epochs = 1, batch_size = self.batch_size)
            train_prediction = model.predict(x_valid, self.batch_size * 2)
            print (f'test_number: {test_number}, epoch: {i}, score: {self.scorrer(y_valid, train_prediction)}')
        test_prediction =  model.predict(self.x_test, self.batch_size * 2)
        self.train_predictions.append([train_prediction, valid_i])
        self.test_predictions.append(test_prediction)
        self.score.append(self.scorrer(y_valid, train_prediction))
        print (f"test_number: {test_number}, avg score: {self.score[-1]}")
        test_number += 1
    print (np.mean(self.score))
    self.train_predictions = (
        pd.concat([pd.DataFrame(data=i[0],index=i[1], columns=[self.col_names]) 
                   for i in self.train_predictions]).sort_index())

**公众号“素质云笔记”定期更新博客内容:**
![这里写图片描述](https://imgconvert.csdnimg.cn/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTgwMjI2MTU1MzQ4NTQ1?x-oss-process=image/format,png)
  • 10
    点赞
  • 1
    评论
  • 38
    收藏
  • 一键三连
    一键三连
  • 扫一扫,分享海报

相关推荐
©️2020 CSDN 皮肤主题: 游动-白 设计师:白松林 返回首页
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、C币套餐、付费专栏及课程。

余额充值