python数据分析学习笔记十

第十章 预测性分析和机器学习

监督学习

无监督学习

强化学习

 

1 scikit-learn

略

2 预处理

import numpy as np
from sklearn import preprocessing
from scipy.stats import anderson

# 加载数据
rain = np.load('rain.npy')
rain = .1 * rain
rain[rain < 0] = .05 / 2

# 期望值 标准差和安德森
print("Rain mean", rain.mean())
print("Rain Variance", rain.var())
print("Anderson Rain", anderson(rain))

scaled = preprocessing.scale(rain)
print("Scaled mean", scaled.mean())
print("Scaled Variance", scaled.var())
print("Anderson Scaled", anderson(scaled))

# 把特征值从数值型转换布尔型
binarized = preprocessing.binarize(rain)
print("binarized", np.unique(binarized), binarized.sum())

# 分类标准类别,输出0-62之间的整数
lb = preprocessing.LabelBinarizer()
lb.fit(rain.astype(int))
print(lb.classes_)

 

运行结果如下:

Rain mean 2.17919594267

Rain Variance 18.803443919

Anderson Rain AndersonResult(statistic=inf,critical_values=array([ 0.576, 0.656,  0.787,  0.918, 1.092]), significance_level=array([ 15. ,  10. ,  5. ,   2.5,   1. ]))

Scaled mean 3.41301602808e-17

Scaled Variance 1.0

Anderson ScaledAndersonResult(statistic=inf, critical_values=array([ 0.576,  0.656, 0.787,  0.918,  1.092]), significance_level=array([ 15.,  10. ,  5. ,   2.5,   1. ]))

binarized [ 0.  1.] 24594.0

[ 0 1  2  3 4  5  6 7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 2324

 2526 27 28 29 30 31 32 33 34 35 36 37 38 39 40 42 43 44 45 46 47 48 49 50

 5253 55 58 61 62]

 

3 基于逻辑回归的分类

该算法可以用以预测事件发生的概率,或是事物是否属于一类别的概率

from sklearn.linear_model import LogisticRegression
from sklearn.cross_validation import KFold
from sklearn import datasets
import numpy as np


def classify(x, y):
    # 使用逻辑回归进行分类
    clf = LogisticRegression(random_state=12)
    scores = []
    # k-折交叉验证
    kf = KFold(len(y), n_folds=10)
    # 检查分类的状确性
    for train, test in kf:
        clf.fit(x[train], y[train])
        scores.append(clf.score(x[test], y[test]))
    print(np.mean(scores))


# 加载数据信息
rain = np.load('rain.npy')
dates = np.load('doy.npy')
# 使用日期和降雨量来构建数组
x = np.vstack((dates[:-1], rain[:-1]))
# 无雨，小雨，雨
y = np.sign(rain[1:])
classify(x.T, y)

iris = datasets.load_iris()
x = iris.data[:, :2]
y = iris.target
classify(x, y)

 

运行结果如下:

0.576726256477

0.413333333333

 

4 基于支持向量机的分类

支持向量机 Support vector machines SVM

支持向量回归 Support vector Regression SVR

可以用来进行回归分析,也可以用来分类

示例代码如下:

from sklearn.svm import SVC
from sklearn.grid_search import GridSearchCV
from sklearn import datasets
import numpy as np
from pprint import PrettyPrinter


def classify(x, y):
    # 进行网格搜索
    clf = GridSearchCV(SVC(random_state=42, max_iter=100), {'kernel': ['linear', 'poly', 'rbf'], 'C': [1, 10]})
    clf.fit(x, y)
    print("Score", clf.score(x, y))
    PrettyPrinter().pprint(clf.grid_scores_)


rain = np.load('rain.npy')
dates = np.load('doy.npy')

x = np.vstack((dates[:-1], rain[:-1]))
y = np.sign(rain[1:])
classify(x.T, y)

iris = datasets.load_iris()
x = iris.data[:, :2]
y = iris.target
classify(x, y)

 

运行结果如下:

#天气数据

Score 0.559660687823

[mean: 0.42879, std: 0.11308, params:{'kernel': 'linear', 'C': 1},

 mean: 0.55570, std: 0.00559, params:{'kernel': 'poly', 'C': 1},

 mean: 0.36939, std: 0.00169, params:{'kernel': 'rbf', 'C': 1},

 mean: 0.30658, std: 0.03034, params:{'kernel': 'linear', 'C': 10},

 mean: 0.41673, std: 0.20214, params:{'kernel': 'poly', 'C': 10},

 mean: 0.49195, std: 0.08911, params:{'kernel': 'rbf', 'C': 10}]

 

#鸢属花样本数据

Score 0.82

[mean: 0.80000, std: 0.03949, params:{'kernel': 'linear', 'C': 1},

 mean: 0.58667, std: 0.12603, params:{'kernel': 'poly', 'C': 1},

 mean: 0.80000, std: 0.03254, params:{'kernel': 'rbf', 'C': 1},

 mean: 0.74667, std: 0.07391, params:{'kernel': 'linear', 'C': 10},

 mean: 0.56667, std: 0.13132, params:{'kernel': 'poly', 'C': 10},

 mean: 0.79333, std: 0.03467, params:{'kernel': 'rbf', 'C': 10}]

 

5 基于elasticNetCV的回归分类

弹性网格正则化 Elasic net Regularization  降低回归分析的过拟合风险

实际上是LASSO(The Least Absolute Shrikage and Selection Operator)算法和岭回归方法的线性组合

示例代码如下：

from sklearn.linear_model import ElasticNetCV
import numpy as np
from sklearn import datasets
import matplotlib.pyplot as plt


def regress(x, y, title):
    clf = ElasticNetCV(max_iter=200,  # 最大迭代次数
                       cv=10,  # 包总量
                       l1_ratio=[.1, .5, .7, .9, .95, .99, 1]  # 0表示只使用岭回归,1表示只使用 LASSO回归,否则使用混合算法
                       )
    clf.fit(x, y)
    print("Score", clf.score(x, y))

    pred = clf.predict(x)
    plt.title("Scatter plot of prediction and " + title)
    plt.xlabel("Prediction")
    plt.ylabel("Target")
    plt.scatter(y, pred)

    if "Boston" in title:
        plt.plot(y, y, label="Perfect Fit")
        plt.legend()
    plt.grid = True
    plt.show()


rain = .1 * np.load('rain.npy')
rain[rain < 0] = .05 / 2
dates = np.load("doy.npy")

x = np.vstack((dates[:-1], rain[:-1]))
y = rain[1:]
regress(x.T, y, "rain data")

boston = datasets.load_boston()
x = boston.data
y = boston.target
regress(x, y, "Boston house prices")

 

运行结果如下：

Score 0.0527838760942

Score 0.683143903455

 

 

6 支持向量回归

示例代码如下：

import numpy as np
from sklearn import datasets
from sklearn.model_selection import learning_curve
from sklearn.svm import SVR
from sklearn import preprocessing
import multiprocessing
import matplotlib.pyplot as plt


# 错误信息
# D:\Python35\lib\site-packages\sklearn\svm\base.py:220: ConvergenceWarning: Solver terminated early (max_iter=800).  Consider pre-processing your data with StandardScaler or MinMaxScaler.
#  % self.max_iter, ConvergenceWarning)
def regress(x, y, ncpus, title):
    X = preprocessing.scale(x)
    Y = preprocessing.scale(y)
    clf = SVR(max_iter=ncpus * 200)
    # 根据cpu数量创建作业数
    train_sizes, train_scores, test_scores = learning_curve(clf, X, Y, n_jobs=ncpus)

    # 求平均数,然后画出得分
    plt.figure()
    plt.title(title)
    plt.plot(train_sizes, train_scores.mean(axis=1), label="Train score")
    plt.plot(train_sizes, test_scores.mean(axis=1), '--', label="Test score")
    print("Max test score " + title, test_scores.max())
    plt.grid(True)
    plt.legend(loc='best')
    plt.show()


def main():
    rain = .1 * np.load('rain.npy')
    rain[rain < 0] = .05 / 2
    dates = np.load('doy.npy')

    x = np.vstack((dates[:-1], rain[:-1]))
    y = rain[1:]
    ncpus = multiprocessing.cpu_count()
    regress(x.T, y, ncpus, "Rain")

    boston = datasets.load_boston()
    x = boston.data
    y = boston.target
    regress(x, y, ncpus, "Boston")


if __name__ == '__main__':
    main()

 

运行结果如下：

Max test score Rain -0.0272088393925

Max test score Boston 0.662188537037

7 基于相似性传播算法的聚类分析

聚类分析就是把数据分成一些组,这些组就是所谓的聚类

聚类分析，属无监督学习

相似性传播 affinity propagation

 

示例代码如下:

# 生成三个数据块
x, _ = datasets.make_blobs(n_samples=100, centers=3, n_features=2, random_state=10)
# 创建矩阵
S = euclidean_distances(x)
# print(S)

# 根据矩阵,给数据标注其所属聚类
aff_pro = cluster.AffinityPropagation().fit(S)
labels = aff_pro.labels_

# 绘制图形
styles = ['o', 'x', '^']
for style, label in zip(styles, np.unique(labels)):
    print(label)
    plt.plot(x[labels == label], style, label=label)

plt.title("Clustering Blobs")
plt.grid(True)
plt.legend(loc='best')
plt.show()

 

运行结果如下:

0

1

2

 

 

8 均值漂移算法

一种不需要估算聚类数的聚类算法(可以应用于图像处理,是不是等同于中值滤波)

示例代码如下:

import numpy as np
from sklearn import cluster
import matplotlib.pyplot as plt
import pandas as pd

# 加载数据
rain = .1 * np.load('rain.npy')
rain[rain < 0] = .05 / 2
dates = np.load('doy.npy')
x = np.vstack((dates, rain))

# 创建dataFrame,并计算平均值
df = pd.DataFrame.from_records(x.T, columns=['dates', 'rain'])
df = df.groupby('dates').mean()
df.plot()

# 均值漂移算法
x = np.vstack((np.arange(1, len(df) + 1), df.as_matrix().ravel()))
x = x.T
ms = cluster.MeanShift()
ms.fit(x)
labels = ms.predict(x)

# 绘制图形
plt.figure()
grays = ['0', '0.5', '0.75']

for gray, label in zip(grays, np.unique(labels)):
    match = labels == label
    x0 = x[:, 0]
    x1 = x[:, 1]
    plt.plot(x0[match], x1[match], lw=label + 1, label=label)
    plt.fill_between(x0, x1, where=match, color=gray)

plt.grid(True)
plt.legend()
plt.show()

 

运行结果如下:

 

 

9 遗传算法

可用于搜索和优化方面

示例代码如下:

 

运行结果如下:

Gen(交叉概率)        nevals(突变率)         max(最大代数)       

0    400        0.000484774

1    222        0.000656187

2    246        0.00745961

3    239        0.00745961

4    240        0.0184182 

5    216        0.0309736 

6    237        0.06957   

7    243        0.06957   

8    231        0.224381  

9    226        0.224381  

10   247        0.224381  

11   228        0.247313  

12   241        0.28318   

13   242        0.354144  

14   246        0.46282   

15   239        0.46282   

16   266        0.480937  

17   233        0.648529  

 ....

76   230        0.998861  

77   252        0.998861  

78   232        0.998861  

79   243        0.998861  

80   235        0.998861  

0.9988605380058289

 

 

10 神经网络

人工神经网络 ANN: 由神经元组成的网络,每个神经元都有输入和输出功能

程序报错,需要调试

 

11 决策树

示例代码如下:

# 所属模块发生变化
# from sklearn.cross_validation import train_test_split
# from sklearn.grid_search import RandomizedSearchCV
from sklearn.model_selection import train_test_split
from sklearn import tree
from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import randint as sp_randint
import pydot
# import StringIO
from io import StringIO
import numpy as np
from tempfile import NamedTemporaryFile

# 加载数据信息
rain = .1 * np.load('rain.npy')
rain[rain < 0] = .05 / 2

dates = np.load('doy.npy').astype(int)
x = np.vstack((dates[:-1], np.sign(rain[:-1])))
x = x.T

y = np.sign(rain[1:])

# 创建测试集和训练集数据
x_tain, x_test, y_train, y_test = train_test_split(x, y, random_state=37)

# 验证各参数的取值范围
clf = tree.DecisionTreeClassifier(random_state=37)
params = {"max_depth": [2, None],
          "min_samples_leaf": sp_randint(1, 5),
          "criterion": ["gini", "entropy"]}
rscv = RandomizedSearchCV(clf, params)
rscv.fit(x_tain, y_train)

# 绘制决策树的对象
sio = StringIO()
tree.export_graphviz(rscv.best_estimator_, out_file=sio, feature_names=['day-of-year', 'yest'])
dec_tree = pydot.graph_from_dot_data(sio.getvalue())
with NamedTemporaryFile(prefix='rain', suffix='.png', delete=False) as f:
    # dec_tree.write_png(f.name)
    dec_tree[0].write_png(f.name)
    print("Written figure to", f.name)

print('Best Train Score', rscv.best_score_)
print('Test Score', rscv.score(x_test, y_test))
print("Best params", rscv.best_params_)

 

运行结果如下:

Written figure toC:\Users\ADMINI~1\AppData\Local\Temp\rainmys2nqfh.png

Best Train Score 0.703164923517

Test Score 0.705058763413

Best params {'min_samples_leaf': 1,'criterion': 'entropy', 'max_depth': 2}