数据分析(4)sklearn入门

如何选择机器学习方法

http://scikit-learn.org/stable/tutorial/machine_learning_map/index.html

通用学习模式

只需要先定义 用什么model学习,然后再 model.fit(数据), 这样 model 就能从数据中学到东西. 最后还可以用 model.predict() 来预测值.

from sklearn import datasets
from sklearn.cross_validation import train_test_split
from sklearn.neighbors import KNeighborsClassifier
iris &＃61; datasets.load_iris()
iris_X &＃61; iris.data
iris_Y &＃61; iris.target
&＃39;&＃39;&＃39;
输入有四个属性&＃xff1a;[[ 5.1 3.5 1.4 0.2] [ 4.9 3. 1.4 0.2] ...]
输出类别&＃xff1a;[0 0 0 ... 1 1 1 ... 2 2 2 ...]
&＃39;&＃39;&＃39;
X_train,X_test,ytrain,y_test &＃61; train_test_split(iris_X,iris_Y,test_size&＃61;0.3) # 顺序也被打乱&＃xff0c;按7:3
knn &＃61; KNeighborsClassifier()
knn.fit(X_train,ytrain) # 训练
print(knn.predict(X_test)) # 预测
print(y_test)

sklearn 的 datasets 数据库

Sklearn 提供了很多的有用的数据库,既有真实数据也有你可以编造的数据!特别的强大.http://scikit-learn.org/stable/modules/classes.html#module-sklearn.datasets

from sklearn import datasets
from sklearn.linear_model import LinearRegression
import matplotlib.pyplot as plt
loaded_data &＃61; datasets.load_boston()
data_X &＃61; loaded_data.data
data_y &＃61; loaded_data.target
model &＃61; LinearRegression() # 里面有参数可以改变
model.fit(data_X,data_y)
print(model.predict(data_X[:4,:]))
print(data_y[:4])
&＃39;&＃39;&＃39;
[ 30.00821269 25.0298606 30.5702317 28.60814055]
[ 24. 21.6 34.7 33.4]
&＃39;&＃39;&＃39;
X, y &＃61; datasets.make_regression(n_samples&＃61;100, n_features&＃61;1, n_targets&＃61;1, noise&＃61;10)
plt.scatter(X, y)
plt.show()

model 常用属性和功能

# y &＃61; 0.1x &＃43; 0.3
print(model.coef_) # 输出0.1
print(model.intercept_) # 输出0.3
print(model.get_params()) # 返回给model认定的参数&＃xff0c;比如{&＃39;copy_X&＃39;: True, &＃39;n_jobs&＃39;: 1, &＃39;normalize&＃39;: False, &＃39;fit_intercept&＃39;: True}
print(model.score(data_X, data_y)) # R^2 coefficient of determination

normalization 标准化数据

normalization 在数据跨度不一的情况下对机器学习有很重要的作用.特别是各种数据属性还会互相影响的情况之下. Scikit-learn 中标准化的语句是 preprocessing.scale() . scale 以后, model 就更能从标准化数据中学到东西.

from sklearn.model_selection import train_test_split
from sklearn.datasets.samples_generator import make_classification
from sklearn.svm import SVC
import matplotlib.pyplot as plt
X, y &＃61; make_classification(n_samples&＃61;300, n_features&＃61;2 , n_redundant&＃61;0, n_informative&＃61;2,random_state&＃61;22, n_clusters_per_class&＃61;1, scale&＃61;100)
plt.scatter(X[:, 0], X[:, 1], c&＃61;y)
plt.show()
X &＃61; preprocessing.scale(X) # normalization step
plt.scatter(X[:, 0], X[:, 1], c&＃61;y)
plt.show()
X_train, X_test, y_train, y_test &＃61; train_test_split(X, y, test_size&＃61;.3)
clf &＃61; SVC()
clf.fit(X_train, y_train)
print(clf.score(X_test, y_test)) # 0.944444444444

cross validation 交叉验证1
sklearn 中的 cross validation 交叉验证 对于我们选择正确的 model 和model 的参数是非常有帮助的. 有了他的帮助, 我们能直观的看出不同 model 或者参数对结构准确度的影响.

from sklearn.datasets import load_iris
from sklearn.cross_validation import train_test_split,cross_val_score
from sklearn.neighbors import KNeighborsClassifier
iris &＃61; load_iris()
X &＃61; iris.data
y &＃61; iris.target
# test train split
X_train, X_test, y_train, y_test &＃61; train_test_split(X, y, random_state&＃61;4)
knn &＃61; KNeighborsClassifier(n_neighbors&＃61;5)
knn.fit(X_train, y_train)
y_pred &＃61; knn.predict(X_test)
print(knn.score(X_test, y_test)) # 0.973684210526
# this is cross_val_score
knn &＃61; KNeighborsClassifier(n_neighbors&＃61;5)
scores &＃61; cross_val_score(knn, X, y, cv&＃61;5, scoring&＃61;&＃39;accuracy&＃39;)
print(scores) # [ 0.96666667 1. 0.93333333 0.96666667 1. ]
print(scores.mean()) # 0.973333333333

import matplotlib.pyplot as plt
k_range &＃61; range(1, 31)
k_scores &＃61; []
for k in k_range:knn &＃61; KNeighborsClassifier(n_neighbors&＃61;k)# loss &＃61; -cross_val_score(knn, X, y, cv&＃61;10, scoring&＃61;&＃39;mean_squared_error&＃39;) # for regressionscores &＃61; cross_val_score(knn, X, y, cv&＃61;10, scoring&＃61;&＃39;accuracy&＃39;) # for classificationk_scores.append(scores.mean())
plt.plot(k_range, k_scores)
plt.xlabel(&＃39;Value of K for KNN&＃39;)
plt.ylabel(&＃39;Cross-Validated Accuracy&＃39;)
plt.show()

cross validation 交叉验证2
sklearn.learning_curve 中的 learning curve 可以很直观的看出我们的 model 学习的进度,对比发现有没有 overfitting 的问题.然后我们可以对我们的 model 进行调整,克服 overfitting 的问题.

from sklearn.learning_curve import learning_curve
from sklearn.datasets import load_digits
from sklearn.svm import SVC
import matplotlib.pyplot as plt
import numpy as np
digits &＃61; load_digits()
X &＃61; digits.data
y &＃61; digits.target
train_sizes, train_loss, test_loss&＃61; learning_curve(SVC(gamma&＃61;0.01), X, y, cv&＃61;10, scoring&＃61;&＃39;mean_squared_error&＃39;,train_sizes&＃61;[0.1, 0.25, 0.5, 0.75, 1])
train_loss_mean &＃61; -np.mean(train_loss, axis&＃61;1)
test_loss_mean &＃61; -np.mean(test_loss, axis&＃61;1)
plt.plot(train_sizes, train_loss_mean, &＃39;o-&＃39;, color&＃61;"r",label&＃61;"Training")
plt.plot(train_sizes, test_loss_mean, &＃39;o-&＃39;, color&＃61;"g",label&＃61;"Cross-validation")
plt.xlabel("Training examples")
plt.ylabel("Loss")
plt.legend(loc&＃61;"best")
plt.show()

cross validation 交叉验证3
连续三节的 cross validation让我们知道在机器学习中 validation 是有多么的重要, 这一次的 sklearn 中我们用到了 sklearn.learning_curve 当中的另外一种, 叫做 validation_curve, 用这一种 curve 我们就能更加直观看出改变 model 中的参数的时候有没有 overfitting 的问题了.这也是可以让我们更好的选择参数的方法.

from sklearn.learning_curve import validation_curve
from sklearn.datasets import load_digits
from sklearn.svm import SVC
import matplotlib.pyplot as plt
import numpy as np
digits &＃61; load_digits()
X &＃61; digits.data
y &＃61; digits.target
param_range &＃61; np.logspace(-6, -2.3, 5)
train_loss, test_loss &＃61; validation_curve(SVC(), X, y, param_name&＃61;&＃39;gamma&＃39;, param_range&＃61;param_range, cv&＃61;10,scoring&＃61;&＃39;mean_squared_error&＃39;)
train_loss_mean &＃61; -np.mean(train_loss, axis&＃61;1)
test_loss_mean &＃61; -np.mean(test_loss, axis&＃61;1)
plt.plot(param_range, train_loss_mean, &＃39;o-&＃39;, color&＃61;"r",label&＃61;"Training")
plt.plot(param_range, test_loss_mean, &＃39;o-&＃39;, color&＃61;"g",label&＃61;"Cross-validation")
plt.xlabel("gamma")
plt.ylabel("Loss")
plt.legend(loc&＃61;"best")
plt.show()

Save
练习好了一个 model 以后总需要保存和再次预测, 所以保存和读取我们的 sklearn model 也是同样重要的一步.

from sklearn import svm
from sklearn import datasets
clf &＃61; svm.SVC()
iris &＃61; datasets.load_iris()
X, y &＃61; iris.data, iris.target
clf.fit(X, y)
# method 1: pickle
import pickle
# save
with open(&＃39;save/clf.pickle&＃39;, &＃39;wb&＃39;) as f:pickle.dump(clf, f)
# restore
with open(&＃39;save/clf.pickle&＃39;, &＃39;rb&＃39;) as f:clf2 &＃61; pickle.load(f)print(clf2.predict(X[0:1]))
# method 2: joblib
from sklearn.externals import joblib
# Save
joblib.dump(clf, &＃39;save/clf.pkl&＃39;)
# restore
clf3 &＃61; joblib.load(&＃39;save/clf.pkl&＃39;)
print(clf3.predict(X[0:1]))