github:https://github.com/gaowanlu/MachineLearning-DeepLearning
Github 机器学习与深度学习例程 项目
依赖:
tensorflow、matplotlib、numpy、sklearn、pickle
关键词
SGDClassifier 、RandomForestClassifier 、PR曲线、精准率、召回率、F1值、ROC曲线
效果
不同阈值下的精准率与召回率
PR曲线
SDG ROC曲线
SDG与随机森林 ROC曲线对比
代码
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
from sklearn.linear_model import SGDClassifier
from sklearn.preprocessing import StandardScaler
import pickle
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
print(len(x_train))#60000个数据集
print(len(y_test))#10000个测试集
#print(x_train[0]) #二维List
some_digit = x_train[0]
some_digit_image = some_digit.reshape(28, 28)
#plt.imshow(some_digit_image, cmap="binary")
#plt.axis("off")
#plt.show()
y_train = y_train.astype(np.uint8)
y_test = y_test.astype(np.uint8)
print("x_train is ",y_train[0])
#转变数据集形式
x_train_transed=[]
for index in range(len(x_train)):
x_train_transed.append(x_train[index].reshape(-1))
#print("疯狂转变中 ",(index/len(x_train))*100,"%")
print("转变完了")
sgd_clf = SGDClassifier(max_iter=1000, tol=1e-3, random_state=42)
#将数字标签转换为bool型标签,List内item的转换
y_train_5 = (y_train == 5)
print("训练中")
sgd_clf.fit(x_train_transed,y_train_5)
print("训练完了")
print("正在保存模型")
with open('./5_image_test.model', 'wb') as fw:
pickle.dump(sgd_clf, fw)
print("正在加载模型")
with open('./5_image_test.model','rb') as fr:
test_5_model=pickle.load(fr)
print("使用模型中")
for i in range(10):
some_digit_image = x_train[i].reshape(28, 28)
#plt.imshow(some_digit_image, cmap="binary")
print("image smaple ",i,"predict result ,is 5 :",\
test_5_model.predict([x_train_transed[i]]))
#plt.axis("off")
#plt.show()
#交叉验证
from sklearn.model_selection import StratifiedKFold
from sklearn.base import clone
#StratifiedKFold分层交叉验证法
skfolds =StratifiedKFold(n_splits=3,random_state=42,shuffle=True)
#分成三份 两份作为训练集 一份作为测试集
for train_index ,test_index in skfolds.split(x_train_transed,y_train_5):
print(len(train_index),len(test_index))
clone_clf=clone(sgd_clf)
x_train_folds=np.array(x_train_transed)[train_index]
y_train_folds=np.array(y_train_5)[train_index]
x_test_fold=np.array(x_train_transed)[test_index]
y_test_fold=np.array(y_train_5)[test_index]
clone_clf.fit(x_train_folds,y_train_folds)
y_pred=clone_clf.predict(x_test_fold)
n_correct=sum(y_pred==y_test_fold)
print(n_correct/len(y_pred))
#使用cross_val_score进行交叉验证
from sklearn.model_selection import cross_val_score
result=cross_val_score(sgd_clf,x_train_transed,y_train_5,cv=3,scoring="accuracy")
print("cross_val_score result ",result)
#准确率通常无法成为分类器的首要性能指标，特别是处理有偏数据集时，
# 大约有10%的图片为5，则随意一张图片不是5，则有90%的概率都是猜正确了
#混淆矩阵
from sklearn.model_selection import cross_val_predict
from sklearn.metrics import confusion_matrix
y_train_pred=cross_val_predict(sgd_clf,x_train_transed,y_train_5,cv=3)
#cross_val_predict同样执行k-折交叉验证，返回的不是评估分数，而是每个折叠的预测
#这意味着对于每个实例都可以得到一个干净的预测
# (干净的意思是模型预测时使用的数据在其训练期间从未见过)
print(len(y_train_pred))
matrix=confusion_matrix(y_train_5,y_train_pred)
print(matrix)
# [[53892 687] 53892非5预测为非5 687非5被预测为5
# [ 1891 3530]] 1891 5预测为非5 3530 5预测为5
'''
精度=（TP）/（TP+FP）T原来为正 P预测为正 F原来为负 N预测为负
召回率=(TP)/(TP+FN)
F1=(2)/((1/精度)+(1/召回率))
'''
from sklearn.metrics import precision_score,recall_score,f1_score
precisionScore=precision_score(y_train_5,y_train_pred)
recallScore=recall_score(y_train_5,y_train_pred)
f1Score=f1_score(y_train_5,y_train_pred)
print("精度：",precisionScore,"召回率：",recallScore,"F1 ：",f1Score)
#如何在精度与召回率之间做抉择
#阈值调整调参优化
y_scores=sgd_clf.decision_function([x_train_transed[0]])
print(y_scores)
#设置阈值为0
threshold=0
temp_pred=(y_scores>threshold)#只留下分数大于阈值的项
print(temp_pred)
#设置阈值为8000
print((y_scores>8000))
#使用cross_val_predict()函数获取训练集中所有实例的分数
y_scores=cross_val_predict(sgd_clf,x_train_transed,y_train_5,cv=3,\
method="decision_function")
#使用precision_recall_curve()函数来计算可能的阈值的精度和召回率
from sklearn.metrics import precision_recall_curve
precisions,recalls,thresholds=precision_recall_curve(y_train_5,y_scores)
def plot_precision_recall_vs_threshold(precisions, recalls, thresholds):
plt.plot(thresholds, precisions[:-1], "b--", label="Precision")
plt.plot(thresholds, recalls[:-1], "g-", label="Recall")
[...] # highlight the threshold and add the legend, axis label, and grid
plot_precision_recall_vs_threshold(precisions, recalls, thresholds)
plt.show()
#假设要将精度设为90%，可以找到精度90%的最低阈值
#阈值越高 召回率越低
threshold_90_percent=thresholds[np.argmax(precisions>=0.9)]
print("保证精度90%以上的最低阈值 ",threshold_90_percent)
#则有了分数阈值，我们也可以使用分数阈值 进行二分类
#Y_train_pred_90=(y_scores>=threshold_90_percent)
#大于阈值可预测为是5
#PR曲线
#画y:精度 x：召回率图像
def plot_precision_vs_recall(precisions, recalls):
plt.plot(recalls, precisions, "b-", linewidth=2)
plt.xlabel("Recall", fontsize=16)
plt.ylabel("Precision", fontsize=16)
plt.axis([0, 1, 0, 1])
plt.grid(True)
plot_precision_vs_recall(precisions,recalls)
plt.show()
#ROC曲线(特征曲线)
#真正类率(召回率)和假正类率(FPR)
#FPR=1-TNR
#TNR:正确分类为负类的负类实例比率
from sklearn.metrics import roc_curve
fpr,tpr,thresholds=roc_curve(y_train_5,y_scores)
#画ROC曲线
#y:真正率(召回率) x:假正率
def plot_roc_curve(fpr, tpr, label=None):
plt.plot(fpr, tpr, linewidth=2, label=label)
plt.plot([0, 1], [0, 1], 'k--') # Dashed diagonal
[...] # Add axis labels and grid
plot_roc_curve(fpr, tpr)
#绘制了所有可能阈值的假正率与真正率的关系
plt.show()
#计算ROC曲线下的面积，最理想的ROC曲线下面积为1
#纯随机分类器ROC AUC为0.5
from sklearn.metrics import roc_auc_score
roc_auc_score(y_train_5,y_scores)
#换个随机森林分类器来看看ROC曲线
from sklearn.ensemble import RandomForestClassifier
forest_clf=RandomForestClassifier(random_state=42)
y_probas_forest = cross_val_predict(forest_clf, x_train_transed, y_train_5, cv=3,
method="predict_proba")
#得到的不是分数，是概率，是5的概率
y_scores_forest = y_probas_forest[:, 1] # score = proba of positive class
fpr_forest, tpr_forest, thresholds_forest = roc_curve(y_train_5,y_scores_forest)
plt.plot(fpr, tpr, "b:", label="SGD")
plot_roc_curve(fpr_forest, tpr_forest, "Random Forest")
plt.legend(loc="lower right")
plt.show()
roc_auc_score(y_train_5,y_scores_forest)#计算ROC曲线下面的面积
#随机森林的ROC曲线比SDG分类器ROC曲线，更靠近左上，AUC值更大，效果更好些
扫一扫
2702
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
