1.前言
sklearn神经网络,进行多分类,数字识别。
2.python代码
(1)数据集用的sklearn自带,数字0~9分类
(2)采用MLPClassifier
(3)执行代码如下multi_class_nn.py:
from sklearn.neural_network import MLPClassifier
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
# 测试集,画图对预测值和实际值进行比较
def test_validate(x_test, y_test, y_predict, classifier):
x = range(len(y_test))
plt.plot(x, y_test, "ro", markersize=5, zorder=3, label=u"true_v")
plt.plot(x, y_predict, "go", markersize=8, zorder=2, label=u"predict_v,$R$=%.3f" % classifier.score(x_test, y_test))
plt.legend(loc="upper left")
plt.xlabel("number")
plt.ylabel("true?")
plt.show()
# 神经网络数字分类
def multi_class_nn():
digits = datasets.load_digits()
x = digits['data']
y = digits['target']
# 对数据的训练集进行标准化
ss = StandardScaler()
x_regular = ss.fit_transform(x)
# 划分训练集与测试集
x_train, x_test, y_train, y_test = train_test_split(x_regular, y, test_size=0.1)
clf = MLPClassifier(solver='lbfgs', alpha=1e-5, hidden_layer_sizes=(5,), random_state=1)
clf.fit(x_train, y_train)
# 模型效果获取
r = clf.score(x_train, y_train)
print("R值(准确率):", r)
# 预测
y_predict = clf.predict(x_test) # 预测
print(y_predict)
print(y_test)
# 绘制测试集结果验证
test_validate(x_test=x_test, y_test=y_test, y_predict=y_predict, classifier=clf)
multi_class_nn()
3.验证结果
下面给出不同隐藏单元数量结果,运行结果
(1)红点是测试集真实结果,绿点是预测结果,红框部分出现了红绿点不重合部分数据(表示预测不准确),看到正确率r=0.872,预测正确/总数。
MLPClassifier参数:hidden_layer_sizes=(5,),结果如图
(2)MLPClassifier参数:hidden_layer_sizes=(15,),结果如图。准确率大幅提高