[人工智能]深度学习------tensorflow多分类

1.自定义数据集（三分类）

import tensorflow as tf
import pandas as pd
import numpy as np
from sklearn.datasets import load_iris
from sklearn.preprocessing import StandardScaler,OneHotEncoder
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt

x_data = [[1, 2, 1, 1],
          [2, 1, 3, 2],
          [3, 1, 3, 4],
          [4, 1, 5, 5],
          [1, 7, 5, 5],
          [1, 2, 5, 6],
          [1, 6, 6, 6],
          [1, 7, 7, 7]]
# 3类别
y_data = [[0, 0, 1],
          [0, 0, 1],
          [0, 0, 1],
          [0, 1, 0],
          [0, 1, 0],
          [0, 1, 0],
          [1, 0, 0],
          [1, 0, 0]]

#3.#定义占位符\权重和斜率
x=tf.placeholder(tf.float32,shape=[None,4])
y=tf.placeholder(tf.float32,shape=[None,3])

w=tf.Variable(tf.random_normal([4,3]))
b=tf.Variable(tf.random_normal([3]))

#4.定义前向传播函数
# model=tf.matmul(x,w)+b
hypothesis=tf.nn.softmax(tf.matmul(x,w)+b)

#5.代价函数
#方法一
cost=-tf.reduce_mean(y*tf.log(hypothesis),axis=1)
# 方法二
# cost=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_data, logits=model))
# 6.梯度向下函数
gradeDecline=tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)

#7.计算精度
predicted=tf.argmax(hypothesis,axis=1)
true_value=tf.argmax(y_data,axis=1)
acc=tf.reduce_mean(tf.cast(tf.equal(predicted,true_value),dtype=tf.float32))

#8.初始化全局变量
sess=tf.Session()
sess.run(tf.global_variables_initializer())

#9.画代价函数图
j=[]
for i in range(3000):
    cost_val,_,acc_val=sess.run([cost,gradeDecline,acc],feed_dict={x:x_data,y:y_data})
    j.append(cost_val)
    if i%100==0:
        print(i,cost_val)
plt.plot(j)
plt.show()

2. iris数据集(三分类)

import tensorflow as tf
import pandas as pd
import numpy as np
import os
from sklearn.datasets import load_iris
from sklearn.preprocessing import StandardScaler,OneHotEncoder
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
plt.rcParams['font.family']='SimHei'
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
tf.set_random_seed(777)

#1.加载数据集
data=load_iris()
x_data=data.data
y_data=data.target
y_data=np.c_[y_data]
#2.特征缩放、独热
sobj=StandardScaler()
x_data=sobj.fit_transform(x_data)
train_x,test_x,train_y,test_y=train_test_split(x_data,y_data,train_size=0.7)

oobj=OneHotEncoder()
y_data=oobj.fit_transform(y_data).toarray()
train_y=oobj.fit_transform(train_y).toarray()
test_y=oobj.fit_transform(test_y).toarray()

#3.#定义占位符\权重和斜率
x=tf.placeholder(tf.float32,shape=[None,4])
y=tf.placeholder(tf.float32,shape=[None,3])

w=tf.Variable(tf.random_normal([4,3]))
b=tf.Variable(tf.random_normal([3]))

#4.定义前向传播函数
model=tf.matmul(x,w)+b
h=tf.nn.softmax(model)

#5.代价函数
#方法一
cost=-tf.reduce_mean(y*tf.log(h),axis=1)
# 方法二
# cost=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_data, logits=model))
# 6.梯度向下函数
gradeDecline=tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)

#7.计算精度
predicted=tf.argmax(model,axis=1)
def acc(y):
    true_value=tf.argmax(y,axis=1)
    acc=tf.reduce_mean(tf.cast(tf.equal(predicted,true_value),dtype=tf.float32))
    return acc
train_acc=acc(train_y)
test_acc=acc(test_y)

j=[]
#8.初始化全局变量
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
#9.画代价函数图
    for i in range(2001):
        cost_val, w_val, b_val, _ ,acc_val= sess.run([cost, w, b, gradeDecline,train_acc], feed_dict={x: train_x, y: train_y})
        if i % 100 == 0:
            print(i,cost_val)
            j.append(cost_val)

    # 4.打印代价图
    plt.title('代价图')
    plt.plot(j)
    plt.show()
    h_val,predict_val,acc_val=sess.run([h,predicted,train_acc], feed_dict={x: train_x, y: train_y})
    print(acc_val)
    h_v, pre_val, acc_v = sess.run([h, predicted, test_acc], feed_dict={x: test_x, y: test_y})
    print(acc_v)

注意：没有用softmax优化器效果，代价函数好多条.

import tensorflow as tf
import pandas as pd
import numpy as np
import os
from sklearn.datasets import load_iris
from sklearn.preprocessing import StandardScaler,OneHotEncoder
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
plt.rcParams['font.family']='SimHei'
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
tf.set_random_seed(777)

#1.加载数据集
data=load_iris()
x_data=data.data
y_data=data.target
y_data=np.c_[y_data]
#2.特征缩放、独热
sobj=StandardScaler()
x_data=sobj.fit_transform(x_data)
train_x,test_x,train_y,test_y=train_test_split(x_data,y_data,train_size=0.7)

oobj=OneHotEncoder()
train_y_o=oobj.fit_transform(train_y).toarray()
test_y_o=oobj.fit_transform(test_y).toarray()

#3.#定义占位符\权重和斜率
x=tf.placeholder(tf.float32,shape=[None,4])
y=tf.placeholder(tf.float32,shape=[None,1])

w=tf.Variable(tf.random_normal([4,3]))
b=tf.Variable(tf.random_normal([3]))

#4.定义前向传播函数
model=tf.matmul(x,w)+b
h=tf.nn.softmax(model)

#5.代价函数
#方法一
# cost=-tf.reduce_mean(y*tf.log(h),axis=1)
# 方法二
cost=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=train_y_o, logits=model))
# 6.梯度向下函数
gradeDecline=tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(cost)

#7.计算精度
predicted=tf.argmax(model,axis=1)
def acc(y):
    true_value=tf.argmax(y,axis=1)
    acc=tf.reduce_mean(tf.cast(tf.equal(predicted,true_value),dtype=tf.float32))
    return acc
train_acc=acc(train_y_o)
test_acc=acc(test_y_o)

j=[]
#8.初始化全局变量
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
#9.画代价函数图
    for i in range(2001):
        cost_val, w_val, b_val, _ ,acc_val= sess.run([cost, w, b, gradeDecline,train_acc], feed_dict={x: train_x, y: train_y})
        if i % 100 == 0:
            print(i,cost_val)
            j.append(cost_val)

    # 4.打印代价图
    plt.title('代价图')
    plt.plot(j)
    plt.show()
    h_val,predict_val,acc_val=sess.run([h,predicted,train_acc], feed_dict={x: train_x, y: train_y})
    print(acc_val)
    h_v, pre_val, acc_v = sess.run([h, predicted, test_acc], feed_dict={x: test_x, y: test_y})
    print(acc_v)

注意：用softmax优化器效果最好,代价函数一条。

3. zoo数据集（7分类）

from sklearn.preprocessing import OneHotEncoder
import numpy as np
import os
import matplotlib.pyplot as plt
import tensorflow as tf
from sklearn.model_selection import train_test_split
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
plt.rcParams['font.family']='SimHei'
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)

data=np.loadtxt(r'E:\ana\envs\tf14\day07\data-04-zoo.csv',delimiter=',')
print(data)
x_data=data[:,:-1]
y_data=data[:,-1:]
print(x_data.shape)
print(y_data.shape)
tf.set_random_seed(777)
train_x,test_x,train_y,test_y=train_test_split(x_data,y_data,train_size=0.7)
oobj=OneHotEncoder()
y_data=oobj.fit_transform(y_data).toarray()
train_y_o=oobj.fit_transform(train_y).toarray()
test_y_o=oobj.fit_transform(test_y).toarray()

x=tf.placeholder(dtype=tf.float32,shape=[None,16])
y=tf.placeholder(dtype=tf.float32,shape=[None,1])

w=tf.Variable(tf.random_normal([16,7]))
b=tf.Variable(tf.random_normal([7]))

model=tf.matmul(x,w)+b
h=tf.nn.softmax(model)
# cost=-tf.reduce_mean(y*tf.log(h),axis=1)
cost=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=train_y_o,logits=model))
gradeDecline=tf.train.GradientDescentOptimizer(learning_rate=0.1).minimize(cost)

#计算准确率
y_predict=tf.argmax(model,axis=1)
def acc(y):
    y_true = tf.argmax(y, axis=1)
    acc =tf.reduce_mean(tf.cast(tf.equal(y_true,y_predict),dtype=tf.float32))
    return acc
train_acc=acc(train_y_o)
test_acc=acc(test_y_o)

sess=tf.Session()
sess.run(tf.global_variables_initializer())
j=[]
for i in range(10000):
    cost_val,_,acc_val=sess.run([cost,gradeDecline,train_acc],feed_dict={x:train_x,y:train_y})
    j.append(cost_val)
    if i%500==0:
        print(i,cost_val)
plt.plot(j)
plt.show()
h1,p1,a1=sess.run([h,y_predict,train_acc],feed_dict={x:train_x,y:train_y})
print(a1)

h,p,a=sess.run([h,y_predict,test_acc],feed_dict={x:test_x,y:test_y})
print(a)