Tensorflow:softmax处理Iris鸾尾花分类

经典的分类模型，鸾尾花的分类。
数据集样例：

length	sepal width	petal length	petal width	class
5.1	3.5	1.4	0.2	Iris-setosa
4.9	3	1.4	0.2	Iris-versicolor
4.7	3.2	1.3	0.2	Iris-virginica

主要是使用softmax来做多类别分类，
残差函数使用的是熵: -y*log(predicY)
代码：

__author__ = 'jmh081701'
#coding:utf-8
import tensorflow as tf
import numpy as np
import pandas as pd
import os

W =tf.Variable(np.random.randn(4,3),dtype=tf.float32,name='W')
b =tf.Variable(0,dtype=tf.float32,name='bias')

def softmax(x):
    x=tf.to_float(x)
    y=[]
    y=tf.exp(x)
    z=tf.reduce_sum(y,1)
    z=tf.transpose([z,z,z])
    z=y/z
    #tensorflow的矩阵除法：同一位置的元素相除
    return z

def load_csv(filename):
    df=pd.read_csv(filename)
    return df
def inputs(df,tables):
    X=[]
    Y=[]
    length =len(df)
    for i in range(0,length):
        x=[df.ix[i]['sepal length'],df.ix[i]['sepal width'],df.ix[i]['petal length'],df.ix[i]['petal length']]
        y=df.ix[i]['class']
        if(y=='Iris-setosa'):
            y=[1,0,0]
        elif (y=='Iris-versicolor'):
            y=[0,1,0]
        else:
            y=[0,0,1]
        X.append(x)
        Y.append(y)
    X=tf.to_float(X)
    Y=tf.to_float(Y)
    return X,Y
def combine(X):

    rst = tf.add(tf.matmul(X,W),b,name='combine_add_as_x')
    #print(sess.run([X,rst]))
    return rst
def loss(X,Y):
    predictY=inference(X)
    logy=tf.log(predictY)
    l=-Y*logy
    rst=tf.reduce_sum(l)
    #print("rst",sess.run(rst))
    return rst

def train(total_loss):
    rate=0.0001
    return tf.train.GradientDescentOptimizer(learning_rate=rate).minimize(total_loss)

def evaluate(X,Y,sess=None):
#评估函数
    predictY=inference(X)

    predictY=tf.to_int32(tf.arg_max(predictY,1)) #按列取最大下标
    Y=tf.to_int32(Y)
    Y=tf.to_int32(tf.arg_max(Y,1))
    #按列取最大下标
    rate =sess.run(tf.reduce_mean(tf.to_float(tf.equal(Y,predictY))))
    #reduce_mean是将矩阵的每一个元素相加求和

    print("evaluate:",rate)
    return rate
def load_model(sess,saver,filedir,filename):
    ckpt=tf.train.get_checkpoint_state(filedir)

    if( ckpt and ckpt.model_checkpoint_path):
        saver.restore(sess,filedir+".\\"+filename)  #载入模型

def inference(x):
    return softmax(combine(x))
if __name__ == '__main__':
    saver=tf.train.Saver()
    dir=r"D:\\data\\Iris\\"
    modelfile="my-model.cpt"
    with tf.Session() as sess:
        init =tf.initialize_all_variables()
        sess.run(init)
        coord= tf.train.Coordinator()
        threads =tf.train.start_queue_runners(sess,coord)
        df =load_csv(dir+"train.csv")
        X,Y=inputs(df, None)
        total_loss =loss(X,Y)
        train_op=train(total_loss)
        step =0
        ratesum=0
        i=0
        sameMax=20
        load_model(sess,saver,dir,modelfile)

        while True:
            step=step+1
            l,op=sess.run([total_loss,train_op])
            if(step%20==0):
                rate=evaluate(X,Y,sess)
                ratesum=ratesum+rate
                i=i+1
                if(i%sameMax==0):
                    #如果sameMax次的测试率都是一样的，近似于收敛,那么停止训练
                    ratesum=ratesum/sameMax
                    if(ratesum==rate):
                        break
                        saver.save(sess,dir+modelfile)
                    else:
                        ratesum=0
                        i=0
            if(step>10000):
                saver.save(sess,dir+modelfile)
                break
        coord.request_stop()
        coord.join(threads)
        print("training over!!!")
        evaluate(X,Y,sess)
        saver.save(sess,dir+modelfile)

最后准确率是94%左右