[ML]keras和tensorflow波士顿房价预测模型

from keras.datasets import boston_housing
from keras.models import Sequential
from keras.layers import Dense,BatchNormalization,Dropout,Reshape,Flatten
from keras.layers.convolutional import Conv2D,MaxPooling2D,Conv1D,MaxPooling1D
import numpy as np
import os

(x_train, y_train), (x_test, y_test) = boston_housing.load_data()

def build0():
    model=Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

def build1():
    model=Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Dense(13))
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

def build2():
    model=Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Dense(13,activation='tanh'))
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

def build3():
    model=Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Dense(13,activation='sigmoid'))
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

def build4():
    model=Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Reshape((13,1)))
    model.add(Conv1D(filters=13,strides=1,padding='same',kernel_size=2,activation='sigmoid'))
    model.add(Conv1D(filters=26, strides=1, padding='same', kernel_size=2, activation='sigmoid'))
    model.add(MaxPooling1D(pool_size=2,strides=1,padding='same'))
    model.add(Conv1D(filters=52, strides=1, padding='same', kernel_size=2, activation='sigmoid'))
    model.add(Conv1D(filters=104, strides=1, padding='same', kernel_size=2, activation='sigmoid'))
    model.add(MaxPooling1D(pool_size=2, strides=1, padding='same'))
    model.add(Flatten())
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

def build5():
    model = Sequential(name='boston')
    model.add(BatchNormalization(input_shape=(13,)))
    model.add(Reshape((13, 1,1)))
    model.add(Conv2D(filters=13, strides=1, padding='same', kernel_size=1, activation='sigmoid'))
    model.add(Conv2D(filters=26, strides=2, padding='same', kernel_size=2, activation='sigmoid'))
    model.add(MaxPooling2D(pool_size=2, strides=1, padding='same'))
    model.add(Conv2D(filters=52, strides=1, padding='same', kernel_size=1, activation='sigmoid'))
    model.add(Conv2D(filters=104, strides=2, padding='same', kernel_size=2, activation='sigmoid'))
    model.add(MaxPooling2D(pool_size=2, strides=1, padding='same'))
    model.add(Flatten())
    model.add(Dropout(0.5))
    model.add(Dense(1))
    return model

for i in range(5,6):
    model=eval("build"+str(i)+"()")
    model.compile('adam','mae')#3.0895
    history=model.fit(x_train,y_train,batch_size=404,epochs=10000,verbose=1,validation_data=(x_test,y_test))
    print(history.history)

    f=open("result.txt",'a')
    f.write(str(history.history['val_loss'][-1])+"
")
    f.close()

其中build5方法预测最准确mae loss最小

tensorflow

import tensorflow as tf
import numpy as np
from keras.datasets import boston_housing

(x_train, y_train), (x_test, y_test) = boston_housing.load_data()


X=tf.placeholder('float32',[None,13])
y=tf.placeholder('float32',[None])

keep_prob=tf.placeholder('float32')

normal1=tf.layers.batch_normalization(X)

reshape1=tf.reshape(normal1,shape=(-1,13,1,1))
conv1_1=tf.layers.conv2d(reshape1,filters=13,strides=1,padding='same',kernel_size=1,activation=tf.nn.sigmoid)
conv1_2=tf.layers.conv2d(conv1_1,filters=26,strides=2,padding='same',kernel_size=2,activation=tf.nn.sigmoid)
pool1=tf.layers.max_pooling2d(conv1_2,pool_size=2,strides=1,padding='same')

conv2_1=tf.layers.conv2d(pool1,filters=52,strides=1,padding='same',kernel_size=1,activation=tf.nn.sigmoid)
conv2_2=tf.layers.conv2d(conv2_1,filters=104,strides=2,padding='same',kernel_size=2,activation=tf.nn.sigmoid)
pool2=tf.layers.max_pooling2d(conv2_2,pool_size=2,strides=1,padding='same')

reshape2=tf.reshape(pool2,shape=(-1,416))
dropout1=tf.nn.dropout(reshape2,keep_prob=keep_prob)

dense1=tf.layers.dense(dropout1,units=1)
reshape3=tf.reshape(dense1,shape=(-1,))

loss=tf.reduce_mean(tf.abs(y-reshape3))
step=tf.train.AdamOptimizer(0.001).minimize(loss)


with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    for i in range(30000):
        sess.run(step,feed_dict={X:x_train,y:y_train,keep_prob:0.5})
        # print(loss_val)

        loss_val=sess.run(loss,feed_dict={X:x_test,y:y_test,keep_prob:1})
        print(str(i)+"=>"+str(loss_val))