TensorFlow多层感知机boston房价预测

今天写了一个代码用tensorflow的多层感知机来实现boston房价的预测，感知器模型的搭建采用tensorflow.contrib.layers提供的函数

数据的读取和预处理用到了sklearn、panda等

#2021.10.16 HIT ATCI LZH
#TensorFlow多层感知机实现Boston房价预测,优化器采用AdadeltaOptimizer测试集的预测结果散点图看上去效果并不是很好，原因未知，而AdamOptimizer优化器的效果要好很多。
''' pandas 是基于numpy构建的含有更高级数据结构和工具的数据分析包，提供了高效地操作大型数据集所需的工具。
pandas有两个核心数据结构 Series和DataFrame，分别对应了一维的序列和二维的表结构。而describe()函数就是返回这两个核心数据结构的统计变量。
其目的在于观察这一系列数据的范围、大小、波动趋势等等，为后面的模型选择打下基础。
'''
from scipy.sparse.construct import random
import tensorflow as tf
import tensorflow.contrib.layers as layers #
import matplotlib.pyplot as plt #数据可视化
from sklearn import datasets #用于加载数据
from sklearn.model_selection import train_test_split #sklearn模块用于获取数据集和预处理数据集，并将其分为训练集和预测集
from sklearn.preprocessing import MinMaxScaler #用于归一化
import pandas as pd #可以用来分析数据集
import seaborn as sns #用于数据可视化
#加载数据集并创建Pandas数据帧来分析数据
boston = datasets.load_boston()
df = pd.DataFrame(boston.data, columns=boston.feature_names)#这部分是采用pd库进行数据处理
df['target'] = boston.target
print(df.describe())

#plotting correlation
# colormap_, ax = plt.subplots(figsize = (12, 10))
# corr = df.corr(method= 'pearson')
# cmap = sns.diverging_palette(220, 10, as_cmap=True)
# _ = sns.heatmap(corr, cmap=cmap, square=True, cbear_kws = {
# 'shrink': .9}, ax = ax, annot=True, annot_kws={'fontsize':12
# })
#Create Test Train Split
X_train, X_test, y_train, y_test = train_test_split(df[['RM',
'LSTAT', 'PTRATIO']],df[['target']], test_size=0.3,
random_state = 0)  #测试集占总样本的30%
print('X_train = {0}, X_test = {1}, y_train = {2}, y_test ={3}'.format(len(X_train), len(X_test), len(y_train), len(y_test)))
#数据归一化
X_train = MinMaxScaler().fit_transform(X_train)
y_train = MinMaxScaler().fit_transform(y_train)
X_test = MinMaxScaler().fit_transform(X_test)
y_test = MinMaxScaler().fit_transform(y_test)
#Network Parameters
n_hidden = 20 #隐层节点数为30个
n = 3 #特征数
m = len(X_train)
#Hyperparameters
eta = 0.01 #学习率 
max_epoch = 1000 #最大训练次数
#定义一个单隐层的多层感知器模型
def multilayer_perceptron(x):
    fc1 = layers.fully_connected(x, n_hidden, activation_fn = tf.nn.relu, biases_initializer = tf.zeros_initializer(), scope = 'fc1')#隐层的传递函数为relu函数
    #fc2 = layers.fully_connected(fc1, 256, activation_fn = tf.nn.relu, biases_initializer = tf.zeros_initializer(), scope = 'fc2')
    out = layers.fully_connected(fc1, 1, activation_fn = tf.sigmoid, biases_initializer = tf.zeros_initializer(), scope = 'out')#输出层的传递函数为sigmoid
    return out
#申明训练数据的占位符和损失函数及优化器
#定义占位符
x = tf.placeholder(tf.float32, [None, n], name = 'placeholder_x')
y = tf.placeholder(tf.float32, name = 'placeholer_y')
y_hat = multilayer_perceptron(x)
#mse作为损失函数
correct_prediction = tf.square(y - y_hat)
mse = tf.reduce_mean(tf.cast(correct_prediction, 'float'))
#优化器
train = tf.train.AdamOptimizer(learning_rate=eta).minimize(mse)
#变量初始化
init = tf.global_variables_initializer()
Loss_epoch = []
with tf.Session() as  sess:
    sess.run(init)
    for epoch in range(max_epoch):
        _, c = sess.run([train, mse], feed_dict = {x: X_train, y: y_train})
        Loss_epoch.append(c)
        print("Epoch %02d, Loss = %.6f" %(epoch, c))
    print('Training Done')
    pre_mse, pre_y = sess.run([mse, y_hat], feed_dict = {x:X_test, y:y_test})
print('Optimization Finished!')
print('pre_ytesr_mse = ', pre_mse)
plt.figure(1)
plt.title('loss')
plt.plot(Loss_epoch)
plt.show()
plt.figure(2)
#plt.title('loss')
plt.scatter(y_test,pre_y)
plt.xlabel('y_test')
plt.ylabel('pre_y_test')
plt.show()

 

当散点图的分布趋近于45°线的时候说明预测效果非常好。