TensorFlow实现波士顿房价多层感知机(MLP)

一、代码

# coding=utf-8
import pandas as pd  # 用于分析数据集
import seaborn as sns  # 可视化
import tensorflow as tf
import matplotlib.pyplot as plt  # 可视化
import tensorflow.contrib.layers as layers
from sklearn import datasets  # 用于获取数据集
from sklearn.preprocessing import MinMaxScaler  # 用于数据归一化
from sklearn.model_selection import train_test_split  # 用于划分训练集、测试集

# 数据集
boston = datasets.load_boston()  # 读取波士顿房价，返回Bunch对象
df = pd.DataFrame(boston.data, columns=boston.feature_names)  # 创建Pandas的数据结构DataFrame
df['target'] = boston.target

print(df.describe())  # 数据细节

# 画图看特征间的线性相关性
_, ax = plt.subplots(figsize=(12, 10))  # 分辨率1200×1000

corr = df.corr(method='pearson')  # 使用皮尔逊系数计算列与列的相关性
cmap = sns.diverging_palette(220, 10, as_cmap=True)  # 在两种HUSL颜色之间制作不同的调色板。图的正负色彩范围为220、10，结果为真则返回matplotlib的colormap对象
_ = sns.heatmap(
    corr,  # 使用Pandas DataFrame数据，索引/列信息用于标记列和行
    cmap=cmap,  # 数据值到颜色空间的映射
    square=True,  # 每个单元格都是正方形
    cbar_kws={'shrink': .9},  # `fig.colorbar`的关键字参数
    ax=ax,  # 绘制图的轴
    annot=True,  # 在单元格中标注数据值
    annot_kws={'fontsize': 12})  # 热图，将矩形数据绘制为颜色编码矩阵

plt.show()

X_train, X_test, y_train, y_test = train_test_split(df[['RM', 'LSTAT', 'PTRATIO']], df[['target']], test_size=0.3,
                                                    random_state=0)  # 创建训练集和测试集，测试集占0.3，随机种子0

X_train = MinMaxScaler().fit_transform(X_train)  # 归一化，缩放到0-1
y_train = MinMaxScaler().fit_transform(y_train)
X_test = MinMaxScaler().fit_transform(X_test)
Y_test = MinMaxScaler().fit_transform(y_test)

m = len(X_train)  # 训练集数
n = 3  # 特征数
n_hidden = 20  # 隐藏层数

batch_size = 200  # 每批训练批量大小
eta = 0.01  # 学习率
max_epoch = 1000  # 最大迭代数


# 定义模型
def multilayer_perceptron(x):
    fc1 = layers.fully_connected(x, n_hidden, activation_fn=tf.nn.relu, scope='fc1')  # 单隐藏层，激活函数为ReLU
    out = layers.fully_connected(fc1, 1, activation_fn=tf.sigmoid, scope='out')  # 输出层，激活函授为Sigmoid
    return out


def accuracy(a, b):
    correct_prediction = tf.square(a - b)
    return tf.reduce_mean(tf.cast(correct_prediction, "float"))


x = tf.placeholder(tf.float32, name='X', shape=[m, n])  # 占位符
y = tf.placeholder(tf.float32, name='Y')
y_hat = multilayer_perceptron(x)

mse = accuracy(y, y_hat)  # 均方差
train = tf.train.AdamOptimizer(learning_rate=eta).minimize(mse)  # 优化器使用Adam优化算法

# 训练
init = tf.global_variables_initializer()
with tf.Session() as sess:
    sess.run(init)
    writer = tf.summary.FileWriter('graphs', sess.graph)  # 将摘要与图形写入graphs目录

    for i in range(max_epoch):
        _, l, p = sess.run([train, mse, y_hat], feed_dict={x: X_train, y: y_train})
        if i % 100 == 0:
            print('Epoch {0}: Loss {1}'.format(i, l))

    print("Training Done")
    print("Optimization Finished!")

    # 评估
    print("Mean Squared Error (Train data):", mse.eval({x: X_train, y: y_train}))

    plt.scatter(p, y_train)
    plt.ylabel('Estimated Price')
    plt.xlabel('Actual Price')
    plt.title('Estimated vs Actual Price Train Data')
    plt.show()

    writer.close()

二、结果

Mean Squared Error (Train data): 0.006424182