dnn预测房价

1.导入库

import tensorflow as tf
import pandas as pd
import numpy as np
from sklearn.datasets import load_boston
import matplotlib.pyplot as plt
from sklearn.preprocessing import scale
from sklearn.model_selection import train_test_split

2.获取数据并进行标准化

boston = load_boston()
X_train, X_test, y_train, y_test = train_test_split(boston.data, boston.target, test_size=0.1, random_state=0)
X_train = scale(X_train)
X_test = scale(X_test)
y_train = scale(y_train.reshape((-1,1)))
y_test = scale(y_test.reshape((-1,1)))
print(X_train.shape, y_train.shape)

3.构建nn层

def build_nn(inputs, input_size, output_size, acticity_function=None, keep_prob = 1):
    with tf.variable_scope('softmax_w'):
        softmax_w = tf.Variable(tf.random_normal([input_size, output_size]))
    with tf.variable_scope('softmax_b'):
        softmax_b = tf.Variable(tf.zeros([1, output_size]) + 0.1)
    with tf.name_scope('result'):
        result = tf.matmul(inputs, softmax_w) + softmax_b
    with tf.name_scope('dropout'):
        dropout = tf.nn.dropout(result, keep_prob=keep_prob)
    if acticity_function is None:
        return dropout
    else:
        return acticity_function(dropout)

4.构建dnn

inputs = tf.placeholder(tf.float32, [None, X_train.shape[1]], 'inputs')
targets = tf.placeholder(tf.float32, [None, 1], 'targets')
l1 = build_nn(inputs, X_train.shape[1], 10, tf.nn.relu)
l2 = build_nn(l1, int(l1.shape[1]), 13)
l3 = build_nn(l2, int(l2.shape[1]), 1)
pred = tf.add(l3, 0, name='pred')
loss = tf.reduce_mean(tf.reduce_sum(tf.square(l3 - targets),reduction_indices=[1]))
tf.summary.scalar("loss", tensor=loss)
train_op = tf.train.AdamOptimizer().minimize(loss)

刚开始在l1,l2,l3那儿报错，因为刚开始的代码是

l1 = build_nn(inputs, X_train.shape[1], 10, tf.nn.relu)
l2 = build_nn(l1, l1.shape[1], 13)
l3 = build_nn(l2, l2.shape[1], 1)

报错为TypeError: Failed to convert object of type <class ‘list’> to Tensor. Contents: [Dimension(10), 13]. Consider casting elements to a supported type.
原因是l1.shape[1]是Dimension(10)，不是10，要用int强制转换才可以
也就是这样

l1 = build_nn(inputs, X_train.shape[1], 10, tf.nn.relu)
l2 = build_nn(l1, int(l1.shape[1]), 13)
l3 = build_nn(l2, int(l2.shape[1]), 1)

5.画图

# draw pics
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.plot(range(50), y_train[0:50], 'b')  #展示前50个数据
ax.set_ylim([-2, 5])
plt.ion()
plt.show()

# parameters
keep_prob = 0.8  # 防止过拟合，取值一般在0.5到0.8
ITER = 5000  # 训练次数

6.训练

def fit(X, y, ax, n, keep_prob):
    feed_list = {inputs: X, targets: y}
    with tf.Session() as sess:
        saver = tf.train.Saver(tf.global_variables(), max_to_keep=15)
        merged = tf.summary.merge_all()
        writer = tf.summary.FileWriter(logdir="nn_boston_log", graph=sess.graph)  # 写tensorbord
        sess.run(tf.global_variables_initializer())
        for i in range(n):
            _loss, _ = sess.run([loss, train_op], feed_dict=feed_list)
            if i % 100 == 0:
                print("epoch:%d\tloss:%.5f" % (i, _loss))
                y_pred = sess.run(pred, feed_dict=feed_list)
                rs = sess.run(merged, feed_dict=feed_list)
                writer.add_summary(summary=rs, global_step=i)  # 写tensorbord
                saver.save(sess=sess, save_path="nn_boston_model/nn_boston.model", global_step=i)  # 保存模型
                try:
                    ax.lines.remove(lines[0])
                except:
                    pass
                lines = ax.plot(range(50), y_pred[0:50], 'r--')
                plt.pause(1)

        saver.save(sess=sess, save_path="nn_boston_model/nn_boston.model", global_step=n)  # 保存模型

fit(X_train, y_train, ax, ITER, keep_prob )

运行结果：
epoch:0 loss:342.51898
epoch:100 loss:49.48788
epoch:200 loss:19.40265
epoch:300 loss:9.06275
epoch:400 loss:5.07183
…
epoch:4700 loss:0.13370
epoch:4800 loss:0.13194
epoch:4900 loss:0.13034