神经网络Python练习

用TensorFlow创建FNN神经网络模型，梯度下降采用Adagrad，使用dropout防止过拟合，尝试保存模型再调用的操作。

import tensorflow as tf
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn import metrics

#定义读取数据的函数
def load_data(train_path=""):
    train=pd.read_csv(train_path,names=["feature1",	"feature2"	,"feature3"	,"feature4"	,"feature5",
                                        "feature6"	,"feature7"	,"feature8",	"feature9"	,"feature10",
                                        "feature11"	,"feature12",	"feature13"	,"feature14",	"feature15"	,
                                        "feature16"	,"feature17"	,"feature18",	"feature19",	"feature20",
                                        "feature21"	,"feature22",	"feature23"	,"feature24"	,"feature25",
                                        "feature26",	"feature27",	"feature28"	,"feature29",	"feature30",
                                        "feature31"	,"feature32",	"feature33",	"feature34"	,"feature35",
                                        "feature36",	"feature37",	"feature38",	"feature39"	,"feature40",
                                        "feature41"	,"feature42",	"feature43"	,"feature44",	"feature45",
                                        "feature46",	"feature47",	"feature48"	,"feature49",	"feature50"	,
                                        "feature51",	"feature52"	,"feature53",	"feature54"	,"feature55",
                                        "feature56"	,"feature57",	"feature58",	"feature59",	"feature60"	,
                                        "feature61",	"feature62",	"feature63",	"feature64",	"feature65"	,
                                        "feature66",	"feature67",	"feature68",	"feature69"	,"feature70"	,
                                        "feature71"	,"feature72"	,"feature73","y"],header=0)
    dict_label={"y":train.pop("y")}
    train_x,train_y=train,(pd.DataFrame(dict_label))
    return (train_x,train_y)

X_train,y_train=load_data("X8.csv")
X_test,y_test=load_data("X8_test.csv")
print(X_train,y_train)

#占位符
xs = tf.placeholder(shape=[None,X_train.shape[1]],dtype=tf.float32,name="inputs")
ys = tf.placeholder(shape=[None,1],dtype=tf.float32,name="y_true")

#将将添加神经网络层的操作封装起来
def add_layer(inputs,input_size,output_size,activation_function=None):
    with tf.variable_scope("Weights"):
        Weights = tf.Variable(tf.random_normal(shape=[input_size,output_size]),name="weights")
    with tf.variable_scope("biases"):
        biases = tf.Variable(tf.zeros(shape=[1,output_size]) + 0.1,name="biases")
    with tf.name_scope("Wx_plus_b"):
        Wx_plus_b = tf.matmul(inputs,Weights) + biases
    #dropout是一个防止过拟合的trick：dropout是指在深度学习网络的训练过程中,对于神经网络单元,按照一定的概率将其暂时从网络中丢弃。
    with tf.name_scope("dropout"):
            Wx_plus_b = tf.nn.dropout(Wx_plus_b)
    #激活函数
    if activation_function is None:
        return Wx_plus_b
    else:
        with tf.name_scope("activation_function"):
            return activation_function(Wx_plus_b)

#添加神经网络层
with tf.name_scope("layer_1"):
    l1 = add_layer(xs,73,30,activation_function=tf.nn.relu)
with tf.name_scope("layer_2"):
    l2 = add_layer(l1,30,10,activation_function=tf.nn.relu)
with tf.name_scope("y_pred"):
    pred = add_layer(l2,10,1)

#loss函数是均方误差
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys - pred),reduction_indices=[1]))  # mse
#梯度下降采用Adagrad
train_op = tf.train.AdagradOptimizer(learning_rate=0.05).minimize(loss)



init = tf.global_variables_initializer()



with tf.Session() as sess:
    saver = tf.train.Saver(tf.global_variables(), max_to_keep=1)
    sess.run(init)
    for i in range(50000):
        _loss, _ = sess.run([loss, train_op], feed_dict={xs: X_train[0:99], ys: y_train[0:99]})
        if i % 100 == 0:
            print("epoch:%d\tloss:%.5f" % (i, _loss))
   # saver.save(sess=sess, save_path="NN/DNN.model", global_step=99999)  # 保存模型
   # builder=tf.saved_model.builder.SavedModelBuilder("NNmodel")
   # builder.add_meta_graph_and_variables(sess,[tf.saved_model.tag_constants.TRAINING])#保存整张网络及其变量，这种方法是可以保存多张网络的，在此不作介绍，可自行了解
    #builder.save()#完成保存
    y_pred = sess.run(pred, feed_dict={xs: X_train[0:99], ys: y_train[0:99]})
    pred_y = sess.run(pred, feed_dict={xs: X_test})
    #mse = tf.reduce_mean(tf.square(pred_y - y_test))
    print(pred_y)
    print("RMSE:", np.sqrt(metrics.mean_squared_error(y_test, pred_y)))
    #print("MSE: %.4f" % sess.run(mse))
    #print(y_pred)
sess.close()

# draw pics
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.plot(range(99),y_train[0:99],'b')
ax.plot(range(99), y_pred[0:99], 'r--')
   # plt.pause(1)
ax.set_ylim([-2,5])
plt.xlabel("Train size")

plt.ion()
plt.show()

# show data
plt.plot(range(len(y_test)),y_test,'b')
plt.plot(range(len(pred_y)),pred_y,'r--')
plt.xlabel("Test size")

plt.show()

过拟合：

添加处理后：


效果不好，还得改进。