具有张量流的2层神经网络的损失不变[预测贷款决策]]

Question

我的数据如下所示：Loan Decision Data

我正在尝试根据前4个特征（Gen，Age，Psco，SocCon）预测贷款决定（0或1）Gen和SocCon是分类的，因此我使用了一键编码。

这里是代码：

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
import tensorflow as tf

loan_data = pd.read_csv('data/LoanDecision.csv')
loanNN = pd.get_dummies(loan_data, columns=['Gen','SocCon'])

X_dataNN = loanNN[['Gen_1', 'Gen_2', 'AGE', 'Psco', 'SocCon_1', 'SocCon_2', 'SocCon_3']]
y_dataNN = loanNN[['Decision']]

X_trainNN, X_testNN, y_trainNN, y_testNN = train_test_split(X_dataNN, y_dataNN, random_state=1011)

x = tf.placeholder(tf.float32,[None,7])
t = tf.placeholder(tf.float32,[None,1])

w1 = tf.Variable(tf.random_normal([7,7]), name = "weight1")
w0_1 = tf.Variable(tf.random_normal([7]),  name = "bias1")

f1 = tf.matmul(x,w1)+w0_1
p1 = tf.nn.softmax(f1)


W = tf.Variable(tf.random_normal([7,1]), name = "weights")
b = tf.Variable(tf.random_normal([1]), name = "bias")
f = tf.matmul(p1,W)+b
p = tf.sigmoid(f)

loss = tf.reduce_sum(tf.square(t-p))

#train_step = tf.train.GradientDescentOptimizer(0.5).minimize(loss)
train_step = tf.train.AdamOptimizer(0.01).minimize(loss)

#predict = tf.equal(tf.cast(p > 0.5,tf.float32), t)
predict = tf.equal(tf.round(p), tf.round(t))

accuracy = tf.reduce_mean(tf.cast(predict,tf.float32))

sess = tf.Session()
sess.run(tf.global_variables_initializer())

i=0
for _ in range(10000):
    i+=1
    sess.run(train_step, feed_dict = {x: X_dataNN, t: y_dataNN})

    if i % 1000 == 0:
        loss_val, acc_val = sess.run([loss, accuracy], feed_dict = {x: X_dataNN, t: y_dataNN})
        print('Step: %d, Loss: %f, Accuracy: %f '%(i, loss_val, acc_val))

sess.close()
我不确定出什么问题，但是我一直得到相同的损失值，准确度没有改变。

Step: 1000, Loss: 473.172150, Accuracy: 0.625062 
Step: 2000, Loss: 473.172150, Accuracy: 0.625062 
Step: 3000, Loss: 473.172150, Accuracy: 0.625062 
Step: 4000, Loss: 473.172150, Accuracy: 0.625062 
Step: 5000, Loss: 473.172150, Accuracy: 0.625062 
Step: 6000, Loss: 473.172150, Accuracy: 0.625062 
Step: 7000, Loss: 473.171844, Accuracy: 0.625062 
Step: 8000, Loss: 473.171906, Accuracy: 0.625062 
Step: 9000, Loss: 473.172821, Accuracy: 0.625062 
Step: 10000, Loss: 473.170105, Accuracy: 0.625062 
使用单层时是相同的。有任何想法吗？是损失函数，训练还是仅仅是不良数据？任何帮助表示赞赏。

我的数据看起来像这样：贷款决策数据我试图基于前4个特征（Gen，Age，Psco，SocCon）来预测贷款决策（0或1）Gen和SocCon是分类的，因此我使用了-热...

Answer 1

通常来说，很少怪数据（除非您有理由怀疑它？），而且我不确定您对培训的含义，它看起来很合理。您的损失正在缓慢地移动，所以我想您需要学习学习率/损失功能