使用Tensorflow Python预测数组中的下一个数字

时间：2019-01-17 09:26:58

标签： java python tensorflow prediction

这是交易，我得到一个python脚本，作为输入，我有2个数组，一个数组具有一个函数的值（x_train），另一个数组具有测量值的时间（y_train）。 该脚本是从Java调用的，其参数如下所示：

a）

[38.0, 45.0, 42.0, 35.0, 41.0, 45.0, 69.0, 35.0, 33.0, 34.0]
[85.0, 86.0, 87.0, 88.0, 89.0, 90.0, 91.0, 92.0, 93.0, 94.0]

b）

[38.0, 45.0, 42.0, 35.0, 41.0, 45.0, 69.0, 35.0, 33.0, 34.0]
[1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0]

c）

[38.0, 45.0, 42.0, 35.0, 41.0, 45.0, 69.0, 35.0, 33.0, 34.0]
[1.562571106116E12, 1.562657506116E12, 1.562743906116E12, 1.562830306116E12, 1.562916706116E12, 1.563003106116E12, 1.563089506116E12, 1.563175906116E12, 1.563262306116E12, 1.563348706116E12]

我需要做一个9级函数并预测数组的下一个元素，即函数的系数（a9，a8，a7，a6，a5，a4，a3，a2，a1， a0和损失），然后用它们计算第11个数字。

问题在于，当我使用选项（a）和（c）时，由于该过程，我得到所有系数的NAN，但是当我使用选项（b）时，我得到了一些结果（这是问题函数的启动参数，我将在稍后介绍），并使用这些参数进行工作。

该脚本不会引发任何错误，我是这个主题的新手，可以使用一些帮助。

这是整个python脚本：

import tensorflow  
import sys

s1 = sys.stdin.readline().strip()
s2 = sys.stdin.readline().strip()
s3 = sys.stdin.readline().strip()

#Trainable Parameters  
startParam0 = 2 
startParam1 = 0.1
startParam2 = 0.1
startParam3 = 0.00016
startParam4 = 0.0000002
startParam5 = 0.0000000000238
startParam6 = 0.000000003202899 
startParam7 = 0.000000000000041
startParam8 = 0.000000000000001
startParam9 = 0.0000000000000001

a0 = tensorflow.Variable([startParam0], dtype=tensorflow.float32)  
a1 = tensorflow.Variable([startParam1], dtype=tensorflow.float32) 
a2 = tensorflow.Variable([startParam2], dtype=tensorflow.float32) 
a3 = tensorflow.Variable([startParam3], dtype=tensorflow.float32)  
a4 = tensorflow.Variable([startParam4], dtype=tensorflow.float32)  
a5 = tensorflow.Variable([startParam5], dtype=tensorflow.float32)  
a6 = tensorflow.Variable([startParam6], dtype=tensorflow.float32)  
a7 = tensorflow.Variable([startParam7], dtype=tensorflow.float32)  
a8 = tensorflow.Variable([startParam8], dtype=tensorflow.float32)  
a9 = tensorflow.Variable([startParam9], dtype=tensorflow.float32)  
a10 = tensorflow.Variable([startParam2], dtype=tensorflow.float32)  
a11 = tensorflow.Variable([startParam3], dtype=tensorflow.float32) 
a12 = tensorflow.Variable([startParam4], dtype=tensorflow.float32)  
a13 = tensorflow.Variable([startParam4], dtype=tensorflow.float32)  
a14 = tensorflow.Variable([startParam3], dtype=tensorflow.float32)  
a15 = tensorflow.Variable([startParam2], dtype=tensorflow.float32)  
a16 = tensorflow.Variable([startParam1], dtype=tensorflow.float32)  
a17 = tensorflow.Variable([startParam1], dtype=tensorflow.float32)  
a18 = tensorflow.Variable([startParam2], dtype=tensorflow.float32)  
a19 = tensorflow.Variable([startParam3], dtype=tensorflow.float32)  

variableArray = [a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11, a12, a13, a14, a15, a16, a17, a18, a19]

curr_a0 = 0
curr_a1 = 0
curr_a2 = 0
curr_a3 = 0
curr_a4 = 0
curr_a5 = 0
curr_a6 = 0
curr_a7 = 0
curr_a8 = 0
curr_a9 = 0
curr_a10 = 0
curr_a11 = 0
curr_a12 = 0
curr_a13 = 0
curr_a14 = 0
curr_a15 = 0
curr_a16 = 0
curr_a17 = 0
curr_a18 = 0
curr_a19 = 0
curr_loss = 0

outputArray = [curr_a0, curr_a1, curr_a2, curr_a3, curr_a4, curr_a5, curr_a6, curr_a7, curr_a8, curr_a9, curr_a10, curr_a11, curr_a12, curr_a13, curr_a14, curr_a15, curr_a16, curr_a17, curr_a18, curr_a19]

#Training Data (inputs/outputs)  
x = tensorflow.placeholder(dtype=tensorflow.float32)  
y = tensorflow.placeholder(dtype=tensorflow.float32)  

# recive parameters from java
lines = s1.split("[")[1].split("]")[0].split(",");
y_train = []
counter = 0
for l in lines:
    s = l.strip()
    y_train.append(float(l.strip()))
    counter = counter + 1

lines = s2.split("[")[1].split("]")[0].split(",");
x_train = []
counter = 0
for l in lines:
    s = l.strip()
    x_train.append(float(s))
    counter = counter +1

grade = int(s3.strip());
linear_model = tensorflow.Variable([0], dtype=tensorflow.float32)  
sessionInput = [ None for k in range( grade + 2 ) ]
sessionOutput = [ None for k in range( grade + 2 ) ]

for i in range(grade + 1):
    s = 1
    for j in range (i): 
        s = s * x 
    linear_model = linear_model + variableArray[i] * s
    sessionInput[i] = variableArray[grade - i] 
    sessionOutput[i] = outputArray[grade - i]

#Linear Regression Loss Function - sum of the squares  
squared_deltas = tensorflow.square(linear_model - y_train)  
loss = tensorflow.reduce_sum(squared_deltas) 
curr_loss = 0
sessionInput[grade+1] = loss
sessionOutput[grade+1] = curr_loss

#Gradient descent optimizer  
#optimizer = tensorflow.train.GradientDescentOptimizer(learning_rate=0.01)  
optimizer = tensorflow.train.AdamOptimizer(learning_rate=0.25)  
train = optimizer.minimize(loss=loss)  

#Creating a session 
sess = tensorflow.Session()  

writer = tensorflow.summary.FileWriter("/tmp/log/", sess.graph)  

#Initializing variables  
init = tensorflow.global_variables_initializer()  
sess.run(init)  

#Optimizing the parameters  
for i in range(10000):  
    sess.run(train, feed_dict={x: x_train, y: y_train})  

#Print the parameters and loss  
outputArray = sess.run(sessionInput, {x: x_train, y: y_train})  
s = ""
for i in range(0, grade + 2):
    s = s + str(outputArray[i])
    if not(i == grade + 1):
        s = s + ","

print(s)  

writer.close()  

sess.close()

和Java代码：

BufferedReader stdInput = new BufferedReader(new InputStreamReader(process.getInputStream()));
PrintWriter stdOutput = new PrintWriter(new OutputStreamWriter(process.getOutputStream()));

System.out.println(values); //y_train
stdOutput.println(values);
stdOutput.flush();

System.out.println(dates); //x_train
stdOutput.println(dates);
stdOutput.flush();

stdOutput.println(grade); //grade=9
stdOutput.flush();

// for reading the ouput from stream
String s1 = null;

List<Double> coefList = new ArrayList<Double>();

while ((s1 = stdInput.readLine()) != null) {
    System.out.println(s1);
    if (s1.indexOf("\\[") == -1) {
        Integer n = s1.split("\\[").length;
        if (n == grade + 2) { 
            try {
                for(int i = 0; i < grade+1; i++){
                    coefList.add( Double.parseDouble(s1.split("\\[")[1 + i].split("\\]")[0]));
                }
            } catch(NumberFormatException E) {
                System.err.println("Received NAN instead of a number for the parameters!");
            }
        }
    }
}

最后，它会打印系数。

在选项（b）中，结果是这样的：

[4.3143527e-06],[-1.3367743e-05],[5.2168118e-05],[0.00026263],[0.0007228],[0.00161756],[0.00351851],[0.10692083],[0.11532306],[2.0412211],17321398.0

第11个数字的预测值是：58.70016770335398（似乎还可以）

在选项（a）和（c）中，结果是：[nan]，[nan]，[nan]，[nan]，[nan]，[nan]，[nan]，[nan]，[ nan]，[nan]，nan

没有错误！请帮我解决这个问题...因为我需要参数以毫秒为单位（如选项（c）所示）

让我知道您是否还需要其他任何内容。只要结果保持在参数范围内，就可以考虑使用任何选项，甚至可以降低功能等级或其他功能。

谢谢。

0 个答案:

没有答案
相关问题
最新问题