如何使用我的NeuRal网络模型? [英] How To Use My NeuRal Network Model?

问题描述

我是NN的新手，喜欢用它测试一些策略.我正在使用一个用于NN的C ++建模代码来设计NN.
该代码为我提供了权重和NN的其他重要参数.
我的问题是如何使用此信息来预测我的其他数据并测试我的模型如何工作?
例如我有2000个数据.
3输入和1输出
使用1个隐藏层.
模型为我做6个重量.
我想知道如何使用这些数据来测试我的NN吗?
问候，

Hi,
I am new in NN and like test some strategies with that. I am using one C++ Modelling Code for NN that design NN.
this code makes for me weights and other important parameters of NN.
my problem is how use this info for predict other my data and test how my model will work?
For example I have 2000 data.
3 Inputs and 1 output
Using 1 hidden layer.
model makes 6 weights for me.
I want know How I can use this data for test my NN?
Regards,

推荐答案

无法确定神经网络何时100％完成学习"给定方程.输入所有输入后，您可以随机输入一些数据，看看给定的结果是否符合您的期望.如果重复多次，则可以确定神经网络的正确性"，例如10中的9.

-Daan

Hi,

There isn''t a way to determine when your Neural Network is 100% done "learning" the given equation. After feeding it all input you have, you can randomly input some data, see if the given result is what you expect. If repeated muliple times you can determine a "correctness" of the Neural Network, for example 9 out of 10.

-- Daan

有很多方法可以测试神经网络是否正常工作.

我遇到的最常见的是:

您可以有一个学习集，然后当错误的总和下降到某个阈值以下时，将其评估为完成学习.
这种方法的问题在于过度训练可能会成为一个问题. IE.神经网络仅学习您呈现的模式，而不学习一般的问题解决方案.

您还可以拥有一组验证数据，该数据不在学习集中，而是用于测试神经网络是否已了解一般问题的解决方案，因此，仅将这个错误与阈值进行比较.这可能会导致训练时间延长，因为如果验证集异常，那么无论训练有多熟练，某些Nets都将无法学习验证集.

您还可以为要学习的纪元数设置任意限制.这是最糟糕的方法，因为您不能保证它在时限内根本没有学到任何东西.

There are many ways to test that a Neural Network is working correctly.

The most common I have come across are:

You could have a learning set and then assess it to be finished learning when the sum square of the errors drops below a certain threshold.
The problem with this approach is over training can become an issue. I.E. The Neural Net learns only the patterns you present and not that general problem solution.

You can also have a set of validation data which is not in the learning set but is used to test that the neural net has learnt the general problem solution and so it''s only the error on this set which is compared to the threshold. This can lead to long training times as some Nets will not learn the validation set no matter how well trained if the validation set is unusual.

You could also set an arbitrary limit on the number of epochs, that you will learn for. This is the worst method because you cannot guarantee that it has learnt anything at all in the time limit.

这篇关于如何使用我的NeuRal网络模型?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！