Error evaluation on convolutional net

Question

Simon 2017-3-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/331246-error-evaluation-on-convolutional-net

评论： Jay 2017-5-18

I have trained a conv net, using imds function as datastore and trainNetwork for the training. I would like to plot something similar to the matlab function "plotperform", but I cannot get that function working, since it requires another type of input, which the trainNetwork does not output. I need to know, how to plot the validation error together with the training error.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Nanda Gupta 2017-3-30

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/331246-error-evaluation-on-convolutional-net#answer_261061

I understand that you want to plot the validation error along with the training error for a ConvNet, to generate a graph similar to the one generated by "plotperform" function.

In the training options for "trainNetwork", you can set the "OutputFcn" to a custom function so that you can explicitly plot the accuracy at each iteration of the training process. Check out this documentation for more information: https://www.mathworks.com/help/nnet/ref/trainingoptions.html#bvnn789

2 个评论
显示无隐藏无

Johannes Kvam 2017-4-28

As far as I can tell this only passes the training info structure which you then can use to show the training accuracy. For validation testing you'd need to also pass the current network to calculate its performance on an independent validation set to get the validation performance. Then show the validation performance together with the training performance. Is there anyway of doing this?

Jay 2017-5-18