How to change the input data/built in code for Matlab's Regression CNN example'??

4 次查看（过去 30 天）

Sho Wright 2021-3-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/784381-how-to-change-the-input-data-built-in-code-for-matlab-s-regression-cnn-example

评论： Sho Wright 2021-4-6

I've been using the in built 'digitTrain4DArray' and 'digitTest4DArray' to train my own regression network, by replacing my own training images and validation images in the folder that these two commands pull from. In the original example, there are also 2 excel files that was used, 'digitTrain' and 'digitTest', which had both predictors (images and digits) and response (angle of rotation). I've collected numerical validation data of 'true' values to replace 'digitTest' (I believe this is right?), but I'm not sure what to do about replacing 'digitTrain'. My supervisor has said that I don't need any numerical data such as 'digitTrain', and that I can alter the built-in code to train the network with simply my training images, and comparing these against the validation images/values will be enough.

I've looked at the relevant codes and 'digitTrain4DArrayData' is dependent on converting a .csv file into a 4D array. Is there any way I can remove the numerical vector 'digitTrain', and just train using my images?

Sorry for the paragraph, would really appreciate some help, I'm quite confused a fairly stressed about this!

2 个评论
显示无隐藏无

Mahesh Taparia 2021-3-31

Can you tell what you want to do? Are you using digitTrain4DArrayData or other dataset?

Sho Wright 2021-4-6

Hi Mahesh,

I have resolved the problem. I've replaced the 'digit' dataset in the file path for digitTrain4DArray with my own now. At the time of writing this question, I was really confused because I did not know what my numerical training data should be. I then realised that my training data should be collected the same as my validation data, as I'm doing supervised learning. It really was a silly moment!

Many thanks for the response

请先登录，再进行评论。

请先登录，再回答此问题。