Can I train the SAME neural network with multiple datasets using Neural Fitting Tool?

13 次查看（过去 30 天）

Georgios Etsias 2018-7-17

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/410780-can-i-train-the-same-neural-network-with-multiple-datasets-using-neural-fitting-tool

评论： Emily LaPrime 2021-4-1

After training the network for the first time I get the option of either retraining it with the same data set, or by pressing NEXT just to test it over a new one. If I go back and add a new data set it seems that the former network is lost and training begins from scratch.

2 个评论
显示无隐藏无

Aquatris 2018-7-17

You can combine all the data sets you want to use to train your NN into a single dataset simply by cascading them.

For instance, you have data set 1 with input u1 and output y1, and data set 2 with u2 and y2. than you create a new data set by simply defining u = [u1;u2] and y = [y1;y2].

Georgios Etsias 2018-7-18

The training datasets should have the same dimension with the dataset I'll use for prediction though?

请先登录，再进行评论。

请先登录，再回答此问题。

回答（1 个）

Greg Heath 2018-7-17

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/410780-can-i-train-the-same-neural-network-with-multiple-datasets-using-neural-fitting-tool#answer_329207

在 MATLAB Online 中打开

In order to not forget a learned dataset while learning a new one:

 Continue training with a mixture (NOT a concatenation!) of the 
two. The training ratio should be approximately the same as the 
expected ratio in the implementation space.

Hope this helps.

Thank you for formally accepting my answer

Greg

3 个评论
显示 1更早的评论隐藏 1更早的评论

Greg Heath 2019-3-1

See my answer before your comment.

Any specific questions?

Greg

Emily LaPrime 2021-4-1

Greg,

Could you elaborate on creating a mixture of the two data sets? I am trying to train a network to classify using 4 sets of training data. I need the network to learn each set of data so that when I use the testing data the output is classified using the results of the data used to train.

请先登录，再进行评论。

请先登录，再回答此问题。