How to compute RMSE on training set and validation set in LSTM regression?

Now I'm trying to do seq2seq regression using LSTM. I can't understand the algorithm of RMSE. My input is a 100*1 cell, and evey cell is 2*100 double matrix. The output is a 100*1 cell, and every cell is 1*100 double vector. This means, numFeatures=2 and numResponses=1. I also set validation set by trainingOptions: 'ValidationData',{XValid,YValid}, XValid and YValid both are 100*1 cell. But what confuses me is how to compute RMSE on training set and validation set. Under my understanding, if my input is a 100*1 cell, then my output is also a 100*1 cell. Every cell of output is 1*100 double. So I don't know how to calculate the RMSE between two cells, I mean output and Yvalid. I think it will retrun a 100*1 cell, every cell is a number of the RMSE between two vector, but it returns just a number. Is the mean of every cell or other statistic?Or is there some other algorithm?

Answer 1

[Taylor]

You'll want to look at the info object returned by trainnet.See this previous post for more info.