How does division of data occurs in multi sequence data sets in neural networks?

I have used the catsample() command to create a 1x400s cell each cell has a 2x30 data points. I am using the multiple sequence concatenation option because I will be predicting gait of different conditions (walking, stairs up and stairs down) so I will have desperate trials of different lengths.

Currently, I have 10 trials of each that is why I have 30 point within a cell and in the order of 1 walking trial --> 1 stair up --> 1 stair down --> 1 walk and so on. I need to have full control of which data points are being used for training, which ones for validations and which ones for testing so I have the same number of trials per condition in each division.

I understand how the net.divideFcn works and the 4 types of division in CONCURRENT series. However, there is no documentation on how multiple sequences are affected and indexed. Currently I am using the 'divideblock' with 70/20/10 and I am assuming that the way I have organized my data set, I am taking 7 complete trials per condition for training. But it might not be the case.

1. Can someone explain me how each division type works in multiple sequences? How is the indexing ordered?

2. Should I use 'divideind' instead?

3. If I want to use

[trainInd,valInd,testInd] = divideblock(Q,trainRatio,valRatio,testRatio)

what do I feed as Q? all the 2x30x400s?

Thank you,

Answer 1

[ErikaZ]

For default 'divideblock', it takes 70 percent of 400s time points for training. If needed to take 70% of the Q = 30, the mode must be changed to sample

net.divideMode = 'sample';

This will take 21 points from every of the 400s time points for training.

Indexes and masks can be checked with

   [net,tr] = train(net,inputs,targets,inputStates,layerStates);

    tr.trainInd 
    tr.valInd
    tr.testInd
    tr.trainMask
    tr.valMask
    tr.testMask