Correlation values shown by 'nntraintool -> regression plots' differs when calculated manually for 'dividerand' function

Hello all, I ran my ANN model to train 60 training sets out which 42 were used for training, 9 used for validation and 9 were used for testing. This values were shown in 'nntraintool - > regression plots'. I tried to calculate them manually also and they were matching for divideblock function.

But same thing if I do for dividerand function with same ratio as used above then I am not getting the values of correlation which were displayed in 'nntraintool - > regression plots'.

Please help me regarding how to use dividerand in NN and use appropriate values for model evaluation.

Regards,

Answer 1

[Sean de Wolski]

dividerand uses random indices.

From the doc

Divide targets into three sets using random indices

So of course the values will be a little different.

[Kalpesh Patil]

But I am saving the indices used for every epoch so my correlation values with that indices should match.

[Kalpesh Patil]

Thanks Sean, Its giving same value now, I retrieved the indices after training from object 'tr'.