Groups of observations for Neural Network?
2 visualizzazioni (ultimi 30 giorni)
Mostra commenti meno recenti
I have an IxN matrix of input data and an OxN matrix of output data. The N columns are observations. Each observation column is unique, however they exist in groups of an unknown number of observation columns per day. Each of these groups of obs from one day should be analyzed by the tree or network as a single related entity.
I just found the grouping variables functionality in Matlab, which is exactly what I need: http://www.mathworks.com/help/stats/grouping-variables.html Is there anyway to incorporate this into either NNTBX or ClassificationTree.fit?
0 Commenti
Risposta accettata
Greg Heath
il 25 Giu 2013
Input and output vector dimensions are fixed.
If you wish to have variable dimensions, the only way I can think of effectively relaying that info to the net is to fix the maximum dimensions and use zeros when vectors are shorter than the maximum. If you use NaNs, the outputs will be NaNs.
4 Commenti
Greg Heath
il 25 Giu 2013
The grouping variables in the reference are useful for UNSUPERVISED learning.
Your problem is one of SUPERVISED learning. You teach the net, via training, that if this vector is the input, then that target vector should be the output.
If your training set is representative of the total data set and categories don't significantly overlap, then you should be OK.
Più risposte (1)
Greg Heath
il 18 Giu 2013
To use the NNTBX, variables are rows and observations are columns.
You train the net by giving it pairs of input vectors and corresponding output targets.
The sizes of the input matrix and corresponding target matrix are
[ I N ]= size(input)
[ O N ] =size(target)
for N I-dimensional input vectors and the corresponding N O-dimensional output target vrctors.
Are you able to explain the relationship that you want to have between a typical I-dimensional input vector and the correspondidng O-dimensional output vector?
Greg
Vedere anche
Prodotti
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!