how to determine hidden layer of patternnet using tainscg function (backpropagation) for better accuration

My input is a 127x108 matrix of 108 attributes (the features) of 127 training dataset. the features are from image zoning 3x3. And then

My target is a 127x36 (row x column) matrix, column indicate what letter it is. (i.e : 0, 1, 2, 3, 4, 5, 6, 7, a, b, c,... i, I, j, k, l,... u, U, v,.. z). Total are 36 characters.

Thanks before.

[Bachtiar Muhammad Lubis]

i've just tried to train my data set, but the cross-entropy and percent error is so far to reach 0. i've tried to set the hidden layers value to 10, 30, 35. but havn't gotten any proper result.

would you master help me how to solve this issue? please.