The objective is offline handwritten Devnagari numeral recognition for which some features are extracted. · Training Data: handwritten hindi numerals - 10 numerals with 200 samples for each numeral. · Testing data: 10 numerals with 40 samples for each numeral.
nntool inputs- Input Matrix: mxn = 36x (200x10)=72000 where m=no. of features (36) n: size of input vector (200 x 10) i.e. 200 samples per numeral Target Matrix: pxn = 10x(200x10)=20000 where p= number of classes =10 n= vector size=(200x10) i.e. 200 samples per numeral for 10 numerals Test Matrix: mxn=36x(40x10) ) i.e. 40 samples per numeral Network: I tried Probabilistic type, Radial Basis ( exact fit) and more before , but it gives error- “Out of memory. Type help memory for your options”
Hello, can you provide some information about connection map character of used network and some example topologies you tried?
I tried Feed Forward back Propagation network, Training function- TrainSCG, Performance: SSE
Layers:2, Number of neurons in first layer:72
Transfer function: Tansig
..... but the point is whatever I do, network not created saying out of memory....
Okay, so there is no way that it would fill your memory. It is likely that you just have to debug something in your implementation, that incorrectly allocates memory. Probably some incorrect constraint on data structures.
- Badges
- Science topic