Help on Cascade training

[sumit]

Hi
I am making an OCR for my project using fann.
I am using Cascade training .
i am attaching my training code below

Initially i used
fann_set_activation_function_output(ann,FANN_SIGMOID_SYMMETRIC);
and
fann_set_activation_function_hidden(ann,FANN_LINEAR);

I am taking image pixel as input (25*25)
for black pixel : -1
for white pixel : 1
my output is 1 corresponding to right char (what i fixed)
and -1 for all.

It get converged in what it write "Neurons 10 Current Error......."

At last it write training error 0 test error 0 (in my case test file and
train file are same)
but when i pass my OCR the same trained char it gives error .

Is that (training error 0 and testing error 0) not reliable ????

What are the impact of
cascade activation functions[]
cascade activation steepnesses[]
RPROP increase factor
RPROP decrease factor
on training ?????

Should i use default or should i change it ????


Thanks

////////////////////////////////////////////////////////////////CODE////////////////////////////////////////////////////////////////////////////

#include <stdio.h>

#include "fann.h"


int main()
{
struct fann *ann;
struct fann_train_data *train_data, *test_data;
const float desired_error = (const float) 0.0000000;
unsigned int max_neurons = 1500;
unsigned int neurons_between_reports = 1;
const float learning_rate = (const float) 0.001;
int k;

printf("Reading data.\n");

train_data = fann_read_train_from_file("cascadeOCR.train");
test_data = fann_read_train_from_file("cascadeOCR.test");

printf("Creating network.\n");

ann = fann_create_shortcut(2, fann_num_input_train_data(train_data),
fann_num_output_train_data(train_data));

fann_set_training_algorithm(ann, FANN_TRAIN_RPROP);

fann_set_activation_function_hidden(ann,FANN_LINEAR);//FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann,FANN_LINEAR);
//FANN_SIGMOID_SYMMETRIC);
fann_set_train_error_function(ann, FANN_ERRORFUNC_LINEAR);
fann_set_learning_rate(ann, learning_rate);

fann_print_parameters(ann);

printf("Training network.\n");

fann_cascadetrain_on_data(ann, train_data, max_neurons,
neurons_between_reports, desired_error);

fann_print_connections(ann);

printf("\nTrain error: %f, Test error: %f\n\n", fann_test_data(ann,
train_data),
fann_test_data(ann, test_data));

printf("Saving network.\n");

fann_save(ann, "cascadeOCR.net");

printf("Cleaning up.\n");
fann_destroy_train(train_data);
fann_destroy_train(test_data);
fann_destroy(ann);
return 0;
}