Request for a Clarification on Every Run Output Differences

Discussion:

Othman Soufan

2012-03-18 23:07:08 UTC

Dear Group,

After I have installed FANN, I went and tried to run the example on the
"Getting Started" page.
However, each time I run the program on the Xor.data, I get a different
output.
I would like to mention that I have combined the training and testing
programs in one program.

Some output samples are:
Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2505536079. Bit fail 4.
Epochs 26. Current error: 0.0007957527. Bit fail 0.
xor test (-1.000000,1.000000) -> -319531589830587711488.000000

Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2500049174. Bit fail 4.
Epochs 23. Current error: 0.0009584196. Bit fail 0.
xor test (-1.000000,1.000000) -> -0.000002

Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2502186596. Bit fail 4.
Epochs 30. Current error: 0.0009348781. Bit fail 0.
xor test (-1.000000,1.000000) -> 0.000000

I would like to know why I am receiving such weird outputs that differs
from each other.

Regards,
Othman

Steffen Nissen

2012-03-19 01:06:23 UTC

Permalink

Weight are initialized randomly, so different result is expected. The very
large number you observed in the first run is, however, not expected.

Post by Othman Soufan
Dear Group,
After I have installed FANN, I went and tried to run the example on the

"Getting Started" page.

Post by Othman Soufan
However, each time I run the program on the Xor.data, I get a different

output.

Post by Othman Soufan
I would like to mention that I have combined the training and testing

programs in one program.

Post by Othman Soufan
Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2505536079. Bit fail 4.
Epochs 26. Current error: 0.0007957527. Bit fail 0.
xor test (-1.000000,1.000000) -> -319531589830587711488.000000
Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2500049174. Bit fail 4.
Epochs 23. Current error: 0.0009584196. Bit fail 0.
xor test (-1.000000,1.000000) -> -0.000002
Max epochs 500000. Desired error: 0.0010000000.
Epochs 1. Current error: 0.2502186596. Bit fail 4.
Epochs 30. Current error: 0.0009348781. Bit fail 0.
xor test (-1.000000,1.000000) -> 0.000000
I would like to know why I am receiving such weird outputs that differs
from each other.
Regards,
Othman

--
Best Regards,
Steffen Nissen, MSc
http://www.linkedin.com/in/steffennissen

Fábio Blessa

2012-03-19 01:25:16 UTC

Permalink

Send us the code plz...
For sure has a small mistake.

BR,

Fábio

Post by Steffen Nissen
Weight are initialized randomly, so different result is expected. The very
large number you observed in the first run is, however, not expected.

Post by Othman Soufan
Dear Group,
After I have installed FANN, I went and tried to run the example on the

"Getting Started" page.

Post by Othman Soufan
However, each time I run the program on the Xor.data, I get a different

output.

Post by Othman Soufan
I would like to mention that I have combined the training and testing

programs in one program.

--
Best Regards,
Steffen Nissen, MSc
http://www.linkedin.com/in/steffennissen
------------------------------------------------------------------------------
Try Windows Azure free for 90 days Click Here
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Fann-general mailing list
https://lists.sourceforge.net/lists/listinfo/fann-general

Othman Soufan

2012-03-19 01:58:30 UTC

Permalink

Thanks for the immediate response.

The code is as follows:

#include <stdio.h>
#include <stdlib.h>
#include <fann.h>
#include "floatfann.h"
#include "libMSVM.h" // Generic structure and function declarations
#include "libtrainMSVM.h" // Training functions (not required for
predictions only)
#include "libevalMSVM.h" // Evaluation functions (also used during
training)
#include "memory.c"
#include "math.h"

//1. Consider having a matrix for the output layer when we have multiple
class i.e. one-to-all representation

int main()
{
const unsigned int num_input = 2;
const unsigned int num_output = 1;
const unsigned int num_layers = 3;
const unsigned int num_neurons_hidden = 3;
const float desired_error = (const float) 0.001;
const unsigned int max_epochs = 500000;
const unsigned int epochs_between_reports = 1000;

//Training the ANN
struct fann *ann = fann_create_standard(num_layers, num_input,
num_neurons_hidden, num_output);

fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_SIGMOID_SYMMETRIC);

fann_train_on_file(ann, "xor.data", max_epochs, epochs_between_reports,
desired_error);

//Testing the ANN
fann_type *calc_out;
fann_type input[2];

input[0] = -1;
input[1] = 1;
calc_out = fann_run(ann, input);
printf("xor test (%f,%f) -> %f\n", input[0], input[1], calc_out[0]);

fann_destroy(ann);

return 0;
}

Post by FÃ¡bio Blessa
Send us the code plz...
For sure has a small mistake.
BR,
Fábio

Post by Steffen Nissen
Weight are initialized randomly, so different result is expected. The
very large number you observed in the first run is, however, not expected.

Post by Othman Soufan
Dear Group,
After I have installed FANN, I went and tried to run the example on the

"Getting Started" page.

Post by Othman Soufan
However, each time I run the program on the Xor.data, I get a different

output.

Post by Othman Soufan
I would like to mention that I have combined the training and testing

programs in one program.

------------------------------------------------------------------------------
Try Windows Azure free for 90 days Click Here
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Fann-general mailing list
https://lists.sourceforge.net/lists/listinfo/fann-general

--
MS Candidate, Class of 2010
Mathematical and Computer Sciences and Engineering
King Abdullah University of Science and Technology
Tuwal, Jeddah, KSA.
Mobile: +966506134003

Othman Soufan

2012-03-19 12:29:13 UTC

Permalink

I would like to update you that:

const unsigned int num_output = 2;

solves the problem.

So, instead of setting num_output = 1 as listed on the Getting Started page,
num_output = 2 seems to overcome the problem of large output differences.

Currently, whenever I execute the program I get the following output:
xor test (-1.000000,1.000000) -> 0.000000

So, would you kindly confirm if this is the proper solution or am I missing
something?

Regards,
Othman

Post by Othman Soufan
Thanks for the immediate response.
#include <stdio.h>
#include <stdlib.h>
#include <fann.h>
#include "floatfann.h"
#include "libMSVM.h" // Generic structure and function declarations
#include "libtrainMSVM.h" // Training functions (not required for
predictions only)
#include "libevalMSVM.h" // Evaluation functions (also used during
training)
#include "memory.c"
#include "math.h"
//1. Consider having a matrix for the output layer when we have multiple
class i.e. one-to-all representation
int main()
{
const unsigned int num_input = 2;
const unsigned int num_output = 1;
const unsigned int num_layers = 3;
const unsigned int num_neurons_hidden = 3;
const float desired_error = (const float) 0.001;
const unsigned int max_epochs = 500000;
const unsigned int epochs_between_reports = 1000;
//Training the ANN
struct fann *ann = fann_create_standard(num_layers, num_input,
num_neurons_hidden, num_output);
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_SIGMOID_SYMMETRIC);
fann_train_on_file(ann, "xor.data", max_epochs,
epochs_between_reports, desired_error);
//Testing the ANN
fann_type *calc_out;
fann_type input[2];
input[0] = -1;
input[1] = 1;
calc_out = fann_run(ann, input);
printf("xor test (%f,%f) -> %f\n", input[0], input[1], calc_out[0]);
fann_destroy(ann);
return 0;
}

Post by FÃ¡bio Blessa
Send us the code plz...
For sure has a small mistake.
BR,
Fábio

Post by Steffen Nissen
Weight are initialized randomly, so different result is expected. The
very large number you observed in the first run is, however, not expected.

Post by Othman Soufan
Dear Group,
After I have installed FANN, I went and tried to run the example on

the "Getting Started" page.

Post by Othman Soufan
However, each time I run the program on the Xor.data, I get a

different output.

Post by Othman Soufan
I would like to mention that I have combined the training and testing

programs in one program.

Steffen Nissen

2012-03-19 13:12:29 UTC

Permalink

That should not fix anything, as the XOR problem only has one output.

Please let me know how you compile the source, perhaps you are linking with
doublefann instead of floatfann.

You can also try to run some of the other examples in:
http://fann.git.sourceforge.net/git/gitweb.cgi?p=fann/fann;a=tree;f=examples;hb=HEAD

Like:
http://fann.git.sourceforge.net/git/gitweb.cgi?p=fann/fann;a=blob;f=examples/cascade_train.c;h=35bae4b9a0e022770c97995d6301aab6e08680e8;hb=HEAD

Steffen

Post by Othman Soufan
const unsigned int num_output = 2;
solves the problem.
So, instead of setting num_output = 1 as listed on the Getting Started page,
num_output = 2 seems to overcome the problem of large output differences.
xor test (-1.000000,1.000000) -> 0.000000
So, would you kindly confirm if this is the proper solution or am I missing
something?
Regards,
Othman
Thanks for the immediate response.
#include <stdio.h>
#include <stdlib.h>
#include <fann.h>
#include "floatfann.h"
#include "libMSVM.h" // Generic structure and function declarations
#include "libtrainMSVM.h" // Training functions (not required for

predictions only)

Post by Othman Soufan
#include "libevalMSVM.h" // Evaluation functions (also used during

training)

Post by Othman Soufan
#include "memory.c"
#include "math.h"
//1. Consider having a matrix for the output layer when we have multiple

class i.e. one-to-all representation

Post by Othman Soufan
int main()
{
const unsigned int num_input = 2;
const unsigned int num_output = 1;
const unsigned int num_layers = 3;
const unsigned int num_neurons_hidden = 3;
const float desired_error = (const float) 0.001;
const unsigned int max_epochs = 500000;
const unsigned int epochs_between_reports = 1000;
//Training the ANN
struct fann *ann = fann_create_standard(num_layers, num_input,

num_neurons_hidden, num_output);

Post by Othman Soufan
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_SIGMOID_SYMMETRIC);
fann_train_on_file(ann, "xor.data", max_epochs,

epochs_between_reports, desired_error);

Post by Othman Soufan
//Testing the ANN
fann_type *calc_out;
fann_type input[2];
input[0] = -1;
input[1] = 1;
calc_out = fann_run(ann, input);
printf("xor test (%f,%f) -> %f\n", input[0], input[1], calc_out[0]);
fann_destroy(ann);
return 0;
}
Send us the code plz...
For sure has a small mistake.
BR,
Fábio
Weight are initialized randomly, so different result is expected. The

very large number you observed in the first run is, however, not expected.