Parameters learning with jSmile

dtodor · Post by **dtodor** » Tue Jun 02, 2009 8:48 am

Hi,

I'm using Visual C++ Express Edition (Version 9.0.30729.1 SP):

Microsoft Visual Studio 2008
Version 9.0.30729.1 SP
Microsoft .NET Framework
Version 3.5 SP1

Installed Edition: VC Express

Microsoft Visual C++ 2008 91909-152-0000052-60665
Microsoft Visual C++ 2008

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB945282) KB945282
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/945282.

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB946040) KB946040
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/946040.

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB946308) KB946308
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/946308.

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB947540) KB947540
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/947540.

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB947789) KB947789
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/947789.

Hotfix for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU (KB948127) KB948127
This hotfix is for Microsoft Visual C++ 2008 Express Edition with SP1 - ENU.
If you later install a more recent service pack, this hotfix will be uninstalled automatically.
For more information, visit http://support.microsoft.com/kb/948127.

Here is the code of the test application:

Code: Select all

#include <iostream>
#include <vector>

#include "smile.h"
#include "smilearn.h"

int main (int argc, char * const argv[]) {
	
	DSL_network network;
	if (network.ReadFile("abc-h-def-network-randomized.xdsl") != DSL_OKAY) {
		std::cout << "Unable to read network\n";
		return -1;
	}
	std::cout << "1. Successfully read network\n";
	
	DSL_dataset dataset;
	if (dataset.ReadFile("abc-h-def-network-original.txt") != DSL_OKAY) {
		return -2;
	}
	std::cout << "2. Successfully read data set\n";
	
	
	std::string errMsg;
	
	
	std::vector<DSL_datasetMatch> matches;
	if (dataset.MatchNetwork(network, matches, errMsg) != DSL_OKAY) {
		return -3;
	}
	std::cout << "3. Successfully calculated matches\n";
	
	for (unsigned int i=0; i<matches.size(); i++) { 
		const DSL_datasetMatch &m = matches[i]; 
		printf("%d col=%d slice=%d h=%d %s\n", i, m.column, m.slice, m.node, network.GetNode(m.node)->GetId()); 
	} 
	
	std::vector<int> fixedNodes;
	
	DSL_em em;
	//em.SetEquivalentSampleSize(2);
	em.SetRandomizeParameters(true);
	if (em.Learn(dataset, network, matches) != DSL_OKAY) {
		std::cout << errMsg;
		return -4;
	}	
	std::cout << "4. Successfully learnt network\n";
	
	if (network.WriteFile("abc-h-def-network-learnt.xdsl") != DSL_OKAY) {
		return -5;
	}
	std::cout << "5. Successfully written learnt network\n";
	
	for (int h = network.GetFirstNode(); h >= 0; h = network.GetNextNode(h)) 
	{ 
		printf("%d %s\n", h, network.GetNode(h)->GetId()); 
		const DSL_Dmatrix *mtx = network.GetNode(h)->Definition()->GetMatrix(); 
		for (int i = 0; i < mtx->GetSize(); i ++) 
		{ 
			printf("%g ", (*mtx)[i]); 
		} 
		printf("\n"); 
	} 
    
    return 0;
}

Under Mac OS X 10.5 I get the following results:

Code: Select all

	//em.SetEquivalentSampleSize(2);
	em.SetRandomizeParameters(true);

1. Successfully read network
2. Successfully read data set
3. Successfully calculated matches
0 col=0 slice=0 h=0 A
1 col=1 slice=0 h=1 B
2 col=2 slice=0 h=2 C
3 col=3 slice=0 h=3 H
4 col=4 slice=0 h=4 D
5 col=5 slice=0 h=5 E
6 col=6 slice=0 h=6 F
4. Successfully learnt network
5. Successfully written learnt network
0 A
0.5005 0.4995
1 B
0.4925 0.5075
2 C
0.5275 0.4725
3 H
0.195312 0.804688 0.518448 0.481552 0.584179 0.415821 0.86999 0.13001 0.0174197 0.98258 0.808501 0.191499 0.162883 0.837117 0.782767 0.217233
4 D
0.201708 0.798292 0.677407 0.322593
5 E
0.588047 0.411953 0.643097 0.356903
6 F
0.323372 0.676628 0.386902 0.613098

Code: Select all

	em.SetEquivalentSampleSize(2);
	em.SetRandomizeParameters(false);

1. Successfully read network
2. Successfully read data set
3. Successfully calculated matches
0 col=0 slice=0 h=0 A
1 col=1 slice=0 h=1 B
2 col=2 slice=0 h=2 C
3 col=3 slice=0 h=3 H
4 col=4 slice=0 h=4 D
5 col=5 slice=0 h=5 E
6 col=6 slice=0 h=6 F
4. Successfully learnt network
5. Successfully written learnt network
0 A
0.500759 0.499241
1 B
0.492636 0.507364
2 C
0.527896 0.472104
3 H
0.195727 0.804273 0.518448 0.481552 0.584179 0.415821 0.86999 0.13001 0.0174197 0.98258 0.808501 0.191499 0.162883 0.837117 0.782767 0.217233
4 D
0.20216 0.79784 0.677407 0.322593
5 E
0.588571 0.411429 0.643097 0.356903
6 F
0.324728 0.675272 0.386902 0.613098

The results for Windows XP SP3 are as follows:

Code: Select all

	//em.SetEquivalentSampleSize(2);
	em.SetRandomizeParameters(true);

1. Successfully read network
2. Successfully read data set
3. Successfully calculated matches
0 col=0 slice=0 h=0 A
1 col=1 slice=0 h=1 B
2 col=2 slice=0 h=2 C
3 col=3 slice=0 h=3 H
4 col=4 slice=0 h=4 D
5 col=5 slice=0 h=5 E
6 col=6 slice=0 h=6 F
4. Successfully learnt network
5. Successfully written learnt network
0 A
0.5005 0.4995
1 B
0.4925 0.5075
2 C
0.5275 0.4725
3 H
0.195313 0.804688 0.725738 0.274262 0.361217 0.638783 0.289796 0.710204 0.172932 0.827068 0.884956 0.115044 0.451852 0.548148 0.763713 0.236287
4 D
0.201708 0.798292 0.703669 0.296331
5 E
0.588047 0.411953 0.478833 0.521167
6 F
0.323372 0.676628 0.0903104 0.90969

Code: Select all

	em.SetEquivalentSampleSize(2);
	em.SetRandomizeParameters(false);

1. Successfully read network
2. Successfully read data set
3. Successfully calculated matches
0 col=0 slice=0 h=0 A
1 col=1 slice=0 h=1 B
2 col=2 slice=0 h=2 C
3 col=3 slice=0 h=3 H
4 col=4 slice=0 h=4 D
5 col=5 slice=0 h=5 E
6 col=6 slice=0 h=6 F
4. Successfully learnt network
5. Successfully written learnt network
0 A
0.500759 0.499241
1 B
0.492636 0.507364
2 C
0.527896 0.472104
3 H
0.195727 0.804273 0.724004 0.275996 0.362899 0.637101 0.294494 0.705506 0.171772 0.828228 0.884285 0.115715 0.449727 0.550273 0.763873 0.236127
4 D
0.20216 0.79784 0.70362 0.29638
5 E
0.588571 0.411429 0.479142 0.520858
6 F
0.324728 0.675272 0.0908674 0.909133

Thanks for the support!!!

Wed Jun 03, 2009 1:21 pm

Due to issues with our internal network I'm unable to connect to the machine which has VM with Darwin, so I can't directly verify your results. However, I checked that the numbers you reported on XP; they match those I'm getting on my Windows box with Visual C++ and cygwin. Output from the program running on FreeBSD is also the same.

While I'm waiting for the admins to complete their job, can you check the gcc version on your OSX computer with gcc --version ? Also, what's the command line you're using to build the program?

dtodor · Post by **dtodor** » Wed Jun 03, 2009 7:47 pm

I'm actually using XCode with gcc 4:

i686-apple-darwin9-g++-4.0.1 (GCC) 4.0.1 (Apple Inc. build 5493)

I've also tried building from the command line:

g++ -O3 -DNDEBUG -ffast-math main.cpp -o test -Lsmile -Ismile -lsmilearn -lsmile

I'm getting the exactly same results as when using XCode. And these are different from what I get under Windows.

Fri Jun 05, 2009 11:30 am

I have just confirmed your findings on Darwin running in VM (so it's certainly not an issue of OS/compiler version mismatch). We'll be trying to identify the reason for the output discrepancy and will post here as soon as any information is available.

Fri Jun 05, 2009 12:56 pm

At this point it looks like the issue with gcc's optimizer. The output is incorrect only if -O1/-O2/-O3 options are used to build SMILE/SMILearn. However, these options do not cause problems when gcc is used on other platforms (we don't have a system with gcc 4.0.x though).

dtodor · Post by **dtodor** » Sat Jun 06, 2009 1:06 pm

Could you provide a build without any optimizations?

Mon Jun 08, 2009 12:32 pm

dtodor wrote:Could you provide a build without any optimizations?

We'll do it as a last instance. Currently we're trying to determine where the codepaths diverge on different platforms.

Mon Jun 08, 2009 11:06 pm

dtodor wrote:Could you provide a build without any optimizations?

We've found the loop which is executed incorrectly when compiled with optimizations enabled on Darwin. It's quite easy to work around this gcc 4.0.0 bug and we'll post new build for OSX later this week compiled with -O3.

dtodor · Post by **dtodor** » Tue Jun 09, 2009 10:39 am

Thanks for the support! Would it be possible to make a x86_64 build as well?

Tue Jun 09, 2009 10:54 am

dtodor wrote:Thanks for the support! Would it be possible to make a x86_64 build as well?

I believe gcc 4.0.0 on Darwin 8.0.1 only supports i386 and ppc architectures. If you can point me to downloadable Darwin installer CD with gcc support for x86-64 I can try to set it up in a VM later this week.

Thu Jun 18, 2009 2:10 pm

SMILE binaries for OSX have been refreshed. This is a build with -O3.