sharpneat-developers Mailing List for SharpNEAT
Status: Beta
Brought to you by:
colgreen
Menu
▾
▴
sharpneat-developers — Discussion list for SharpNEAT developers, bug reports etc.
You can subscribe to this list here.
| 2005 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
(14) |
Dec
(3) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2007 |
Jan
|
Feb
(2) |
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
(1) |
Sep
|
Oct
|
Nov
|
Dec
|
|
From: Wael M. <le...@gm...> - 2007-08-14 20:19:40
|
hi there, i am trying to implement a neural network for a hardware platform, and i will be using integer weights, i was wondering which activation function should i use i am using a tanh but it gives a wrong attitude its not learning as i use integer weights, but when i use float weights it works fine. can anyone help ????. i will appreciate it Regards -- WILL MALEK |
|
From: Chris S. <chr...@gm...> - 2007-02-20 17:02:51
|
Just noticed there hasn't been any activity in this list for over a year. Is this project still active? Regards, Chris |
|
From: Chris S. <chr...@gm...> - 2007-02-20 16:44:32
|
Hi, I'm new to SharpNeat, and I'm trying to make sense of the current release, but I can't find any documentation, neither on the website nor the archive. There's not even any usage/readme/installation docs. Could someone please provide some guide on how to use this software? Thanks, Chris |
|
From: Colin G. <cg...@ds...> - 2005-12-05 20:23:45
|
Ricardo J. M=E9ndez wrote: >Hey Colin, > >Thanks for the note, I'll get the latest file and work from there. > =20 > Ok good. Everything seems to be progressing nicely :) Colin. |
|
From: <men...@gm...> - 2005-12-05 05:49:30
|
Hey Colin, Thanks for the note, I'll get the latest file and work from there. > A couple of other things I forgot to mention. There was a > WebConfigElementListHandler and a ConfigElementListHandler project in > your zip file but only the former was being used, is this name (prefixed > with 'Web') correct? And could the class in this project actually be > moved into your AppUtils project? Will do. The ConfigElementListHandler being there was a clean-up issue - the class was originally named WebConfigElementListHandler, due to where I was using it. I renamed it given that it's actually useful for reading any sort of .config files, but .Net started giving me some grief about being unable to load the class and I reverted all changes. The directory you got was a result of me not being done cleaning up. Cheers, Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ |
|
From: Colin G. <cg...@ds...> - 2005-12-04 17:51:40
|
Hi Again, A few points further to my last email. Actually I think the experiment title should be in the config, otherwise you have to load all experiments to build the combo! but the description and most other config data can go into the IExperiment. To alleviate the overhead of that this creates I have modified the existing experiment classes so that they don't instantiate an IPopulationEvaluator until one is requested (via the property). The fact that an evaluator was instantiated along with each experiment was a legacy from previous versions of the code. The file paths within the config file are currently full paths to the debug directories, which creates a problem if you switch the solution to 'release' mode. I'm not sure what the best solution is here but as a an initial solution I have made the paths relative (so they just specify the dll name) and added references to the experiment projects to to the SharpNeatGui project, this causes the experiment dll's to get copied into the debug dir alongside the main EXE thus ensuring the relative paths work. A couple of other things I forgot to mention. There was a WebConfigElementListHandler and a ConfigElementListHandler project in your zip file but only the former was being used, is this name (prefixed with 'Web') correct? And could the class in this project actually be moved into your AppUtils project? Anyway for now there is another updated zip file available on this URL: http://sharpneat.sourceforge.net/prerelease/ I'll start putting zips at that URL from now on rather than fill up your inbox! Regards, Colin. |
|
From: Colin G. <cg...@ds...> - 2005-11-28 23:16:51
|
Ricardo J. M=E9ndez wrote: >>Is this likely to cause you any problems Rico (+ new list member!)? >> =20 >> > >It probably would, given I'm working on this on an older version of >Visual Studio for a client's project. I might be able to maintain a >parallel project file while my client moves to the latest version, but >it would of course be an inconvenience. > >Do let me know if you're going to do any major changes to the project >structure - I was planning to work on the library loading re-structure >this week, and file conflicts would likely arise from us both changing >that area. > =20 > OK for now then I'm going to have tinker with VS 2005 and maybe move=20 some files/dirs around but I don't expect to make any major changes to=20 the code for a while anyway. Hopefully any work you submit I can merge=20 into the new directory structure and I'll then be able to make a zip=20 file available. Colin. |
|
From: <men...@gm...> - 2005-11-27 16:53:26
|
> Is this likely to cause you any problems Rico (+ new list member!)? It probably would, given I'm working on this on an older version of Visual Studio for a client's project. I might be able to maintain a parallel project file while my client moves to the latest version, but it would of course be an inconvenience. Do let me know if you're going to do any major changes to the project structure - I was planning to work on the library loading re-structure this week, and file conflicts would likely arise from us both changing that area. Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ |
|
From: Colin G. <cg...@ds...> - 2005-11-26 21:20:08
|
According to an email I received the other day SharpNEAT's existing visual studio project files can be loaded into VS2005 and transformed into VS 2005 format with no problems, and everything compiles and runs fine in .Net 2.0. As such I'll probably be moving to .Net 2.0 and VS 2005 soon. It's probably worth mentioning that the Visual Studio Express editions are currently free to download from here: http://msdn.microsoft.com/vstudio/express/default.aspx Is this likely to cause you any problems Rico (+ new list member!)? I might also take the opportunity to re-arrange where the project and solutions files are located. Cheers, C. |
|
From: Colin G. <cg...@ds...> - 2005-11-24 21:13:42
|
Ricardo J. M=E9ndez wrote: >On 11/22/05, Colin Green <cg...@ds...> wrote: > =20 > >>Yeh this is just wrong. Have made the change here. You probably also >>noticed I often don't call clear right at the start, which is normally >>Ok because the network was normally just built, but in the general case >>it's an assumption and that's bad. >> =20 >> > >I've noticed that in running an experiment, saving a population and >then re-loading it to continue the experiment, networks that were >previously optimal and had remained stable for a while tend to go out >of whack, but I've never really chased the issue. I wonder if this >lacking of an beginning initialization is what's throwing it off. > >Hmmm.... However, I'm always clearing the signals before even setting >the first input signal, so that's probably not it. > =20 > I've chased up this issue in the past but it's one of those things that=20 you have to keep an eye on as it seems to creep back in occasionally. On=20 one occasion it was due to the casting between doubles and floats. The=20 default network class - FloatFastConcurrentNetwork - uses floats, but=20 the genome class uses doubles and the file formats [should] store the=20 full precision of a double. Thus if you are saving/loading genomes and=20 populations then there shouldn't be a problem, but if you saving a=20 network you might lose some precision. It's something to keep in mind=20 anyway, some problem domains are more sensitive to small changes in=20 connection weights than others. Also note sometimes if you have been=20 running a search for a long time with only small improvements in fitness=20 then these improvements can be represented by small weight changes. > =20 > >>Maybe if networks set a flag when >>they are created (which resets on the first activation) we can check >>this to determine if a full clear is really required, then we can call >>clear at the top of all experiments, safe in the knowledge we won't be >>doing an unnecessary clear. >> =20 >> > >Certainly. Another option is having a network preference >"InitialAutoClear" , which defauls to True, which checks said flag on >call to "SetInputSignals" to see if an initial clearing is necessary. > =20 > Could do, but I prefer to leave it to the user to call clear if they=20 want a clear, it's simple and immediately obvious from the point of view=20 of a new coder. Also some experiments want clearing on, say, alternate=20 calls to SetInputs, in cases like that it's easier to just keep it=20 simple and trust that the coder of the experiment knows what they're=20 doing :) C. |
|
From: <men...@gm...> - 2005-11-24 00:31:29
|
On 11/22/05, Colin Green <cg...@ds...> wrote: > Yeh this is just wrong. Have made the change here. You probably also > noticed I often don't call clear right at the start, which is normally > Ok because the network was normally just built, but in the general case > it's an assumption and that's bad. I've noticed that in running an experiment, saving a population and then re-loading it to continue the experiment, networks that were previously optimal and had remained stable for a while tend to go out of whack, but I've never really chased the issue. I wonder if this lacking of an beginning initialization is what's throwing it off. Hmmm.... However, I'm always clearing the signals before even setting the first input signal, so that's probably not it. > Maybe if networks set a flag when > they are created (which resets on the first activation) we can check > this to determine if a full clear is really required, then we can call > clear at the top of all experiments, safe in the knowledge we won't be > doing an unnecessary clear. Certainly. Another option is having a network preference "InitialAutoClear" , which defauls to True, which checks said flag on call to "SetInputSignals" to see if an initial clearing is necessary. Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ |
|
From: <men...@gm...> - 2005-11-23 15:12:26
|
On 11/22/05, Colin Green <cg...@ds...> wrote: > Thanks Ricardo. When I talk about restructuring I'm referring to some > drastic changes I have planned to allow plugging in of different types > of network and genome but I have no concrete designs right now. > Basically I'm going to try and find a quiet corner over the holidays and > see what happens! One area I think you already mentioned was plug-in > experiments. This is almost possible already except I think an evaluator > is needed/instantiated right from the start to obtain the Input and > OuputNeuronCount, and right now this means all experiments are > instantiated in full at startup. + they should be external to the main > GUI exe so a recompile isn't required every time. I think both of these can be accomplished by the independent library / catalog I suggested. If your main focus is currently on the different types of networks and genomes, I'll start working on the experiment separation. > BTW have you created any interesting new experiments? Not yet, but I have a derivate of the Maximizer designed which I'll try to plug in later this week. It gets a list of numbers, and the odds that the number will increase in the next row. If it picks a number from an odd row, it's also picking the same column in the next row. He who ends up with the highest total after we run out of lines, wins. Something like a BlindGambler experiment, since it only knows a current payout and the chance said payout will increase. I'll drop you a note when I get it done, plus, clean up the code for the Maximizer a bit and send it over. Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ |
|
From: Colin G. <cg...@ds...> - 2005-11-22 23:22:09
|
Ricardo J. M=E9ndez wrote: >Hey Colin, > >What are your restructuring plans? I'll be glad to lend a hand. > > =20 > Thanks Ricardo. When I talk about restructuring I'm referring to some=20 drastic changes I have planned to allow plugging in of different types=20 of network and genome but I have no concrete designs right now.=20 Basically I'm going to try and find a quiet corner over the holidays and=20 see what happens! One area I think you already mentioned was plug-in=20 experiments. This is almost possible already except I think an evaluator=20 is needed/instantiated right from the start to obtain the Input and=20 OuputNeuronCount, and right now this means all experiments are=20 instantiated in full at startup. + they should be external to the main=20 GUI exe so a recompile isn't required every time. BTW have you created any interesting new experiments? C. |
|
From: Colin G. <cg...@ds...> - 2005-11-22 23:12:19
|
Ricardo J. M=E9ndez wrote: >No problem, hadn't posted yet. Two notes on the code. > >In SpeciesForm.cs, there's two lines around line 332 which read: > > int xIncrement =3D pnlMain.Width / colCount; > int yIncrement =3D pnlMain.Height / rowCount; > >Ocassionaly, colCount or rowCount can be 0, so the program ends with >an unhandled exception. > =20 > Actually I think I've experienced that, but never found where it was.=20 Thanks. >The second one I'm not sure if it's a bug or not, but on >XorNetworkEvaluator network.ClearSignals() is not called before >the test case for (1,1), while it is called for the next two. I >imagine this can lead to test problems with (1,1), given the >inter-neuron signals are not being checked. > =20 > Yeh this is just wrong. Have made the change here. You probably also=20 noticed I often don't call clear right at the start, which is normally=20 Ok because the network was normally just built, but in the general case=20 it's an assumption and that's bad. Maybe if networks set a flag when=20 they are created (which resets on the first activation) we can check=20 this to determine if a full clear is really required, then we can call=20 clear at the top of all experiments, safe in the knowledge we won't be=20 doing an unnecessary clear. Cheers, Colin. |
|
From: <men...@gm...> - 2005-11-22 21:51:51
|
Hey Colin, What are your restructuring plans? I'll be glad to lend a hand. Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ |
|
From: <men...@gm...> - 2005-11-20 03:08:27
|
No problem, hadn't posted yet. Two notes on the code. In SpeciesForm.cs, there's two lines around line 332 which read: =09=09=09int xIncrement =3D pnlMain.Width / colCount; =09=09=09int yIncrement =3D pnlMain.Height / rowCount; Ocassionaly, colCount or rowCount can be 0, so the program ends with an unhandled exception. The second one I'm not sure if it's a bug or not, but on XorNetworkEvaluator =09=09=09network.ClearSignals() is not called before the test case for (1,1), while it is called for the next two. I imagine this can lead to test problems with (1,1), given the inter-neuron signals are not being checked. Good luck, Ricardo J. M=E9ndez http://ricardo.strangevistas.net/ On 11/19/05, Colin Green <cg...@ds...> wrote: > Hi Ricardo, I see you found the list ok. Apologies for the test emails, > I hadn't realised that as well as being the list owner I also need to be > subscribed as a normal user. If you posted a message before this one > then I won't have seen it - and it won't be in the viewable archives for > a month - so if you did post something could you repost it for me. > > Thanks, > > Colin. > > > ------------------------------------------------------- > This SF.Net email is sponsored by the JBoss Inc. Get Certified Today > Register for a JBoss Training Course. Free Certification Exam > for All Training Attendees Through End of 2005. For more info visit: > http://ads.osdn.com/?ad_id=3D7628&alloc_id=3D16845&op=3Dclick > _______________________________________________ > Sharpneat-developers mailing list > Sha...@li... > https://lists.sourceforge.net/lists/listinfo/sharpneat-developers > -- |
|
From: Colin G. <cg...@ds...> - 2005-11-19 18:22:11
|
Hi Ricardo, I see you found the list ok. Apologies for the test emails, I hadn't realised that as well as being the list owner I also need to be subscribed as a normal user. If you posted a message before this one then I won't have seen it - and it won't be in the viewable archives for a month - so if you did post something could you repost it for me. Thanks, Colin. |
|
From: Colin G. <cg...@ds...> - 2005-11-19 18:14:27
|
|
From: Colin G. <cg...@ds...> - 2005-11-19 18:10:56
|
|
From: Colin G. <cg...@ds...> - 2005-11-19 18:09:49
|
