Re: Machine Learning in PicoLisp
How bad would a pure picolisp implementation be? El lunes, 20 de julio de 2015, Alexander Burger a...@software-lab.de escribió: On Mon, Jul 20, 2015 at 09:00:14AM +0200, klaus schilling wrote: Artificial neural networks, support vector machines, regression, smart classification/regression trees, and som on require heavy number crunching, such as inverting large matrices. Numpy (one of the modules used by python's most popular machine learning module) does this by accessing the standard Fortran libraries (BLAS, LAPACK). Would that be possible from picolisp? Yes. Better in pil64, with 'native' calls. -- UNSUBSCRIBE: mailto:picolisp@software-lab.de javascript:; ?subject=Unsubscribe
Re: Machine Learning in PicoLisp
I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. 2015-07-20 0:51 GMT-07:00 Alexander Burger a...@software-lab.de: On Mon, Jul 20, 2015 at 12:31:15AM -0700, Amaury Hernández Águila wrote: How bad would a pure picolisp implementation be? It depends how heavily the implementation depends on floating point. All right if it can be handled in fixpoint. If true floating point is needed, we must resort to 'native' calls again. On the other hand, the precision in a fixpoint implementation would be unlimited. -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
On Mon, Jul 20, 2015 at 12:31:15AM -0700, Amaury Hernández Águila wrote: How bad would a pure picolisp implementation be? It depends how heavily the implementation depends on floating point. All right if it can be handled in fixpoint. If true floating point is needed, we must resort to 'native' calls again. On the other hand, the precision in a fixpoint implementation would be unlimited. -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
Amaury Hernández Águila amhe...@gmail.com writes: Any suggestions to the name of the library? PicoML sounds good. i would suggest not using 'PicoML' - that sounds like a dialect of the ML programming language: https://en.wikipedia.org/wiki/ML_%28programming_language%29 Alexis. -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
PicoLearn On Mon, Jul 20, 2015 at 11:16 AM, Alexis flexibe...@gmail.com wrote: Amaury Hernández Águila amhe...@gmail.com writes: Any suggestions to the name of the library? PicoML sounds good. i would suggest not using 'PicoML' - that sounds like a dialect of the ML programming language: https://en.wikipedia.org/wiki/ML_%28programming_language%29 Alexis. -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe -- If technology is your thing plan to die reading manuals --Gene Woolsey -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
2015-07-20 2:08 GMT-07:00 Rowan Thorpe ro...@rowanthorpe.com: If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. Will do.
Re: Machine Learning in PicoLisp
I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment. I am currently trying to get PilOS running on Qemu on a Win 8.1 64bit machine. I'd love to have that and computational intelligence libraries working in 64bit PicoLisp! Hey, how about PicoCi or PicoCI? Rob On Mon, Jul 20, 2015 at 5:08 PM, Rowan Thorpe ro...@rowanthorpe.com wrote: On 2015/07/20-01:01, Amaury Hernández Águila wrote: I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. -- Rowan Thorpe PGP fingerprint: BB0A 0787 C0EE BDD8 7F97 3D30 49F2 13A5 265D CCBD There is a great difference between worry and concern. A worried person sees a problem, and a concerned person solves a problem. - Harold Stephens -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: PilOS
I tried several, and when I finally tried version 2.2.90 for windows 64 bit, It went past that error and went to a 'READ ERROR 09' Is that a 09h Attempt to DMA over 64kb boundary using the same Wiki reference you supplied in a previous posting? How can I fix it or keep moving on? Where do I look in the source to troubleshoot this myself? Exciting fun anyhow, so no complaints here! Rob On Mon, Jul 20, 2015 at 1:32 PM, Alexander Burger a...@software-lab.de wrote: Hi Robert, First, I would like to donate some money anyway, seeing how much fun I have had with PicoLisp and PilOS. Thanks, that's very nice! But don't worry, that's not what I'm looking for. I need some stable, long-term project(s) ;) Second, I get a 'Guest has not initialized the display (yet).' when trying to run PilOS in qemu on my i5 Windows 8.1 64bit machine. Any steps I am missing? I am a qemu newbie, and PilOS newbie. I did a short search on the web, and it seems this error appears in qemu in other situations too. Not only on Windows, but also on other guest operating systems. I have no idea what might be the reason. PilOS simply uses the standard VIDEO memory (VGA) on hardware address 0xB8000. Can you try another version of qemu? ♪♫ Alex -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
I appreciate people who know the term computational intelligence. PicoCI sounds good. I know that BLAS and LAPACK are battle-tested, but in that case I would just use other libraries in other programming languages (this is how I feel). I've been doing CI in common lisp using clml, mgl-gpr, mgl, and others, and I even have access to run my models in CUDA GPUs with my current setup. I'd like to see PilOS running CI in a near future, and without the dependencies on fortran's BLAS and LAPACK. I'm still open to constructive criticism. Should we take a purist approach or should we go the battle-tested safer route? 2015-07-20 2:32 GMT-07:00 Robert Herman rpjher...@gmail.com: I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment. I am currently trying to get PilOS running on Qemu on a Win 8.1 64bit machine. I'd love to have that and computational intelligence libraries working in 64bit PicoLisp! Hey, how about PicoCi or PicoCI? Rob On Mon, Jul 20, 2015 at 5:08 PM, Rowan Thorpe ro...@rowanthorpe.com wrote: On 2015/07/20-01:01, Amaury Hernández Águila wrote: I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. -- Rowan Thorpe PGP fingerprint: BB0A 0787 C0EE BDD8 7F97 3D30 49F2 13A5 265D CCBD There is a great difference between worry and concern. A worried person sees a problem, and a concerned person solves a problem. - Harold Stephens -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
Not using BLAS or LAPACK. 2015-07-20 3:53 GMT-07:00 Manuel Cano manutalc...@gmail.com: What gives you more fun? 2015-07-20 12:18 GMT+02:00 Amaury Hernández Águila amhe...@gmail.com: I appreciate people who know the term computational intelligence. PicoCI sounds good. I know that BLAS and LAPACK are battle-tested, but in that case I would just use other libraries in other programming languages (this is how I feel). I've been doing CI in common lisp using clml, mgl-gpr, mgl, and others, and I even have access to run my models in CUDA GPUs with my current setup. I'd like to see PilOS running CI in a near future, and without the dependencies on fortran's BLAS and LAPACK. I'm still open to constructive criticism. Should we take a purist approach or should we go the battle-tested safer route? 2015-07-20 2:32 GMT-07:00 Robert Herman rpjher...@gmail.com: I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment. I am currently trying to get PilOS running on Qemu on a Win 8.1 64bit machine. I'd love to have that and computational intelligence libraries working in 64bit PicoLisp! Hey, how about PicoCi or PicoCI? Rob On Mon, Jul 20, 2015 at 5:08 PM, Rowan Thorpe ro...@rowanthorpe.com wrote: On 2015/07/20-01:01, Amaury Hernández Águila wrote: I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. -- Rowan Thorpe PGP fingerprint: BB0A 0787 C0EE BDD8 7F97 3D30 49F2 13A5 265D CCBD There is a great difference between worry and concern. A worried person sees a problem, and a concerned person solves a problem. - Harold Stephens -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe -- Manuel
Re: Machine Learning in PicoLisp
If you can choose, I don't know what are you waiting for... have fun! 2015-07-20 13:03 GMT+02:00 Amaury Hernández Águila amherag@gmailcom: Not using BLAS or LAPACK. 2015-07-20 3:53 GMT-07:00 Manuel Cano manutalc...@gmail.com: What gives you more fun? 2015-07-20 12:18 GMT+02:00 Amaury Hernández Águila amhe...@gmail.com: I appreciate people who know the term computational intelligence. PicoCI sounds good. I know that BLAS and LAPACK are battle-tested, but in that case I would just use other libraries in other programming languages (this is how I feel). I've been doing CI in common lisp using clml, mgl-gpr, mgl, and others, and I even have access to run my models in CUDA GPUs with my current setup. I'd like to see PilOS running CI in a near future, and without the dependencies on fortran's BLAS and LAPACK. I'm still open to constructive criticism. Should we take a purist approach or should we go the battle-tested safer route? 2015-07-20 2:32 GMT-07:00 Robert Herman rpjher...@gmail.com: I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment. I am currently trying to get PilOS running on Qemu on a Win 8.1 64bit machine. I'd love to have that and computational intelligence libraries working in 64bit PicoLisp! Hey, how about PicoCi or PicoCI? Rob On Mon, Jul 20, 2015 at 5:08 PM, Rowan Thorpe ro...@rowanthorpe.com wrote: On 2015/07/20-01:01, Amaury Hernández Águila wrote: I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. -- Rowan Thorpe PGP fingerprint: BB0A 0787 C0EE BDD8 7F97 3D30 49F2 13A5 265D CCBD There is a great difference between worry and concern. A worried person sees a problem, and a concerned person solves a problem. - Harold Stephens -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe -- Manuel -- Manuel
Re: Machine Learning in PicoLisp
What gives you more fun? 2015-07-20 12:18 GMT+02:00 Amaury Hernández Águila amherag@gmailcom: I appreciate people who know the term computational intelligence. PicoCI sounds good. I know that BLAS and LAPACK are battle-tested, but in that case I would just use other libraries in other programming languages (this is how I feel). I've been doing CI in common lisp using clml, mgl-gpr, mgl, and others, and I even have access to run my models in CUDA GPUs with my current setup. I'd like to see PilOS running CI in a near future, and without the dependencies on fortran's BLAS and LAPACK. I'm still open to constructive criticism. Should we take a purist approach or should we go the battle-tested safer route? 2015-07-20 2:32 GMT-07:00 Robert Herman rpjher...@gmail.com: I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment. I am currently trying to get PilOS running on Qemu on a Win 8.1 64bit machine. I'd love to have that and computational intelligence libraries working in 64bit PicoLisp! Hey, how about PicoCi or PicoCI? Rob On Mon, Jul 20, 2015 at 5:08 PM, Rowan Thorpe ro...@rowanthorpe.com wrote: On 2015/07/20-01:01, Amaury Hernández Águila wrote: I think this will be an exciting project. I'll try a pure PicoLisp implementation and see how far I can go. Any suggestions to the name of the library? PicoML sounds good. Currently, I would start with a fuzzy logic toolbox, genetic programming and an architecture to create multi-agent systems. The second step would be to create neural networks. If you will develop on a public repo, please do send this thread a link to it when you feel it is at a point that others could send pull-requests to (or open issues for) to help with the progress. -- Rowan Thorpe PGP fingerprint: BB0A 0787 C0EE BDD8 7F97 3D30 49F2 13A5 265D CCBD There is a great difference between worry and concern. A worried person sees a problem, and a concerned person solves a problem. - Harold Stephens -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe -- Manuel
Re: PilOS
Hi Rob, I tried several, and when I finally tried version 2.2.90 for windows 64 bit, It went past that error and went to a 'READ ERROR 09' Is that a 09h Attempt to DMA over 64kb boundary using the same Wiki reference you supplied in a previous posting? Yes, indeed. And interesting: This seems to be the first time that error number 09 appears. How can I fix it or keep moving on? Where do I look in the source to troubleshoot this myself? Exciting fun anyhow, so no complaints here! The place where this happens is pilos/x86-64/beg.l, line 26 ff mov $DAP, %si # Disk Address Packet mov $0x42, %ah# Extended Read Sectors int $0x13 # Drive interrupt jc readError This is the standard BIOS call to read a given number of sectors from a drive. https://en.wikipedia.org/wiki/INT_13#INT_13h_AH.3D43h:_Extended_Write_Sectors_to_Drive This above error is correct, in that indeed more than 64 KiB are being read from the drive (the size of PilOS is 670 KiB currently). So the BIOS which is emulated in your version of Qemu somehow cannot handle it. Unfortunately, I don't know of any other way, except from writing your own BIOS. ♪♫ Alex -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: PilOS
Wow, thanks. It's trying to read in the entire thing 670kb of PicoLisp. I'll try other versions of qemu. There are a few between this one and the others that failed. Is there a way to change a qemu bios file config? Could I change beg.l to load it in segments, or would it still require a 670kb read to put it together again? Thanks. Rob On Mon, Jul 20, 2015 at 6:33 PM, Alexander Burger a...@software-lab.de wrote: Hi Rob, I tried several, and when I finally tried version 2.2.90 for windows 64 bit, It went past that error and went to a 'READ ERROR 09' Is that a 09h Attempt to DMA over 64kb boundary using the same Wiki reference you supplied in a previous posting? Yes, indeed. And interesting: This seems to be the first time that error number 09 appears. How can I fix it or keep moving on? Where do I look in the source to troubleshoot this myself? Exciting fun anyhow, so no complaints here! The place where this happens is pilos/x86-64/beg.l, line 26 ff mov $DAP, %si # Disk Address Packet mov $0x42, %ah# Extended Read Sectors int $0x13 # Drive interrupt jc readError This is the standard BIOS call to read a given number of sectors from a drive. https://en.wikipedia.org/wiki/INT_13#INT_13h_AH.3D43h:_Extended_Write_Sectors_to_Drive This above error is correct, in that indeed more than 64 KiB are being read from the drive (the size of PilOS is 670 KiB currently). So the BIOS which is emulated in your version of Qemu somehow cannot handle it. Unfortunately, I don't know of any other way, except from writing your own BIOS. ♪♫ Alex -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe
Re: Machine Learning in PicoLisp
Yes, go for it! Are you going to try and translate Fortran or C to PicoLisp, or are you going full hog, and try to implement from scratch? On Mon, Jul 20, 2015 at 7:17 PM, Manuel Cano manutalc...@gmail.com wrote: If you can choose, I don't know what are you waiting for... have fun! 2015-07-20 13:03 GMT+02:00 Amaury Hernández Águila amhe...@gmail.com: Not using BLAS or LAPACK. 2015-07-20 3:53 GMT-07:00 Manuel Cano manutalc...@gmail.com: What gives you more fun? 2015-07-20 12:18 GMT+02:00 Amaury Hernández Águila amhe...@gmail.com: I appreciate people who know the term computational intelligence. PicoCI sounds good. I know that BLAS and LAPACK are battle-tested, but in that case I would just use other libraries in other programming languages (this is how I feel). I've been doing CI in common lisp using clml, mgl-gpr, mgl, and others, and I even have access to run my models in CUDA GPUs with my current setup. I'd like to see PilOS running CI in a near future, and without the dependencies on fortran's BLAS and LAPACK. I'm still open to constructive criticism. Should we take a purist approach or should we go the battle-tested safer route? 2015-07-20 2:32 GMT-07:00 Robert Herman rpjher...@gmail.com: I would welcome the results of your efforts, and contribute where I could, but I think it would be best to make calls to BLAS and LAPACK, since they are battle-tested. I am currently working my way through a book 'Handbook of Neuroevolution through Erlang', but I prefer Lisp. Erlang is just better at the fault tolerance, distributed thing. Lush2 Lisp was used for heavy numerics, so you may want to look there for some guidance, however the Sourceforge site is down at the moment
Re: PilOS
On Mon, Jul 20, 2015 at 08:30:01PM +0800, Robert Herman wrote: Is there a way to change a qemu bios file config? The man page of qemu-system says QEMU uses the PC BIOS from the Seabios project and the Plex86/Bochs LGPL VGA BIOS. But there is also a -bios file option. I haven't tried. Could I change beg.l to load it in segments, or would it still require a 670kb read to put it together again? This is surely possible, but quite tedious. You would have to modify the 'DAP' structure (starting at line 87) for varying Start sectors, Offsets and smaller Number of sectors values, and call int 13h repeatedly. So you could read it all in e.g. 11 pieces, each = 64 KiB. Perhaps not worth the effort ... ♪♫ Alex -- UNSUBSCRIBE: mailto:picolisp@software-lab.de?subject=Unsubscribe