> More seriously, I think rapid experimentation will also help with the classification vs. anomaly detection conundrum as discussed during the last office hours. If you can actually see representations in each hierarchy, you can also understand how classifications are built automagically in an unsupervised fashion and how persistent they are.
> David, I think that good visualizations are one of the most important tools for this brand of AI and this is a great first step. I am especially interested in NLP because I'm trying to solve some problems in the area of social media business intelligence. The Bayesian and linguistic approaches have hit their ceilings and I'm confident that the neocortical approach will solve many of the problems that plague modern NLP. But to get sophisticated I think we're going to need to be able to model not just words, but grammar which will probably require hierarchichies, The type of tool you've developed will be very helpful in this area. Hi Mika and Alexander, Thank you again, guys! Certainly my wish and of many here it's that NuStudio attend the the largest possible number of scenarios involving HTM. So I look forward for your feedback on any adjustments that you think that are needed! All: I finished 3 chapters of the tutorial right now: "Introduction", "Overview" (explanation of how NuStudio is organized and its main features) and "Simulation" (an interative explanation where you will understand 3D elements, their colors, statistics,etc). I'll try finish the last chapter (about how create a project from scratch) this weekend. Fergal and David Ray kindly volunteered to revise the English text and with markdown. So please be patient until we conclude it! In the meantime I would like ask to those that had a successful installation to list the package versions installed on your machines. This way, I'll put into pip requirements only those versions that are presenting better results for most users. This said, to get the version of all required packages, just type: python setup.py develop or python setup.py install This script will show which packages already are installed and their respective versions. Only in the case of PyQt4 that you should use: >>> from PyQt4.QtCore import QT_VERSION_STR >>> print QT_VERSION_STR My package list is: numpy==1.8.1 scipy==0.14.0 PyOpenGL==3.1.0 PyOpenGL-accelerate==3.1.0 PyQt4==4.8.6 pyqtgraph==0.9.8 Erik Blias's list: numpy==1.9.0 scipy==0.14.0 PyOpenGL==3.1.0 PyOpenGL-accelerate==3.1.0 pyqtgraph==0.9.8 Your turn... On 11 September 2014 00:49, Mika Schiller <[email protected]> wrote: > David, I think that good visualizations are one of the most important > tools for this brand of AI and this is a great first step. I am especially > interested in NLP because I'm trying to solve some problems in the area of > social media business intelligence. The Bayesian and linguistic approaches > have hit their ceilings and I'm confident that the neocortical approach > will solve many of the problems that plague modern NLP. But to get > sophisticated I think we're going to need to be able to model not just > words, but grammar which will probably require hierarchichies, The type of > tool you've developed will be very helpful in this area. > > Mika > > On Wed, Sep 10, 2014 at 9:20 PM, David Ragazzi <[email protected]> > wrote: > >> Hi guys, >> >> Thank you for all your supportive messages, this is very motivating!! >> >> >>Does the visualization come from one of the hackathons? I can’t recall >> who did that back then. Maybe you already. Since then I hoped something >> like this will come along. >> For the encoder thing. Would it be possible to include a python code >> window where you could write encoder functions on the fly? I think rapid >> feedback has a lot of merit when prototyping encoders. >> >> Alexander: Actually not, this visualization came from my mind. :-p Of >> course.. some things like statistics I migrated from OpenHTM project which >> was a part nicely written by my friend Uwe. But the nodes selector/modifier >> (UI tree) was originally written by me in C#: I had this idea while OpenHTM >> was active and then decided migrate it to NuStudio because it is really >> multi-funtional and very intuitive to use. Although OpenHTM had a 3D >> simulator, it was only for .NET environment and also was very tricksy to >> model simple things, even a sphere or cube! So the 3D simulator I had write >> letter by letter because I wanted use components that were easy to consume >> and code (in this case, I've chosen PyQtGraph). Futhermore, I wanted a 3D >> color scheme that required little effort to someone follow the elements >> states, so I choose: gray for inactive elements, yellow for >> active/connected (because it is a color really appeling), and green for >> predicted elements (because green means a natural transition to yellow), by >> the way I'm open to suggestions. About encoder, the initial plan is give >> freedom to the researcher create her own encoders, for this purpose >> NuStudio already provide a super class called "Encoder" with two functions >> to be overrided: encodeToArray(rawData) and decodeFromArray(arrayData) >> which keeps backward compatibility with NuPIC encoders (I'll explain this >> in details in the tutorial). >> >> >>I'd really like to attach this to the Hot Gym example. It would be a >> good tutorial to explain how to take the One Hot Gym Prediction data and >> get it into NuStudio. This will be a goal of mine, and I'll try to get you >> feedback and help improve the tutorials you create as I go. >> >> Matt: I agree absolutely with you! We could replicate the Hot Gym example >> on NuStudio, this way anyone could read Hot Gym example on NuPIC wiki and >> at same time see it on NuStudio alive! About you help me with tutorials, >> needless to say that you're my GOD on that and other subjects like Travis, >> Git and others! >> >> All: As I suspected, some installation issues arose, the next step is >> check which package versions work well together, and then update PIP >> package. This said, still it is expected some manual installation to get >> the things workings. But relax, "apt", "brew" and other package manages are >> playing well their roles. >> >> Next week I'll post some draft tutorial to help people create a HTM from >> scratch quickly. By the way you already play with the existing project >> ("squares") on this weekend and report bugs and sugestions of improvement. >> >> Cheers! >> >> On 10 September 2014 20:44, <[email protected]> wrote: >> >>> This is awesome work dude! Super Inspiring!! >>> >>> Sent from my iPhone >>> >>> On Sep 9, 2014, at 7:39 PM, David Ragazzi <[email protected]> >>> wrote: >>> >>> Hello NuPICers! >>> >>> Just for inform that in all these days I was out, I was creating an a >>> tool that I believe will be a breakthrough for the community. This tool >>> calls "NuPIC Studio", or just NuStudio. NuStudio is an all-in-one >>> tool that allows users create a HTM neural network from scratch, train it, >>> collect statistics, and share it among the members of the community. It is >>> not just a visualization tool but an HTM builder, debugger and laboratory >>> for experiments. It is ideal for newbies with little intimacy with NuPIC >>> code as well as experts that wish a better productivity. Among its features >>> and advantages I list: >>> - Users can open, save, or change their "HTM projects" or of other >>> developers. A typical project contains data to be trained, neural network >>> configuration, statistics, etc, which can be shared to be analysed or >>> integrated with other projects. >>> - The HTM engine is the own original NuPIC libray (Python distribution). >>> This means no port, no bindings, no re-implementation, etc. So any changes >>> in the original nupic source can be immediatedly viewed. This helps users >>> that wish test improvements like new encoders or even hierarchy, attention, >>> and motor integration. >>> - The project is pip-installable and for desktop use. >>> >>> Project details: >>> - All code is implemented in Python. When I say "all code", I really >>> mean that code like GUI, 3d simulation, etc, are writen in Python and use >>> open-source and cross platform components. >>> - All GUI stuff (forms, controls, etc) is implemented in PyQT. PyQT is >>> the version of the famous QT adapted for python apps, which is open-source, >>> cross-platform, complete, and very easy to code and change. >>> - All utilities like XML serialization, canvas 2d, etc, also are >>> implemented in PyQT. >>> - A front-end object-oriented layer also written in Python is provided >>> in order to ease the understanding of the NuPIC code and keep backwards >>> compatibility. This means that someone with less experience in NuPIC could >>> re-use this front-end in her projects. >>> >>> The project's site is: >>> https://github.com/DavidRagazzi/nupic.studio >>> >>> and you can see some screen-shots attached to this message. >>> >>> Surely it still is not 100%, and it's expected that some installation >>> issues arise, but I hope optimize the process as soon as possible. By the >>> way, it works reasonably well and any people is welcome to try it or even >>> improve it. >>> >>> I look forward for your feedback, it is very important! >>> >>> Cheers, >>> David >>> >>> -- >>> David Ragazzi >>> MSc in Sofware Engineer (University of Liverpool) >>> Try NuPIC Studio! https://github.com/DavidRagazzi/nupic.studio >>> -- >>> "I think James Connolly, the Irish revolutionary, is right when he says that >>> the only prophets are those who make their future. So we're not >>> anticipating, we're working for it." >>> >>> <splash.png> >>> >>> <screenshot.png> >>> >>> >> >> >> -- >> David Ragazzi >> MSc in Sofware Engineer (University of Liverpool) >> OS Community Commiter at Numenta.org >> -- >> "I think James Connolly, the Irish revolutionary, is right when he says that >> the only prophets are those who make their future. So we're not >> anticipating, we're working for it." >> > > -- David Ragazzi MSc in Sofware Engineer (University of Liverpool) OS Community Commiter at Numenta.org -- "I think James Connolly, the Irish revolutionary, is right when he says that the only prophets are those who make their future. So we're not anticipating , we're working for it."
