Add to Watchlist

Comportex Notebook [DEMO #7]


Citation of segment
Embed Code
Purchasing a DVD Cite video
on the other end and and this is is
something that you have yet I guess still so it's work on combining and putting it together into an output format of the goal is to put the code closer to the visualization of it so that the execution of it being aware of what's going on is right next to the couch and I think that this is a good way of bringing them together I really like notebooks of this format of the to like and in case anyone doesn't know like it's sorta like of IPython notebook you can evaluate this but it's in the document forms on other kinds of just real quick the comport axes of HTM implemented and closure but my feel and can protect this similarly he made a bunch of really cool specializations which I still profusely so and is going to go ahead and jump and so that's 1 gets constraint on the you see like this is basically command command line output you see on this model the model just like this is a stack up here it's creating a model and
rather than evaluating into some big data structure a variable name you can see it and so you can you can like toy with that you can change the speckle it'll be like if I want to have like 8 100 columns or you can do multiple let you choose the cells per column as well right it's to lower than that and you can do a bunch of regions in series
and and so on and so on
can and in now and in some ways this is really boring no because I was working Texas is really cool is when it brings time into the picture so I created sequence which is
the simulation and ahead visualize that so when you see across like left to right you see time this is completely feel it's like that this is like I I stole this from comport explains nite inserted into a notebook so and this of the simulation is actually happening on the server and then it's talking to a browser which is so tossing these models across the wire like that the data and has being visualized here and all although John this is just like a toy example which so is this but had I decided that pattern data through and and this is all today's starting fresh with like a different implementational and I knew compartments already but starting college and today I I didn't it works but I had to know a lot more if I had a bunch more time but you can sort of see the investigative process of how you would start to tuning so here's the model of start with what's going on when a few things helper functions will see the
beginning of the model on all of these red because this is the beginning of the bill start to turn blue and purple was the predicted and so is the beginning there's no predictive model yet another thing that that's going to be useful to
know later as these inputs the way encoders are often like this we can sort of read them like reading the matrix of this 1 is the time the hour of the day this 1 is whether it's the weekend or not and this is the the consumption and i should show those numbers for people who haven't seen I'll show a lot of so the corresponding numbers to that were in the red line is the actual data blue dots are predictions of predict because there's no model for it hasn't forms of distal synapses of jump ahead so that the reason is taking time discuss colleges Khanakul right now there's just sequence the set up like as you read from it it evaluates the and it evaluates the simulation so you could actually have like an infinite sequential which represents a simulation and this is a fun way to work with data you set so this is 500 time later you see it's starting to predict obstruct 2000 and similarly as much as 500 types the summertime
to take and and so is predicting the lattice and these that's like I see this and I am not a super experience but I see it it's like a lot of the predictions are after the fact is kind of just saying what was before so it's not a it's it's not that good on so I can would like to drill a little bit and see what the columns are right now and see if I could glean maybe a little bit of insight so we would be a good yeah the 1st century at this point in Figure was certainly 1 thousand before and you like short time steps works and you can jump up and down the and we were looking at the early ones on 1 thing that you're seeing here that you the different visualisations this this this this was a very quick to generate like you can do more in depth 1 of well I creatively called the steps and steps plus plus 1 over 4 so the reason I brought up the encoders is when you're starting to look at the columns and wondering why they're Wu Y which ones are predicted or it's not very useful to look at the ones that are connected to the upper encoders only and because it's only consumption that we care about the school and we want to predict correctly so I like and I think I saw toward the bottom and it's sort of on it's it's sort of topographical and so you could you know to look at those and you certainly looking that looking around seeing which ones are bursting the red ones reversing and so you see this 1 was somewhat predicted that thing but not totally on Caesar getting an idea of is this thing predicted amassing a lot of blue blue means that was predicted but not active residents active but not predicted and it just gives you and you you kind of learn how it all works a lot better just using this but I think a guides what further questions you ask and I think that's like a really useful thing and this is about as far as I got on and and hot and so what is going on in the leave it there but I think you can see how that like I just how it would guide the next questions so another future we will when you have of the from
just so that you see right now it's running on local host which means it's useless to most of the world right now and I will drop the stress on a on a file share basically onto a static website for it's going and so what I'm doing now is I'm posting this to the web and it's going to be in a different form where it's a little bit of work less powerful in that you can execute new code but has
30 seconds had that are gonna talk a little bit so it's it was pastoralists of so basically what I've just done is something I was writing code but I was also writing a blog post or an essay or you could see paras slides slides for presentations in fact once it's there and 10 seconds why is it so it's a little bit they yeah there's that a lot of the work was making it were these were enormous you learn all of a hundred ways to hang a browser when you don't think you pick models and to extent models into so that should be there downloading Macintosh hopefully that still and then I will discuss a little bit more about and like I said it with is running on on the H T M in this case is running on a server and visualisations in the browser which is new for is that is basically the model you use with new kind of wisdom cortex and visualizations of it because it conveniently can be compiled on the closest of all the simulations of this point ran in the browser and which is like often good but there are often reasons you would wanna do it this way as well and that was a big part of the work here was making that possible so that load and as you can see this is no longer like this text input area but the other the way it the school what this will do that and is still interactive seat pursuant to this and like that you on the trade in front of you and I think it's so it opens up new scenarios it's good to get away from local host cell and like this and that was that was the trivia little from compartments there's so as most all of that I think I've seen is often really good at making you notice questions you should have been asking are the did not ask and I know what's a really good dev environment like what you just saw was a of environment that was like I wrote code in that are constantly this this weekend and with these powers combined I want 1 way that this is I would say more convenient can cortex this is in we we were both struck by this is like we're confuse that something was going on there and they were like we were in coding place right now we can probe in right now and like look at it it was a mind shift so as to bring the 2 together like closer I think makes cool stuff happens so that's like at which the questions and to him
really can't notebooks and now I just use the technology was behind like using other source technologies used to call on you it would itself comport Barillo rebel which is basically someone so I kind of like the empirical problem was this is simply because I use this to make him except pull requests but has a but anyway and the that is getting process so but yes cultural rebel it's is the similar 1 foreclosure we find the we we


Formal Metadata

Title Comportex Notebook [DEMO #7]
Title of Series 2015 Spring NuPIC Hackathon
Number of Parts 19
Author Lewis, Marcus
License CC Attribution 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
DOI 10.5446/18051
Publisher Numenta Platform for Intelligent Computing (NuPIC)
Release Date 2015
Language English

Content Metadata

Subject Area Information technology
Abstract Marcus Lewis shows off a tool much like the iPython Notebook, but for Comportex.

Related Material



  320 ms - page object


AV-Portal 3.7.0 (943df4b4639bec127ddc6b93adb0c7d8d995f77c)