Numenta Live Streams, 19 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Got the SP initialized, now to compute something...

Description

Creating a C++ Spatial Pooler in Python 3 through pybind. -- Watch live at https://www.twitch.tv/rhyolight_

A

And I did end up reform because I can't stand okay, that's fine! So now that I have a special puller I mean this is where I wanted to get it's a sit right. This is it revoke special puller I mean I could run it, but I'm sort of confident that if, if I'm got this far with the bindings, it's probably gonna work but I bet there is a compute function right. Let's take a look, let's see if I can just fake it out.

A

Okay, so I have to initialize it. Where is this the same?.

A

Let me read through this a bit initialized special floor using given parameters. Oh that's, that's the same thing: okay yeah. This is the initializer.

A

This is a C thing, the compute, so we're gonna get asked to our input that should be of the input, dimension, right, learn and then something forget to put the output in right. So an SDR representing the winning columns after inhibition size of SDR is equal to the number of columns. So I should be able to call this compute function, pass it in two arrays.

A

Let me do this on the right track. You guys compute I'm, gonna, send in an input and I'm gonna, say, learn true and then output. Let's call this active columns.

A

Get pizza, yes, I was given two pizza boxes and a huge monitors end up recycling them yeah those very worthless, they're worthless. My biggest achievement on that ultra sixty was setting up Samba networking with my windows tower SMB Samba, whatever that was so hard that was so hard and all I was trying to do. Is you know, be able to copy it like move files from one to the other?

A

I just want to be able to see that disk space from my Windows machine, it's so what was so hard to do back then I felt like a computer scientist just being able to network these two computers together liked it anyway. So let's so, let's create an input and it's going to be a an array.

A

So a pythonic way of creating an array Oh input is a special, is a special keyword in Python. So I can't use that, let's put let's just call it input, data I, don't know input data and the input data is going to is going to be array.

A

Active columns is also going to be an array and I need to create an array of zeros or array of a certain size and then a Python 3 Python 3, create array of size stack overflow I've already been here before, so you can tell that so. Okay, I've done this 9 times 5. So I could do this. That's easy enough! So 0 times input, dimensions, input, dimensions actually 0 right, because the way we just pass it in is we can be multi-dimensional or not.

A

We're just we're doing a one dimensional non topological input right, see this for SDR use, SDR test, ok, I'll, come I'll, come back to that, because I want to see if I can get. If I can do this, my active columns, I'm gonna, fill in the zeros to and put column dimensions 0 now. This should work great.

A

Then I can say prints action problems. Let's see what happens. Oh sweets, wait a minute wait a minute, so I got okay. That's my oh! This is the debug right right right, so I. So it's calling spatial or invoked with that, and it's because I built a debug build. So this hasn't actually been doing anything so I'm calling it with zero input and it's ending up with zero active columns. Obviously, now what if I fill it with a full input?

A

Are there gonna be any active columns? It's gonna. Take a few runs right. So let's do this. Let's do four wait. Wait! Okay, Python, okay, I'm going to be doing this a lot because I'm trying to learn Python. So there's a bunch of things. I'm learning, Python C++ react: Twitter, twitch, Python 3. What was I doing? Oh yeah run run x times, something like that or pathetic wanna process X times for blank and blank and range. That's great! That's perfect!

A

That's what so, let's run this 50 times sure why not alright compute 50 times up up up up? No that didn't work. I just did one.

A

What happened, what did you wrong shouldn't this work shouldn't this run fifty times.

A

So that worked I'm confused now.

A

Oh, oh I had a trace back on this day. Compute data, okay, great incompatible, function, arguments following argument: types are supported.

A

Are one array are our boolean array so I'm sending an array and a boolean, an array of I, not I,.

A

Don't need this anymore yeah STR array, so, oh, it needs to be an SDR that needs to be an STR. That's what you're trying to tell me right! Mark okay, I can do that and you said see this for SDR use in my bindings PI test, SDR. Okay! Let's look at that. So this is a little different right because you guys got your own SDR class. Now you'll need PI R a for the arcs or said SDR I'll, try the STR STR! No, we want to look at the tests.

A

Bindings tests, STR STR tests- okay, so oh yeah I, like this I like this, so I'm just gonna create. So let me get this out. I'll create I'll. Get this I'll put that in my run, I already got it already got it all right and what I want to do is create an SDR with some dimensions.

A

They see some other ways, so I can or I can just send the dimensions in. So this is what I'm going to do like that and just say hundred.

A

So this is going to be input, dimensions, Oh, just input dimensions and then I'll make infant dimensions. I've been array or a tuple of work right. So I should just put infant dimensions here and the same thing here: Str with comm dimensions, maybe I, don't know, I will see I'm, not sure I'm, not certain. That's gonna work, okay,.

A

Not invoked with spatial puller SDR objects through a stereo object, so it still says it wants an array. Not an SDR object. Do I need to call like to something on it. Let me go look at the tests again test. Constructor, test zero test, dense test dense in place, test, sparse test coordinates just keep alive.

A

Just set, get get some get sparsity or NP dot array, yeah I guess I could do that. Does this have like a to numpy.

A

Where's, the so the Python code for this SDR class isn't. Is it in new pic CPP.

A

Intersection test.

A

Okay, I'm gonna I'm gonna go back since I'm, not familiar with the SDR I'm gonna go back and create an up your a I'll just do that from import numpy as P, so I think I can say this.

A

Right, this might be right, my god I thought three numpy array of size and D array, size I'm. That I don't know nothing that well either honestly, so I can do a bunch of zeros.

A

That's what I want zero.

A

Three fight- that's a so I only want that to be my dimensions. Okay,.

A

Yeah, okay, so if I do this and I use infant dimensions here, I would hope that would work. Do I really need a D type. I, don't know, I'll, try it with and without.

A

That's right with first so we've got an array of zeros here which PI code for STR STR is in C++. You look at pi bindings, okay, okay, okay, that's fine! That makes sense. I thought that it was a Python class.

A

Okay, you expect you meant okay, so I think I can just change this right to unsigned in 32.

A

Aha success! All right! So that's weird though it did not did it run.

A

Did it run hello? Did it run? Yes, but it's not okay, but it just didn't print out the computer, which is kind of weird I thought it wouldn't so I'm, basically sending the same input data over and over and over. Oh, it's all zeros of course. So let me just randomly set some bits.

A

So let's do 10 times 4 I, guess: I can't use the same thing so X and range 10 times I'm going to let's okay, so the input I'm going to set every every time, I'm going to have new input and the active columns is also always going to be new zeros right, I. Think yeah I shouldn't do this every time and then compute. So this is just going to be for like input, data at some random index equals one and the random index. Okay. So now I gotta go Python.

A

Three random number, Rand int I want let's, let's use it.