Coursera Study Group

datageist

06:36

I was signed up for the PGM class, but the MITx circuits one is taking up all my ambient time at the moment.

I went through the first version of Machine Learning though. Pretty good.

Ivo Flipse

06:58

@datageist Hows the MIT one?

datageist

Much more time consuming than the coursera stuff was, mainly because they don't pull any punches, technically speaking. However, the coursera stuff felt MUCH more coherent.

They've got one professor doing the lectures, and another doing the assignments, and the content doesn't seem to connect 100%.

Ivo Flipse

Its like their opencourseware stuff then?

datageist

It's supposed to be like a "real" MIT class, so it's more immersive. I actually enjoyed watching the OCW circuits lectures more than I'm enjoying this class, if that makes sense.

Ivo Flipse

The problem there was always the lack of assignments (or rather getting solutions to them) or interactivity and the fact that they require MIT level pre-requisites which obviously not everybody has

datageist

The lectures felt coherent.

Ivo Flipse

07:08

I have that feeling with Udacity

A lot of the stuff from earlier classes gets used for solutions in later ones

datageist

Right, that's the main difference--prerequisites are assumed. Diff eq's, some electronics background, etc.

Ivo Flipse

the problem is that there isn't OCW content for those prereqs :P

datageist

The interesting thing about MITx vs. Coursera is that the infrastructure is more comprehensive.

They have a full circuit simulator in the browser, for example.

Ivo Flipse

or at least not in a way that if you have no real prior knowledge, that you could learn enough from it just by watching video's khan academy style

@datageist Coursera and Udacity are rather 'hurried', I mean the code editor for javascript on Codecademy is a lot more user friendly for example

They just have a smash hit on their hands and needed to get this content out there ASAP

datageist

Interesting. I haven't tried anything from Udacity yet.

Ivo Flipse

07:10

while MITx probably had this is the works for much longer

well it will depend on your background if its any use

datageist

Yeah, they've really got some manpower behind it, and some really good programmers.

The infrastructure is more coherent/mature than the content.

Ivo Flipse

Such is life

datageist

I think future verisons of the course will be better--this is kind of the prototype.

Yep.

Ivo Flipse

I think after the next wave of courses though, I'll have to 'take it easy' and get some work done too

even though they keep adding more and more

datageist

Yeah, this MIT one is wearing me out. I was signed up for 3 coursera courses when this one started, and had to drop out of all of them :)

The nice thing about Coursera is that all the content was contained in the lectures.

Ivo Flipse

07:13

@datageist I wasted like 1.5 day on NLP, just because I wanted to get a good score

and because I didn't use version control until half way through, so I 'lost' the code for my best effort :P

datageist

Yeah, even when I took the Machine Learning one, I started it almost 2 months late, so even though I technically got a perfect score, they wouldn't grade it that way.

Ivo Flipse

today I'm going to work on PGM

@datageist that sucks

datageist

Which just seems capricious and arbitrary. Online courses should be asynchronous.

Ivo Flipse

I thought ML was very rewarding though, because they allowed you to keep trying till you got a perfect score

Definitely, they claim you need deadlines to keep the pressure on, which I think is bollocks

datageist

There's too much of an effort to re-create a "classroom" experience, and it doesn't transfer.

Ivo Flipse

07:16

you hear people say: if you didn't need deadlines, people would have used OCW

datageist

Well, deadlines are OK if they're relative rather than absolute.

Ivo Flipse

well the current content is a lot more user friendly for new people than OCW, so its not a fair comparison

I think they main reason they use deadlines is for test results

datageist

Deadlines provide a minor motivation, but good content is better.

Ivo Flipse

Which they probably do because they want to sell your info, which they can only do if your grade is at least worth something

deadlines make cheating a little bit harder

though thanks to github, I can cheat a lot of times

datageist

Ha :)

Ivo Flipse

07:18

I'm here to learn, if they prevent me from doing that by giving some cockblock exercise, I'll try to get around it :P

datageist

Yeah, I've got mixed feelings about these online courses compared to how I learn on my own.

Ivo Flipse

which is also why I always hate that books never have solutions

great way to mess with anyone not in a class room

datageist

Yeah, I like to verify most of my knowledge with programming (whenever possible), which I find makes it stick more than exercises. Unless the exercises happen to be really well designed...

The main problem with books (in relation to self-study) is that they are too formal.

In other words, there's lot of ambient wisdom that gets conveyed during lectures that most people would never put into a book.

Ivo Flipse

That's for sure

datageist

And that's the real value of the all OCW stuff (for me at least).

Ivo Flipse

07:24

In class they sometimes take the time to give a for dummies explanation, which nobody in their right mind would write down

and indeed, a lot of books try to be 'exact' especially when it comes down to the math

datageist

Yeah, exactly. Or say controversial things like "nobody ever uses this".

Ivo Flipse

so they use a lot of formula's and infinity signs, which only make it harder to understand

first give me a basic example, then bore me with the details

datageist

Well, learning just about any difficult subject tends to require finding at least 3 really good books from different authors.

In the absence of good lectures, that is.

That way one author will hopefully fill in the gaps left by the others.

Another trick I like is finding graduate classes with a module that reviews some prerequisite. Those "reviews" often teach the subject better than an entire class.

How is the PGM class so far?

Ivo Flipse

Tough

I lack almost every prereq :P

the book is helpful, but tough as well, plus I follow so many courses, that its hard to catch up with all the reading

datageist

Yeah, I was really looking forward to that one. Maybe next time around. Like what are the pre-reqs for it?

Ivo Flipse

07:35

she probably covers half or more of her entire book :\

well I've never worked with probabilities before, apart from AI-class

so that and probably some Math related to that, like set-theory and such would help

I've had plenty of Math, but never the CS kind

But obviously, its very much worth it, because what you can do with those models is awesome

so I'm just doubling down on it

datageist

Interesting. The one I really want them to put in coursera format is Linear Dynamical Systems.

Ivo Flipse

that's a linear algebra math course right?

But I agree, more of those courses would help make the rest more accessible

datageist

It's like linear algebra combined with differential equations.

I learned more heavy linear algebra from watching a few of those lectures than any of the MIT OCW stuff.

The Stanford Fourier Analysis lectures are amazing as well.

Ivo Flipse

I'd also like to brush up my statistics

though most courses stop at the basic stuff, while all the more interesting stuff is left out

Though I reckon even my basics could use a refresh :P

Ivo Flipse

14:48

@Phonon strange, for some reason my output for ComputeAllSimilarityFactors keeps getting rejected :S

scratch that, I was using the wrong index for j

Mohammad

17:14

@Phonon @IvoFlipse Afternoon all!

Ivo Flipse

Hello :)

Just finished Week 3 of PGM

Mohammad

@datageist Ah, totally true.

Ivo Flipse

now wrapping up the quiz for NLP

and still need to finish that Strongly Connected Components Algorithm for Algo-class, but it keeps crashing on me :S

Mohammad

hehe

@IvoFlipse I forget, your background is DSP or physics?

Ivo Flipse

@Mohammad Human Movement Science :P

fbw.vu.nl/en/programmes/…

Mohammad

17:18

@IvoFlipse Oh wow cool! :)

WHere DSP meets the Gym :P

Ivo Flipse

Cool but useless

Mohammad

@IvoFlipse Nah, I think its rather useful actually, what with all the kinectic and wii centric things coming out

Haptic devices

@IvoFlipse Lots of overlap with robotics I imagine as well

Ivo Flipse

@Mohammad Indeed, but that's like 0.01% of all the available jobs :P

Not at all, though I think we'd do well in a robotics group

Mohammad

@IvoFlipse Well if I ever make an android I will let you know :P

Ivo Flipse

Thanks

Phonon

17:54

@Mohammad Hello

Mohammad

@Phonon Hey dude

Work from home today woohoo! :D

Phonon

@Mohammad Nice

Mohammad

@Phonon Im implementing a TDE using spectral flux today btw.

Phonon

@Mohammad Yeah, I've heard of spectral flux before

Mohammad

@Phonon Seems to be the best performer in a paper/study I saw

@Phonon Well, best in terms of ease of implementation and complexity I shoudl say

Phonon

18:01

@Mohammad Interesting

@Mohammad I actually never came across anything like that

Mohammad

@Phonon Check this one: (dafx.ca/proceedings/papers/p_133.pdf)

@Phonon What do you mean?

Phonon

@Mohammad Using spectral flow for TDOA

Mohammad

@Phonon Oh, yeah I found it under the music community. They pretty much try to detect 'onset' of a signal.

@Phonon Given the high non-linearities non-stationarities it might be useful

Phonon

@Mohammad I see.

@Mohammad So it works in cases where you're sending pulses, but not continuous signals.

@Mohammad With noise and distortions, of course

Mohammad

@Phonon By continous you mean smoothly rising waves?

Phonon

18:07

@Mohammad Sorry. I mean signals without any particular onsets per se. Like if you're just constantly sending white noise.

Mohammad

@Phonon The violin in the paper is an example of a smooth continuous waveform whos 'onset' seems to be hard to pinpoint even with the eyeball, but seems to capture its onsets somewhat ok.

Phonon

@Mohammad I see.

Mohammad

@Phonon Even if you were sending white noise and there was a jump somewhere, I think this would still capture it (spectral flux) because we are looking at difference in spectrum, and that would include amplitude within it as well.

@Phonon Kinda why I want to implement it and take it for a test drive hehe

@IvoFlipse @Phonon Hey what programming languages do yall know btw?

Ivo Flipse

@Mohammad Python, Matlab and C++ (though don't have much experience with that) and I've dabbled some with Ruby for SaaS-class and Javascript for Codecademy

Mohammad

@IvoFlipse I see cool. How good are you in Python vs ML ?

Ivo Flipse

18:21

My Python skills pwn my Matlab skills, if only because I feel like it can do more

Mohammad

@IvoFlipse Haha really?! Hmm. Why do you say that?

Ivo Flipse

@Mohammad Did you every try making a GUI in Matlab?

Mohammad

@IvoFlipse I have yes, but have nothing to compare it to to know if it sucks

@IvoFlipse So I have python, dabbled in it, but I want some IDE or interface where it feels like ML, in the sense of being able to evaluate lines on the fly. What is the best for that?

Ivo Flipse

plus I just don't look at Matlab as a tool to built user oriented software, its more a research tool

@Mohammad Python(x,y)

or Sage if you're on Linux

Mohammad

@IvoFlipse Hmmm (x,y)...very interesting... and the interface is like "QT"? So it 'feels' like matlab then?

Ivo Flipse

18:27

@Mohammad you aren't really dependent on Qt, but yes it feels like Matlab

you get an interpreter, you get to see what variables you have and such

but its Python :)

which is awesome

Mohammad

@IvoFlipse That is very sweet...Ill have to really check it out... the current python I have set up runs on my eclipse IDE and feels to C++ish

too*

@IvoFlipse What interpreter you use for that matlab feel?

Ivo Flipse

I use Eclipse too though, because I feel I have more control over what get's imported and what not

I don't use that Matlab feel :P

though really, it doesn't matter to me

the only 'quirks' are like when you try to plot

and it keeps waiting until you press Enter or something

which it might not do in iPhython

Mohammad

@IvoFlipse Ya, but what is an example of an interpreter in this context ?

Ivo Flipse

I haven't been messing with data lately, so I didn't need the interactivity as much

Mohammad

I see

Ivo Flipse