Coursera Study Group

« first day (308 days earlier) ← previous day next day → last day (652 days later) »

Ivo Flipse

15:39

@Phonon got 26/40 for the final exam

2 hours later…

Phonon

17:26

@IvoFlipse Oh wow dude! are you still getting the certificate?

@IvoFlipse Now that I think about it it's not that bad

I read 26/50 first

Haven't taken mine yet

Will take it this weekend

4 hours later…

Ivo Flipse

21:53

@Phonon I dunno, messed up one of the quizzes too

But at least I passed all the PAs

But I never liked the quizzes for this course very much anyway, so I'm not to worried

I learned some dynamic programming, that's good enough for me :)

Phonon

22:14

@IvoFlipse Agreed. We'll see how I do.

Ivo Flipse

You should be fine

Phonon

@IvoFlipse May not be much better lol

Ivo Flipse

In most cases I just made the wrong choice, because while there are 4 options, its mostly one of each pair

In case of algorithm X this is or isn't NP complete

Phonon

right

I see

Ivo Flipse

I'm now watching video's for Computational Methods for Data Analyses

The Matlab stuff can be interesting

But the Math is too complicated, the code is too simple :P

Phonon

22:18

I like that = )

Should definitely take that class at some point

Ivo Flipse

I don't, I have no clue what some of the Math he talks about is means

then I see the code and am like: is that all?

I did like how he simply makes his own Gaussian to filter the fft domain

I never realized it was basically just something like this: F = exp(-0.0001 * (KX-401).^2 - 0.0001 * (KY-301).^2);

Note that KX and KY are the x's and y's of a meshgrid

I also still have to watch the Data Analysis course video's

I did the quiz at the last moment yesterday

Though while the format of the assignments is interesting, I don't feel like doing the work for the deadlines

Phonon

22:38

wow nice

I'm totally going to take it

Ivo Flipse

Yeah, I never realized how the magic behind the scenes really worked

the Image processing class covered some interesting stuff too

Like how you can easily calculate the mean by using a Gaussian or something like that

Can't recall what the topic was exactly

But sadly very low on applying the stuff in practice

Kortuk

23:02

@IvoFlipse I agree with that. Often the math is too complicated and matlab makes things so simple you sometimes have errors from it attempting to guess what you need, in my opinion.

Ivo Flipse

My biggest problem is that the type of calculus I learned isn't the type of calculus used for this stuff

Kortuk

@IvoFlipse :(

Ivo Flipse

So he writes some kind of formula and then goes like: and now we take the derivative. And what he writes down just looks like gibberish to me

Kortuk

I was wanting to take that class, but as I always have, I ran out of time on the previous class and did not get it done.

@IvoFlipse :(

Ivo Flipse

@Kortuk Luckily there will be plenty of new classes in the future :P

Kortuk

23:04

@IvoFlipse generally, things like gaussians are very simple derivatives, but very messy to write down if you have all the terms.

Ivo Flipse

@Kortuk Exactly

Kortuk

@IvoFlipse I am watching the videos for intro to philosophy right now.

@IvoFlipse I dont think I am going to make this for other reasons.

Ivo Flipse

I no longer have enough hours in a day to keep up with every interesting course

Another problem is that you can't master everything

Kortuk

@IvoFlipse sounds legit, the data analysis class that is starting sounded fun, but by the time I found time, too late. I had a really nastey flu last week and got behind on everything.

Ivo Flipse

Studying more Math and Statistics is all nice and well, but I'm also expected to program some of this stuff (or now how to use the libraries properly), clean up the data or make tools to clean them, know what the data means, design a GUI so non-programmers can view the results, write about what this stuff means in practice

Most people just get to focus on their specialty

And every time I learn something new, I just scratch the surface of yet another topic I don't know anything about

Kortuk

23:07

@IvoFlipse the interesting problems in science are always a problem that crosses a boundary between specialties.

@IvoFlipse That is how the world works.

Ivo Flipse

@Kortuk Auch, are you feeling better now?

And the Data Analysis course seems good, but I can't stand R's syntax

Kortuk

@IvoFlipse Professor once put it, you start learning and realize how little you know, then get to a point where you feel like you know it all, then you keep learning and learn all that no one knows.

Ivo Flipse

But alas, I'll just bite the bullet and try to master it some more

Kortuk

@IvoFlipse Yeah, I wanted to learn R, which is why I signed up, is it nasty?

@IvoFlipse Flu turned into bronchitis, I am not doing better yet.

@IvoFlipse Ha, that is the @IvoFlipse I know.

Ivo Flipse

@Kortuk I don't find it very consistent and it looks ugly

@Kortuk That doesn't sound well indeed

@Kortuk Heck, a year ago I didn't know what a Gaussian was to begin with

Kortuk

23:10

@IvoFlipse That is a pretty amazing step forward, to know how to use a gausian filter as opposed to learning what a gaussian is. Very impressive!

Ivo Flipse

What helped is that so many courses used exp and log for statistics and whatnot

I never had to deal with that kind of stuff

to me statistics used to imply painstakingly copy pasting data into SPSS

Kortuk

@IvoFlipse That makes stat sound really boring :(

Ivo Flipse

@Kortuk It was really boring

And it completely ignores time series, because those in most cases are too complicated for intro to stats courses

One problem I have with my pressure measurements is that there's so many variables and degrees of freedom

I can compare the summed up pressure over time, but not every step has the same duration

I can normalize them, so they have the same length, but that means I can only compare the pattern/shape

Off course, I can do another statistical comparison based on the absolute values of the step durations, but then I have to somehow combine the results afterwards

Kortuk

@IvoFlipse Yes, this is the challenge when you go from analyzing a picture to a video. Sounds like a serious challenge.

Ivo Flipse

I was just about to say, it gets worse when looking at the raw pixel data

I can normalize the size, so I can compare the distribution, but you lose the size information

Or if someone only loads half his foot, and I'd want to compare the distribution with a regular foot, I can't scale his half foot up to the full width, because I'd be comparing the wrong part of the foot

Kortuk

23:17

@IvoFlipse you can simplify one dimension by removing it, but then you lose it. :)

Ivo Flipse

So I'm thinking about simply doing any statistical test I can come up with and then find out which actually make sense

But then I need sufficient data as to not be overfitting

And obviously there's something to say against using a shotgun approach

Kortuk

@IvoFlipse That is an issue.

Ivo Flipse

Indeed, but I bet there's also a publication bias against a lot of variables

So we simply don't know what's interesting and what's not, because the uninteresting results never get published

And most humans aren't very good at keeping tons of variables in their head, so they focus on those that are easily accessible or clearly show differences

Btw the Data Analysis course is indeed nice, it at least takes the time to explain what things 'mean', rather than assuming you already know it

Kortuk

@IvoFlipse hmm. Well I am heading home from Work. I hope you have a great night.

@IvoFlipse I will see you tomorrow.

Ivo Flipse

See you and get well!

Kortuk

23:27

@IvoFlipse I will see you soon, I am sure.

Ivo Flipse

You know where to find me ;)

« first day (308 days earlier) ← previous day next day → last day (652 days later) »