Coursera Study Group

Ivo Flipse

12:41

@Phonon I found a little help with my blocklog function, to make it do what I mentioned last time (AKA make it use the right factor values)

turned out my MMC and Gibbs functions were working just fine

which just goes to show how messed up the assignment is

Phonon

13:07

@IvoFlipse Yeah, I've been there, too.

Ivo Flipse

@Phonon I'm trying to do the accompanying quiz

but it seems I'm missing some steps to get it running properly

Let’s run an experiment using our Gibbs sampling method. As before, use the toy image network and set the on-diagonal weight of the pairwise factor (in ConstructToyNetwork.m) to be 1.0 and the off-diagonal weight to be 0.1. Now run Gibbs sampling a few times, first initializing the state to be all 1’s and then initializing the state to be all 2’s. What effect does the initial assignment have on the accuracy of Gibbs sampling? Why does this effect occur?

I'm getting: Undefined function 'ComputeExactMarginalsBP' for input arguments of type 'struct'.

Phonon

You need to take that from your Week 4

@IvoFlipse I just added my Week 4 folder to the path, and everything worked fine.

Ivo Flipse

Ah crap, the error is that it can't find it instead of that its not working...

sigh

Phonon

yeah

@IvoFlipse Whenever you get undefined function, that's what it means.

Ivo Flipse

13:24

What the hell I have no idea how to interpret those figures :P

Ivo Flipse

14:05

> Undefined function 'scomponents' for input arguments of type 'double'.

How awkward, what function is that anyway

d'oh its strongly connected components, I wrote that in algo-class

ah, those are in gaimc, which I reckon I haven't added to the path

Phonon

@IvoFlipse Yeah, that's right. Don't know why they're in a different directory.

@IvoFlipse Your results look right, although mine didn't look anything like that.

@IvoFlipse What are those the results of?

@IvoFlipse What was your network?

Ivo Flipse

@Phonon TestToy

something going haywire I suppose

@Phonon no clue :P

I'm running it again to finish off the quiz

then I can move on to week 6 \o/

Phonon

@IvoFlipse Oh, I see. With which method was that?

Ivo Flipse

@Phonon TestToy I believe

Phonon

@IvoFlipseOh, I see it says Gibbs

@IvoFlipse Cool.

Ivo Flipse

14:14

yup, which also happens to run forever

Phonon

@IvoFlipse Yes, same for me.

Ivo Flipse

Now I completed all functions and they seem to be working, so it might give different results this time

some of the graphs looked rather weird last time

Not sure what it means either :P

Phonon

14:49

@IvoFlipse Yeah, some of mine looked very similar.

Ivo Flipse

lol after some guessing I managed to get 32/32 :P

I'm a pro at guessing multiple choice

Mohammad

@Phonon @IvoFlipse Morning all!

Ivo Flipse

hi @Mohammad

Phonon

@Mohammad Hey

@IvoFlipse I wish I could say the same = )

Mohammad

@IvoFlipse How do you say good morning in Dutch? Guten Morgen? Thats German no?

Ivo Flipse

14:51

@Mohammad That's German yeah, its Goedemorgen in Dutch

which looks pretty similar I'd say

Mohammad

@IvoFlipse Hehe yup

Phonon

@Mohammad Except you should really be asking how do you say good evening? instead = )

Mohammad

@Phonon I was just thinking that hehe

@Phonon @IvoFlipse Goetentag?

Ivo Flipse

goed = good, morgen = morning, avond = evening, middag = afternoon (mid = half)

Mohammad

@IvoFlipse I know the German evening is 'tag' so suspected it was similar :P

Ivo Flipse

15:02

no, tag = dag = day

abend = avond = evening

as you can see, we're like a mix between German and English

Mohammad

@IvoFlipse Yikes. Im more rusty than I imagined. Either that or I was telling all those Germans good morning around 10pm. lool.

Ivo Flipse

morgen = morning

so you were saying good day, which you can say all day, every day

Mohammad

@IvoFlipse Ah, day as in 'day', not necessarily day time.

Ivo Flipse

@Mohammad well you can abuse the ambiguity to your advantage :P

Mohammad

@IvoFlipse Hehe, true.

Phonon

15:14

@IvoFlipse I like how you demonstrate language mixing geographically.

Ivo Flipse

I could mix in France too, because we use words like paraplu, affaire, après-ski or bagage which are all French

Mohammad

@Phonon @IvoFlipse On a train ride once from Holland to Germany, after 5 or so hours on the train the operator announced we had finally arrived in "Deutchland", and as kid I wondered why we were still in Holland since they speak Dutch there. 0_0 lol

Phonon

@IvoFlipse Jeremy Clarkson one said that Belgium is a country that was put on the map specifically so that the Germans and the English had some place settle their differences. = )

Ivo Flipse

@Mohammad That's rather unlikely, the distance from Amsterdam to our border is like 100 km :P

Mohammad

@IvoFlipse We were going to Hanover

Ivo Flipse

15:17

@Mohammad Then it might be possible, horrible connections

Mohammad

lol

Ivo Flipse

@Phonon Its true :P

I live near a WW2 memorial graveyard :\

we had some important battles in this area

Operation Market Garden

|combatant1= Poland Dutch Resistance |combatant2= Germaansche SS in Nederland |commander1= Bernard Montgomery Lewis H. Brereton Miles Dempsey Frederick Browning |commander2= Gerd von Rundstedt Walter Model Kurt Student Wilhelm Bittrich Gustav-Adolf von Zangen Friedrich Kussin |strength1=41,628 airborne troops |strength2=Unknown |casualties1= b15,326–17,200 casualties88 tanks144 transport aircraft |casualties2=bincomplete estimates:6,315–13,300 casualties |notes=aMore detailed information regarding the outcome of this operation can be found in the 'Aftermath' section. bMore detailed...

Mohammad

@IvoFlipse @Phonon Friends tell me Belgium is 'surprisingly' sectarian, there is lots of linguistic/cultural segmentations

Ivo Flipse

That bottom red arrows stops right at the village I currently live

Mohammad

Sort of like Quebec in Canada

Phonon

15:20

@IvoFlipse Holy cow dude!

Ivo Flipse

@Mohammad Don't get me started, I lived there for 2 years (and still work there)

Phonon

@IvoFlipse I guess the entire Europe is riddled with historic battlefields.

Ivo Flipse

They have a federal government (even though they have 6 million inhabitants or what not)

@Phonon That's just because we have people who could write history when it was made, America didn't :P

@Mohammad Then they have goverments for the Dutch north, the French south, a small German province in the east and Brussels itself too

Mohammad

@IvoFlipse lol wow.

Phonon

@IvoFlipse As Eddy Izzard one famously said in Dressed to Kill, "As you can tell, I am from Europe... where history comes from."

Ivo Flipse

15:22

Those are all split into smaller provincial governments too, which makes you wonder what they need the second layer for

@Phonon The Chinese beg to differ :P

Phonon

@IvoFlipse lol

Ivo Flipse

@Mohammad Its even so bad that the Flemish and Wallonian parts have separate rules for a whole bunch of things, which makes doing business across regions 'interesting'

Mohammad

@IvoFlipse I once met a (I think Dutch) girl who was ranting and raving about 'Flamish' or something.

@IvoFlipse Thats it! Flemmish! Yes she would not stop ranting about them, whoever they are

Ivo Flipse

like a Flemish company can't get Brussels funding when they collaborate with a university from Brussels

@Mohammad They're the Dutch who were stupid enough to remain behind when the Spanish came

Mohammad

@IvoFlipse LOOOL

Ivo Flipse

15:25

We like to make jokes that the people in Belgium are 'dumb/stupid', they call us "cheese-paring" (thank you google translate)

that last word is obviously quite fitting, given everybody thinks all we eat is cheese

which is silly, we don't just eat cheese

we put it on our bread instead ^^

Phonon

@IvoFlipse Oh man, these Jeremy Clarkson quotes just keep popping up in my head = )

Ivo Flipse

@Phonon That's because they're true :P

Mohammad

@Phonon The man seems like he knew what he was talking about! :)

Ivo Flipse

Well after 18+ seasons of Top Gear, you have to shoot episodes elsewhere right?

lazur.hubpages.com/hub/Those-typical-Dutch

Phonon

@IvoFlipse I have no problem with anything listed on that site lol. = )

@IvoFlipse But I doubt I would date the girl in the picture.

Ivo Flipse

15:34

@Phonon Neither would I, I wouldn't want to be found dead near anyone wearing those clothes

Mohammad

@IvoFlipse I hesitate to click on that link while at work... 0_0

Ivo Flipse

The only good thing about her is that she drinks nice beer :P

Its SFW

Mohammad

@Phonon BTW: upload.wikimedia.org/wikipedia/commons/5/53/…

@Phonon As something that would be nice for various DSP concepts

@Phonon The animation, etc

Phonon

@Mohammad Wait, what does it do exactly?

Ivo Flipse

@Mohammad Nice one

Phonon

15:47

@Mohammad Oh, it's just an animation. Got it.

@Mohammad Yeah, that's really nice

Ivo Flipse

Pfff my brain is fried, I can't even think of what to do in the first exercise of PA6

Phonon

@IvoFlipse I'll be starting that tonight

Ivo Flipse

They go like: here's a formula, good luck programming it out

uhhh... ok?

And the input format still makes no sense to me, but that's probably because I never got it to begin with :P

Phonon

@IvoFlipse It's jstu a factor product

@IvoFlipse \delta is decision factor

U is utility function

and the rest are your bayesian factors

D is the set of all parents of your decision function

Ivo Flipse

well decision factor if its 1 or should I just take the product and make that remove anything where the decision was 0?

Phonon

15:52

@IvoFlipse I believe it's the likelihood of you making a certain decision based on other factors, i.e. your parents in the bayesian net

Ivo Flipse

what are those var's because there are only 10 factors in I.RandomFactors

so how can they point to values larger than 10

Phonon

I haven't gotten to the exercises yet, but I believe that in this case D = I.DecisionFactors.var

@IvoFlipse LOL. I actually guessed the name decision factor

Ivo Flipse

16:54

> Yup, SimpleCalcExpectedUtility calculates expected utility for a given set of factors; in particular, for a fixed decision factor. In the case of FullI the decision factor is given to you in the input as Sriram stated. Indeed 9 is the decision variable and 11 is the test result variable, so for example the first line means that "if test result (var 11) is 1, then going by the current decision rule, the probability of making decision 1 (i.e. var 9 = 1) is 1.000000".

Don't worry about the expected utility being negative: this is completely fine (and expected); all it means is that given the current decision rule, the expected value of the utility we will end up with is negative. It doesn't tell you anything about whether and in what cases it is in our interest to have surgery, or whether the given decision rule is optimal or not.

All that SimpleCalcExpectedUtility can tell you is "if you were to go with this fixed decision rule, this is the expected value of your utility that would result." Selecting the best decision rule is done by another function (OptimizeMEU). Hope that helps a bit with the intuition :)

I hate how these variables are so non-descriptive, how on earth was I supposed to know that the 9 was the decision and 11 its result?

Mohammad

17:09

@Phonon Would a question on best DSP unis be kosher you think?

@Phonon as long as it abides by (blog.stackoverflow.com/2010/09/good-subjective-bad-subjective)

Phonon

23:55

@Mohammad I don't think it would be a good question. Even though it may be in general, not on this site. You can try academia.stackexchange.com. I'm sure it will be a much better fit for them.

Transcript for

Coursera Study Group