@Phonon @IvoFlipse Hey all!

@Phonon @IvoFlipse So guys, since we are going to be in ML/CV, lets figure out chat room for those. Put it together with PGM or seperate?

@Mohammad We'll keep it together.

@Mohammad I need to figure out a way to rename the room = )

@Phonon Hehe yup

@Phonon "21st Century Learning" hehe

room topic changed to Coursera.org study group: Study Group for Probabilistic Graphical Models, Machine Learning and Conputer vision online classes (no tags)

room topic changed to Coursera.org study group: Study Group for Probabilistic Graphical Models, Machine Learning and Computer Vision online classes (no tags)

room topic changed to Coursera.org study group (XXI Lentury Learning): Study Group for Probabilistic Graphical Models, Machine Learning and Computer Vision online classes (no tags)

...or something like that

room topic changed to Coursera.org study group (XXI Century Learning): Study Group for Probabilistic Graphical Models, Machine Learning and Computer Vision online classes (no tags)

there

@Phonon I was 0_o and this close || to googling "Lentury" lol

lol

Can't spell today

@Phonon Hehe, I have those days

@Mohammad So I'll finally come back to my DSP time-alignment algorithms at work in a week or so. Been working on some tangent things. I haven't read all the papers you linked me to yet, but I'll start reading them next week, and hopefully give you some feedback

@Phonon No worries, I have been dealing with ... tangential... things as well. I know the feeling. Last night I put together a time delay estimation algo real quick for this question - (dsp.stackexchange.com/questions/1949/…), Tell me what you think.

Hurrah!

@IvoFlipse Did you win the lottery??

No that the room name changed

I'm working on NLP at the moment, had a score of 0.869, but forgot to commit intermittently, then when I came back from grocery shopping I had made some changes I didn't know how to revert and my score dropped again :(

@IvoFlipse Hehe yup, figured we are going to have at least 2 classes in common among ourselves

@Mohammad Yeah, I don't know how I'm going to pull all three classes

@Mohammad Should be fun.

@Mohammad Intense, that is.

@Phonon Yeah. They should made a class called "How to decide and plan which classes to take"

lol

@Mohammad Don't know about a class, but I'm glad they offer these again and again.

@Mohammad So I can come back to some of them later

@Phonon Yeah, Im really glad ML is being offered again, even though I already sat through all the lectures in 1-week sitting. Helps to go through it again

@Mohammad Nice.

@Mohammad I haven't seen those yet.

@Mohammad I hope they won't be as time-consuming as PGM. That class takes ~15 hours a week for me.

@Phonon Yeah, ML I know is not going to be nearly as time consuming as PGM. Not sure about CV though

Yeah, I have to get started on PGM for next week, but getting a good NLP score is taking longer than planned :(

@IvoFlipse Is NLP as time consuming as PGM?

no, but I like this exercise :P

PGM is just way more complicated

Awesome!

I'll take NLP when it's offered again then

We have to program a Maximum Entrophy Markov Model (well only a part of it)

to find names of persons in a body of text

based on features like whether a word has a capital, is preceded by something etc

@Phonon @IvoFlipse You know this is what I love about those classes, they are all so inter-related but its not obvious. Which is why I try and 'take' classes that I have absolutely no interest in, (like NLP), because one lecture at one point hits on markov processes and I learn from that, and apply the same Tool somewhere else. Its so awesome.

Nice

Yeah I used QuickSort in CS101 and Levenstein distance from NLP too

and NLP is a gentle introduction into a lot of probability stuff

and its way more practical: count number of occurrences of X and Y, add and divide blabla

better than a wall of LaTeX

loool

"Wall of Latex". LOL. nice.

@IvoFlipse Surely, you mean wall of LaTeX

That's what I said

Any suggestions for features that make a word unlikely to be a name?

@IvoFlipse Not an NLP person, but do you have features that make a word likely to be a verb? Then inverse that and it CAN be a name...

I just browsed through all words starting with a Capital

think I should change tactics

currently I just check for features and add a tag for that

but it should be conditional

if something is a weekday, month, country etc it is NOT a person

if it contains more than two from [ ' . - ] then its probably not a (normal) name but an abbreviation or a company name

all caps is not done

it shouldn't contain + $ € or any numbers

So only if it passes ALL those tests, then perhaps it could be a name

@IvoFlipse Also I mean are you only assuming Western names? Because so many names can have ' and - in them as well if not Western

yes its pretty western only

but we get to be a bunch of n00bs :P

haha

I don't have arabian names in my training set

Yeah Arabic names can have higher likelihood of having dashes and apostrophes

Although I supposed they are pretty well isolated, like -bin- means 'son of'

So dashes there

well if I had more time, I could encode the most common elements

yeah

it wouldn't like something like O'Learry-Smith either

but being strict weeds out false positives more than false negatives

Hehe yeah Irish names can be problematic too

Wow, I just have two functions and they weed out a whole lot of bad combo's

def allcaps(word):
    if word.isupper() or word[1:2].isupper():
        return False
    return True

def checkChars(word):
    for letter in word:
        if letter in [u"\u0024", u"\u20AC", u"\xA3", '+'] or letter.isdigit():
            return False
    return True

that second function checks for $ € and pound sign

I should download a list with all the names of all countries, that should help weed out a lot

damn I really should use regular expressions to match parts of words :(

Just throwing this out for @ivo:

boingboing.net/2008/02/20/swedish-couple-fined.html

:)

@yoda Welcome! Did you get lost? = )

@yoda lol my program wouldn't find that one :P

@Phonon not as lost as datageist :P

@yoda I see = )

but yeah, I do drop in every day to look around... sometimes without logging in

It surely feels like I come here just to ban people :P

@yoda That's what Fitness feels like to me

@yoda I meant the room, not the site. I know you're around. I feel the force.

@Phonon oh that... the subject caught my attention =)

@yoda I assume you know about the free courses and what not?

I annoyed him about it :P

yeah, @IvoFlipse wouldn't let go of me unless I signed up for it.. :P I'm sure there's a pyramid scam in there somewhere that he benefits =)

@yoda No, he's just very-very lonely. = )

@Phonon Indeed I am ;P

@yoda So what are you taking?

shhh... @ivo is still here ;)

Ahhh, I see

@Phonon Hey man. I am also starting to look at onset detection based on how they do it for musical signals.