|
Bloody posted:i once did cool algorithm lovely matlab -> lovely c++ -> kinda not so bad c++ once see if you do it as part of a university spin off you actually get paid a real wage
|
#
?
Mar 13, 2015 04:09
|
|
|
|
| # ? May 24, 2024 22:17 |
|
|
and worthless stock options! don't forget about those!
|
|
#
?
Mar 13, 2015 04:09
|
|
|
hobbesmaster posted:real wage well..
|
|
#
?
Mar 13, 2015 04:12
|
|
|
grad school makes it look good!
|
|
#
?
Mar 13, 2015 04:13
|
|
|
Bloody posted:what's the algorithm It's based on a thing that may not be an actual algorithm called a support vector machine with some sorta baysiean criteria added so they call it a relevance vector machine. this guy further extends it into using some "kernel basis functions" and starts calling the vectors units and then it's referred to as the RUM. I think in the final stuff that I would be working with it's called CRUM since it's the RUM applied to Classification problems (in this case the classification of ncRNAs) still reading all the literature on it so i have a maximal amount of time invested before i get the program that the guy wrote and it turns out to be incomprehensible mathematician-code akin to the source of dwarf fortress Marzzle fucked around with this message at 04:32 on Mar 13, 2015 |
|
#
?
Mar 13, 2015 04:29
|
|
|
oh so it's a svm. might i interest you in good svm libraries, such as libsvm or lasvm?
|
|
#
?
Mar 13, 2015 04:31
|
|
|
not sure what is meant by the bayesian criteria but the rest is bog-standard svm for nonlinear decision boundaries
|
|
#
?
Mar 13, 2015 04:32
|
|
|
post the yosposest svm libs u got
|
|
#
?
Mar 13, 2015 04:33
|
|
|
cusvm b/c it doesnt work
|
|
#
?
Mar 13, 2015 04:35
|
|
|
by which i mean i never got it to work
|
|
#
?
Mar 13, 2015 04:37
|
|
|
Bloody posted:not sure what is meant by the bayesian criteria but the rest is bog-standard svm for nonlinear decision boundaries yeah, the RVM/bayesian stuff is cited from a 2001 and 2003 paper from some journals randomly pulling "machine, staticstics, intelligence etc..." to form their names so I dunno how wide spread it is yet. i'd imagine I will probably have to end up reading both of them as I have no idea what the bayesian criteria are either but i'd imagine it's sorta working some kind of covarience (only because that was the thing thrown around when I was meeting with people about it) based thing to make its choices Marzzle fucked around with this message at 04:43 on Mar 13, 2015 |
|
#
?
Mar 13, 2015 04:39
|
|
|
Bloody posted:by which i mean i never got it to work I think that's why the dude was doing it in matlab because while it's a proprietary garbage pile that isn't gonna get me the honeys, it probably has some svm stuff built into it with a fun mathworks(tm) tutorial/faq but I haven't actually check yet i suppose i am gonna now though
|
|
#
?
Mar 13, 2015 04:42
|
|
|
hobbesmaster posted:most stuff like that is in matlab because there's an existing function for everything you could possibly imagine and at a university it's all basically free Why'd they want it in c++? People seem pretty chill with just publishing like 4 papers on a bunch of matlab stuff here
|
|
#
?
Mar 13, 2015 04:50
|
|
|
Marzzle posted:are there any with OFF THE SHELF functionality? yeah, libsvm
|
|
#
?
Mar 13, 2015 05:18
|
|
|
Marzzle posted:Why'd they want it in c++? People seem pretty chill with just publishing like 4 papers on a bunch of matlab stuff here typically so you can get results in less than one computer decade
|
|
#
?
Mar 13, 2015 05:18
|
|
|
Bloody posted:typically so you can get results in less than one computer decade what about cloud matlab harvesting all the geothermal power of iceland so no one will have to learn what a memory pointer is
|
|
#
?
Mar 13, 2015 07:29
|
|
|
Someone post the julialang benchmark where Matlab was like 10,000 times slower than java
|
|
#
?
Mar 13, 2015 08:14
|
|
|
(looks at chart) oh hey thats not so bad 'pos you so funny(notices the log scale)  Marzzle fucked around with this message at 09:45 on Mar 13, 2015 |
|
#
?
Mar 13, 2015 09:37
|
|
|
Marzzle posted:however, it means i gotta look at matlab code and that means it wouldn't be very useful if i wanted to make an html5 interface for it so i was thinking of moving it to octive (which is supposed to be the same thing as matlab sorta but OSSier) the only reasons to even consider octave over matlab are: 1. price 2. ideology
|
|
#
?
Mar 13, 2015 12:29
|
|
|
Bloody posted:i once did cool algorithm lovely matlab -> lovely c++ -> kinda not so bad c++ once what was the improvement when you were done?
|
|
#
?
Mar 13, 2015 13:19
|
|
|
Corla Plankun posted:what was the improvement when you were done? it not being in matlab
|
|
#
?
Mar 13, 2015 13:20
|
|
|
even moving to c++ is an improvement
|
|
#
?
Mar 13, 2015 13:20
|
|
|
fritz posted:the only reasons to even consider octave over matlab are: 3. you need a result while waiting on a quote from mathworks
|
|
#
?
Mar 13, 2015 13:27
|
|
|
Corla Plankun posted:what was the improvement when you were done? mostly that you could read the code and plausibly understand what was going on
|
|
#
?
Mar 13, 2015 13:57
|
|
|
woah julia looks pretty sick, i'll have to take a closer look next time i need to do something mathy although honestly i'll probably just keep using f# and learn to use the R type provider because f# fully owns and nothing mathy i do requires good performance
|
|
#
?
Mar 13, 2015 17:43
|
|
|
julia is fully sick but the lack of a large stdlib is really inconvenient. like it doesn't even have builtin plotting functionality. so i use it never. apparently the ffi is pretty dece but if i want to gently caress around all day with ffi i'll just use c#
|
|
#
?
Mar 13, 2015 19:13
|
|
|
Bloody posted:julia is fully sick but the lack of a large stdlib is really inconvenient. like it doesn't even have builtin plotting functionality. so i use it never. i used gadfly and it was fine (stupid name notwithstanding) http://dcjones.github.io/Gadfly.jl/
|
|
#
?
Mar 13, 2015 20:55
|
|
|
Shinku ABOOKEN posted:i used gadfly and it was fine (stupid name notwithstanding) that's the most tech bubble url
|
|
#
?
Mar 13, 2015 21:53
|
|
|
Bloody posted:mostly that you could read the code and plausibly understand what was going on if that's all then that was a huge waste of time
|
|
#
?
Mar 13, 2015 23:53
|
|
|
Corla Plankun posted:if that's all then that was a huge waste of time 
|
|
#
?
Mar 14, 2015 00:16
|
|
|
it's friday and i finished my major feature of the week like 2 hours ago and have been doing gently caress all it feels weird to be doing nothing and there be sunlight should i get my drink on or what
|
|
#
?
Mar 14, 2015 00:18
|
|
|
PleasureKevin posted:it's friday and i finished my major feature of the week like 2 hours ago and have been doing gently caress all yes i got mine done and then we had a party at work now i'm yosposting, i should just go the hell home
|
|
#
?
Mar 14, 2015 01:18
|
|
|
Awia posted:gotta build everything from the ground up! programming should be taught in something like a trade school programming jobs shouldn't require degrees CS degrees can go back to teaching CS
|
|
#
?
Mar 14, 2015 02:22
|
|
|
code:
|
|
#
?
Mar 14, 2015 03:18
|
|
|
~Coxy posted:CS degrees can go back to teaching CS the dream
|
|
#
?
Mar 14, 2015 03:25
|
|
|
that is definetely an issue here. there is a total lack of CS highschool teachers because of how much money you can make not teaching
|
|
#
?
Mar 14, 2015 03:28
|
|
|
~Coxy posted:programming should be taught in something like a trade school I dunno I think half or more of a cs degree is just a sorta specialized in depth intelligence test that hopefully makes sure you aren't too dumb to figure out novel problems. Not that its any good at that, but if you teach programming like a trade then the second some dummy that memorized all the method calls to use for the right situations like some sorta info-plumber will just be entirely lost. Cs may have very little to do do with iINDUSTRY PROGRAMMING but aside from some gattaca brain scan it's probably the current best-yet-flawed way to make sure a person has the aptitude to learn how do figure stuff out
|
|
#
?
Mar 14, 2015 03:31
|
|
|
as I near the end of my degree, the most useful things ive learned over the past 4 years are: • how to code in python and java along with the important java workflow stuff • how to pick up new languages in a reasonable time • knowing how to find the info i need to solve something • lots of agile team projects
|
|
#
?
Mar 14, 2015 03:39
|
|
|
the thing is, at this point you can get by in most programming jobs without ever solving a novel problem sure, if you work for an actual Software Company then maybe you'll be solving some novel problems but that isn't where like 80% of the work is
|
|
#
?
Mar 14, 2015 03:40
|
|
|
|
| # ? May 24, 2024 22:17 |
|
|
Dessert Rose posted:sure, if you work for an actual Software Company then maybe you'll be solving some novel problems but that isn't where like 80% of the work is sometimes not even then lol
|
|
#
?
Mar 14, 2015 03:47
|
|