SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
User avatar
Cuchulainn
Posts: 56690
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 3:53 pm

katastrofa wrote:
3 months? Some prodigy? (It took me 2 years to do my MSc, which is standard, I think.)
Some trend detection in data could be fast and easy.

MSc/MFE.  1 year + project at the end.
A moot point is if you can learn anything in a year. But I digress, as the saying goes.
 
User avatar
ppauper
Posts: 68470
Joined: November 15th, 2001, 1:29 pm

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:03 pm

I know nothing about ML

But a general question:
I realize this is a professional MFE course with taught classes rather than a research degree.
Are you looking for them to do
(a) a traditional thesis which has to be original work and be defended
or
(b) "projects" to learn skills? (and you could even use the same projects year after year)

I don't know how many students are in your course, but I suspect you'll have your work cut out to find projects for all of them either way.
And I don't know the practicalities of it, but if it's "projects" then it might be easier on you to have them work in groups, meaning you would need fewer things for them to do.
 
User avatar
katastrofa
Posts: 6139
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:05 pm

@Cuchulainn Yes, it was very long ago. I even came up with a relatively nice and simple clustering algorithm, but I didn't have time to publish it back then and now I cannot :-/
https://www.averisera.uk/machine-learni ... ering.html

I also used EM clustering,which is a popular alternative to k-means.
 
User avatar
Cuchulainn
Posts: 56690
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:14 pm

ppauper wrote:
I know nothing about ML

But a general question:
I realize this is a professional MFE course with taught classes rather than a research degree.
Are you looking for them to do
(a) a traditional thesis which has to be original work and be defended
or
(b) "projects" to learn skills? (and you could even use the same projects year after year)

I don't know how many students are in your course, but I suspect you'll have your work cut out to find projects for all of them either way.
And I don't know the practicalities of it, but if it's "projects" then it might be easier on you to have them work in groups, meaning you would need fewer things for them to do.

I am not giving a course, but am an external supervisor at a UK university for a while (about 40 theses to date). Till now, we have the usual fiance project but seeing all the hullabaloo around ML it might be a good idea.
It's a type b) project (3 months...)
Assigning two students to a project is good idea. I did it in the past.
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:19 pm

Cuchulainn wrote:
Given all the concern about whether popular ML packages really work

This is outside project scope.

This will NOT be part of the project; the students are NOT beta-testers. They can do plumbing work in a later capacity after they have really understood when it does and does not work.
Rewriting all these off-the-shelf packages from scratch won't teach them "when it does and does not work." Experimenting with the packages will teach them "when it does and does not work."
 
User avatar
Cuchulainn
Posts: 56690
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:19 pm

katastrofa wrote:
@Cuchulainn Yes, it was very long ago. I even came up with a relatively nice and simple clustering algorithm, but I didn't have time to publish it back then and now I cannot :-/
https://www.averisera.uk/machine-learni ... ering.html

I also used EM clustering,which is a popular alternative to k-means.

Excellent. I wish more NN authors could write with your clarity and precision.
BTW a friend of mine from TCD wrote a book on Voronoi
https://www.amazon.com/Generalized-Voro ... 1332131417

I also see Boost supports VD
https://www.boost.org/doc/libs/1_67_0/l ... i_main.htm
 
User avatar
Cuchulainn
Posts: 56690
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:27 pm

Traden4Alpha wrote:
Cuchulainn wrote:
Given all the concern about whether popular ML packages really work

This is outside project scope.

This will NOT be part of the project; the students are NOT beta-testers. They can do plumbing work in a later capacity after they have really understood when it does and does not work.

Rewriting all these off-the-shelf packages from scratch won't teach them "when it does and does not work."  Experimenting with the packages will teach them "when it does and does not work."

You won't get MSc thesis by mucking around with packages. If time permits, use them as 2nd opinion.
We are NOT rewriting packages. I want them to make their own mistakes instead of debugging someone else's . Do you understand now?
 
User avatar
Traden4Alpha
Posts: 23951
Joined: September 20th, 2002, 8:30 pm

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:35 pm

Cuchulainn wrote:
Traden4Alpha wrote:
Cuchulainn wrote:
Given all the concern about whether popular ML packages really work

This is outside project scope.

This will NOT be part of the project; the students are NOT beta-testers. They can do plumbing work in a later capacity after they have really understood when it does and does not work.

Rewriting all these off-the-shelf packages from scratch won't teach them "when it does and does not work."  Experimenting with the packages will teach them "when it does and does not work."

You won't get MSc thesis by mucking around with packages. If time permits, use them as 2nd opinion.
We are NOT rewriting packages. I want them to make their own mistakes instead of debugging someone else's . Do you understand now?
What code do you propose they write? And what percentage of the time will they spend writing code & debugging it rather than experimenting with the performance of that code?
 
User avatar
Cuchulainn
Posts: 56690
Joined: July 16th, 2004, 7:38 am
Location: Amsterdam
Contact:

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 4:41 pm

Traden4Alpha wrote:
Cuchulainn wrote:
Traden4Alpha wrote:
Rewriting all these off-the-shelf packages from scratch won't teach them "when it does and does not work."  Experimenting with the packages will teach them "when it does and does not work."

You won't get MSc thesis by mucking around with packages. If time permits, use them as 2nd opinion.
We are NOT rewriting packages. I want them to make their own mistakes instead of debugging someone else's . Do you understand now?

What code do you propose they write?  And what percentage of the time will they spend writing code & debugging it rather than experimenting with the performance of that code?

I've already addressed these issues with katastrofa AFAIR. See Wed Jun 06, 2018 4:19 pm
They work 24/7 until the project's requirement are satisfied.
 
User avatar
katastrofa
Posts: 6139
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Machine learning for fun, to pay your rent or whatever

June 6th, 2018, 5:04 pm

Cuchulainn wrote:
katastrofa wrote:
@Cuchulainn Yes, it was very long ago. I even came up with a relatively nice and simple clustering algorithm, but I didn't have time to publish it back then and now I cannot :-/
https://www.averisera.uk/machine-learni ... ering.html

I also used EM clustering,which is a popular alternative to k-means.

Excellent. I wish more NN authors could write with your clarity and precision.
BTW a friend of mine from TCD wrote a book on Voronoi
https://www.amazon.com/Generalized-Voro ... 1332131417

I also see Boost supports VD
https://www.boost.org/doc/libs/1_67_0/l ... i_main.htm

Thanks, I try my best. And thanks for the book recommendation. (Apologies for the bugs in the charts - I was trying to make the website work under IE, but it seems it failed in several places. I've patched it for now - I suggest using Firefox or Chrome.)
 
User avatar
katastrofa
Posts: 6139
Joined: August 16th, 2007, 5:36 am
Location: Alpha Centauri

Re: Machine learning for fun, to pay your rent or whatever

July 12th, 2018, 8:56 pm

Software beats animal tests at predicting toxicity of chemicals

Using the read-across method, Hartung’s software compares a new chemical to closely related compounds and assesses the probability of toxic effects by reference to the properties of these known chemicals. Effectively, says Hartung, the software mimics how a toxicologist would size up a new chemical but in automated fashion.

Hartung’s database analysis also reveals the inconsistency of animal tests: repeated testing of the same chemical can give different results, because not all animals react the same way. For some types of toxicity, the software therefore provides more-reliable predictions than any individual animal test, he says.
 
 
User avatar
tagoma
Topic Author
Posts: 18233
Joined: February 21st, 2010, 12:58 pm

Re: Machine learning for fun, to pay your rent or whatever

July 27th, 2018, 2:42 pm

Is XGBoost used in reallife? overfitting   Merci!
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...