Serving the Quantitative Finance Community

 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

February 15th, 2019, 11:30 am

BTW what's a good C++ open-source for HMMs and BNs?
Google suggests this: https://github.com/KorfLab/StochHMM
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 15th, 2019, 12:00 pm

BTW what's a good C++ open-source for HMMs and BNs?
Google suggests this: https://github.com/KorfLab/StochHMM
Thank you very much. Looks very good.
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 17th, 2019, 2:50 pm

BTW what's a good C++ open-source for HMMs and BNs?
Google suggests this: https://github.com/KorfLab/StochHMM
Thank you very much. Looks very good.
Such libraries would benefit greatly by migrating them to Boost.
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 19th, 2019, 4:36 pm

Here is an interesting paper for friends relevant to this thread. An Artificial Neural Network Representation of the SABR Stochastic Volatility Model by William Mcghee.
Unfortunately, the run-time performance of both ANN (3 hours) and FDM(20 days!) are disappointing. Must be chalking up a yuge electricity bill?
With 'traditional' FDM popular consensus says it can be done in 15 minutes.

IMO, I am becoming more and more convinced that NN and PDE don't mix, really. Maybe that 'Eureka moment' will come...

For this kind of problem, I would expect Lewis/Papadoupolis FDM to be a good baseline.

https://arxiv.org/abs/1801.06141
BUMP
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 19th, 2019, 4:36 pm

I just read the paper per diagonals :)

Isn't the 475 hours for solving 300k FDM as the goal was to generate 3 millions volatilities (10 at a time)?
BUMP
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 19th, 2019, 4:37 pm

Update 
Section 4.3 

1. 1st and 2nd derivatives are not calculated using NN directly (by AD, whatever)... no reason given why. I suspect difficult to implement etc.
2. Instead author uses cubic splines and its derivatives. Something I have done some work on while back for FEM.

OK then, mathematically, each time you differentiate a spline it gets worser and worser (overshoot) as I discuss in mathematical detail and in numbers in the recent 2nd edition of my C++ book. 

Two screen shoots coming up.
BUMP
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 19th, 2019, 4:39 pm

Image
BUMP
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

February 19th, 2019, 10:02 pm


Google suggests this: https://github.com/KorfLab/StochHMM
Thank you very much. Looks very good.
Such libraries would benefit greatly by migrating them to Boost.
I'm not sure. It may be a more attractive tool for users if it's free of Boost dependencies. Some organisations ban Boost.
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

February 19th, 2019, 10:15 pm

Politicians fume after Amazon's face-recog AI fingers dozens of them as suspected crooks

https://www.theregister.co.uk/2018/07/2 ... ion_sucks/
I suspect that Amazon made some pretty rookie errors in training their system. If they get a different error rate on black and white people, it's likely that their training dataset was unbalanced, e.g. had different number of white and black defendants. We don't know if they also trained it with photos of people who are not in the database of known suspects. It's very hard to say anything concrete about a closed-source system.
The test done by ACLU was not terribly scientific either. They didn't test for false negatives, only for false positives. A 5% false positive error rate, in itself, is not so bad, if there's 2nd stage vetting. Most people on the street are not wanted criminals. If you have few false negatives, and only 5% false positives which you weed out by manual verification, that's not so bad.
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 20th, 2019, 1:35 pm


Thank you very much. Looks very good.
Such libraries would benefit greatly by migrating them to Boost.
I'm not sure. It may be a more attractive tool for users if it's free of Boost dependencies. Some organisations ban Boost.
Boost libraries are very professional. Not sure if a lot of github projects (e.g. StochHMM) are in the same league.

Many of the 'universally relevant' Boost libraries are now in C++11/C++17.
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 22nd, 2019, 1:36 pm

Microsoft has CNTK using C++ which can't be a bad thing.

https://blogs.microsoft.com/ai/microsof ... -advances/

As the system grew more popular, however, they heard from developers who wanted to combine their own Python or C++ code with the toolkit’s deep learning capabilities.

CNTK v TensorFlow
https://hub.packtpub.com/dl-frameworks- ... w-vs-cntk/
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

February 22nd, 2019, 8:49 pm

"TensorFlow also tries its best to run LSTM RNNs, but in vain."

That's not true.
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 23rd, 2019, 11:54 am

"TensorFlow also tries its best to run LSTM RNNs, but in vain."

That's not true.
Maybe the article was talking about an earlier version. Regarding CNTK, you need Windows 10 and it doesn't work in Windows 7.. And C# is also a great language. I am particularly interested in how the algorithms are implemented, i.e. not black box.

The bigger issue (for me) is how easy is C++ to use with these libraries. I agree Pyrhon is a nice wrapper language for them.

Is it possible to programmatically access TensorFlow through C++?
 
User avatar
ISayMoo
Topic Author
Posts: 2332
Joined: September 30th, 2015, 8:30 pm

Re: If you are bored with Deep Networks

February 23rd, 2019, 8:33 pm

Sure. TensorFlow is C++ under the hood. It is not difficult to write C++ extensions for it.

https://www.tensorflow.org/guide/extend/cc
 
User avatar
Cuchulainn
Posts: 20252
Joined: July 16th, 2004, 7:38 am
Location: 20, 000

Re: If you are bored with Deep Networks

February 25th, 2019, 3:02 pm

Does it (or any library) come with a warranty?