Comments for Mathieu's log

Comment on The Little Machine Learner by Mathieu

Mathieu — Sun, 21 Feb 2010 02:40:32 +0000

Thanks for your comment, Ian. In my case, I guess what I like the most is that the implementation is right below the corresponding explanations. This makes the flow of explanations very natural.

Speaking of numerical analysis, I’ve been looking into optimization lately since it’s so important in Machine Learning. It’s a very interesting field.

Comment on The Little Machine Learner by enj

enj — Sat, 20 Feb 2010 21:11:33 +0000

This is an awesome idea. I really like the format for learning the math and the implementation at the same time. The hands on aspect of it is great!

I could see this being a great way to teach numerical analysis or other computational science tools (interpolation, integration, signal processing) as well.

Really sweet!

Comment on Dynamic Time Warping : theory by Mathieu’s log » Blog Archive » Seam Carving in Python

Mathieu’s log » Blog Archive » Seam Carving in Python — Tue, 09 Feb 2010 15:57:27 +0000

[...] The principle is very simple. Find the connected paths of low energy pixels (”the seams”). This can be done efficiently by dynamic programming (see my post on DTW). [...]

Comment on Fantasdic on Mac OS X install how-to by Ja Ru

Ja Ru — Sat, 23 Jan 2010 01:36:37 +0000

Hello,

Alternative is to use the Dictionary application which is bundled with Mac OS.

Comment on Tegaki 0.2 released by Mathieu

Mathieu — Thu, 24 Dec 2009 17:08:25 +0000

Ask your question on the discussion group.

Comment on Tegaki 0.2 released by Ryan

Ryan — Wed, 23 Dec 2009 13:59:00 +0000

I can’t install the tegaki.I don’t know why.Could you write more detail of the install.

Comment on Using internet users to do something useful by How millions of internet surfers, including you, unknowingly help digitize ancient books | Hermes Technologies Ltd.

Tue, 15 Dec 2009 12:22:37 +0000

[...] Blondel explains the idea behind a CAPTCHA in his article on using internet users to do something useful: The idea of CAPTCHA is to ask users to answer a question that only humans can answer in order to [...]

Comment on First look at Cython by Ian

Ian — Mon, 30 Nov 2009 17:09:58 +0000

That’s interesting. I tried Cython for a homework assignment where I was looping over and doing simple operations on a couple of arrays of 100,000 elements. Using just python and numpy it was horribly slow, taking about 5 seconds to compute.

I tried doing the C style syntax in the python with Cython and got about a 2x speedup when I statically typed my arrays and loop variables, and turned off boundary checking. That was still way too slow since this was part of an iterative method.

I simply rewrote the loop in C++ and it computed instantaneously, but I never got around to wrapping the loop and bringing it back to python. I think I’m going to finish the attempt after some free time.

Comment on Dynamic Time Warping : theory by Mathieu’s log » Blog Archive » First look at Cython

Mathieu’s log » Blog Archive » First look at Cython — Fri, 27 Nov 2009 08:14:25 +0000

[...] try Cython, I’ve written a small extension to perform Dynamic Time Warping (see my recent post). Rather than writing my extension entirely in Cython, I’ve opted for the solution of writing [...]

Comment on Web Canvas by Mathieu

Mathieu — Tue, 17 Nov 2009 15:21:41 +0000

Thank you very much Shawn. Nice to see the webcanvas working on the IPhone.

Actually I’m also a user of Anki. I’ve been using the Maemo port of it on my N810. I also use the online service extensively to keep my laptop (where I prepare my vocab lists) and my N810 in sync.

I had a similar idea to you recently. Since anki is broken down into libanki and ankiqt, it should be possible to use libanki as backend and write a new frontend which uses handwriting recognition as input method. Since both my project Tegaki and Anki are written in Python, that seems like a good match. In ankiqt, you need to choose if you found the question hard, good or easy. With handwriting recognition, we can use the level of confidence returned by the recognizer instead.

What problems are you having with the repo exactly?