Machine Learning with scikit-learn

Most reveal.js based presentations you can press "esc" and get a slide overview, although that's of limited use. Slides in general are simply not as useful as a reasonable web page if they don't have the presentation along with them in audio or video. It's a peculiar part of HN culture to link to just slides, I don't know of any other group that does it.

A far better introduction to sci-kit learn is the project's examples page, http://scikit-learn.org/stable/auto_examples/index.html, which you can use to get example code and data to generate each type of graph. The documentation on the rest of the site is also of very high quality.

I was going to say that it's not really that much of a problem - until I got to page 7 and couldn't scroll down to see the code because I'm on a laptop. Instead I have to Select All - Copy - Paste into editor. That's unforgivable.

Edit: on closer inspection it's just the last couple of lines that I'm missing - but it's still very annoying.

He has only done one competition as noted in his slides.

Here's how to get MNIST, from the sklearn docs:

>>> from sklearn.datasets import fetch_mldata

>>> mnist = fetch_mldata('MNIST original', data_home=custom_data_home)

I think the handwritten digits dataset used in the presentation is just a subset of MNIST; MNIST is 28x28 and the handwritten digits are 8x8.

Here's a link to the other available sample datasets: http://scikit-learn.org/stable/datasets/index.html

SGD is fast but not necessarily more accurate. If you've got lots and lots of data, then a simple yet fast approach is likely to be a good choice to start with.