Visualizing machine learning | Hacker News Reader

Visualizing machine learning

r2d3.us

170 pointssnickmy22 comments

I realise this is probably the way the web is going, but I personally really struggle with this sort of presentation. The content is great, it looks lovely, the actual infographics are very clear, and unlike lots of sites, it actually scrolls very, very smoothly.

But it makes it _so_ hard to bookmark something in your mind, or to skip back to something, or to view an infographic in the context of the next or previous paragraph, or any number of things that have been easy to do with text and images for thousands of years. Trying to change the text size in the hope that you can just view more text breaks everything.

I find it really affects my overall comprehension of a piece, and the speed with which I can get through it. Given the universally positive comments elsewhere, I assume my brain is just wired for the dark ages.

Agreed 100%. I personally hate this effect everywhere that it's implemented. This usage is ok relative to most but it's still awful. Why can't we just get a nice post with some pictures along the side? Everything doesn't have to be fancy and difficult to register.

This "reads" more like a video, rather than a traditional book with graphics. I wonder what a good compromise would look like.

I agree with you, despite having enjoyed that presentation very much.

Allow me to define "web page" as an HTML document optionally enhanced with CSS and JS and "web presentation" as an experience delivered over the web but which fundamentally cannot be rendered as a static document.

"No, I can't explain the dance to you; If I could say it I wouldn't have to dance it." ~Isadora Duncan

As much as I like and enjoy well-made "web presentations" I feel wary of the high praise that doesn't take into account the points you raise above. These things can get away from people. Look how easily light-grey body text swept through the web.

As cool as this is, it's still a far cry from e.g. Alan Kay's "active documents" and it's not really a "web site" (IMO).

Great piece & truly awesome visualization.

I would add, as useful resource, this full course on Amazon Machine Learning: https://cloudacademy.com/amazon-web-services/courses/amazon-...

Pretty visualizations... but many of the animations kill this for me. Some are great, animating the decision-tree-pachinko machine for training/CV data sets is fantastic... but all the swooping and building just made me want to skip half of the charts entirely.

This appeared in Show HN a few weeks ago [1]; really awesome visualizations.

https://news.ycombinator.com/item?id=9955553

Well I didn't learn anything I didn't already know but It was fun to read and I think that for someone completely new to the field it would act as a good introduction to the idea of decision trees.

I think the title promises too much though (I realize that it is probably meant as an overall title for series of posts). There is no machine learning in this presentation, no algorithms for discovering the splits are described so the machine doesn't learn anything.

This site is so epic, really. Good stuff! the animations / visuals on there are great too! really cool!

In the "scatterplot matrix" visualization, I am having trouble understanding what the X & Y axes are. Please explain the triangle of scatterplots being formed using the 7 dimensions?

I'm not the OP.

To find the X dimension for one of the 7 scatterplots in that triangle thing put your mouse on the scatterplot you want and move down till you hit some text, that's the X axis. The Y axis is to the left.

They're just trying to illustrate 7 dimensional data. There are 67 2-dimensional comparisons, scatterplots, 67/2 if you eliminate ones made redundant by symmetry, 6*7/2+7 if you add 1-dimensional histograms for "self comparisons".

So a scatterplot matrix, a splom, displays those 21 scatterplots in a meaningful way. Each "row" and "column" of scatterplots compares a single dimension against all of the others.

You usually look at a thing like this to see if you'll get lucky and find a good "orthogonal" comparator—in other words, that just two normal, human-interpretable dimensions already form a good splitting plane like "elevation cross $/sq-ft" does in the running example.

I'm finding my mind blown with this novel way of using a single point in space to represent both X & Y axes based on context.

Thanks for the explanation!

Brilliant article and fantastic intro, though I think some key concepts aren't discussed at all, such as bias and variance. Looking forward to future articles.

Beautiful - in my opinion, a great example of fancy interactive graphics enhancing both the appearance and the ease of understanding an article. Great work!

Absolutely love this! Someone please link a tutorial on how to do these kinda of paralax/svg graphics, my site NEEDS this!

I couldn't spare time to learn some basics of machine learning. This is awesome to understand general idea.

The whole last part (the tree) was great. Subscribed.

awesome visualizations! a picture tells more than thousand words

Very beautiful.

Very very cool!

Excellent stuff

Reply on news.ycombinator.com