It has been a long time since I touched any IOS code, so I was quite looking forward to a chance to do some when the IOS team within Google+ released the SDK into preview at Google I/O last month. Luckily, the documentation on the SDK page on developers.google.com is excellent, and covers the basics of the setup, but in part because I’ve forgotten almost everything about XCode and IOS, I thought it best to just write down what I did and what I encountered, primarily as an Aide-mémoire for the future. Anyone building a new app should follow the documentation - but perhaps some developers will hit an issue I did and get something out of this post!

In one of his talks at QCon, John Allspaw mentioned using Holt-Winter exponential smoothing on various monitoring instances. Wikipedia has a good entry on the subject, of course, but the basic idea is to take a noisy/spikey time series and smooth it out, so that unexpected changes will stand out even more. That’s often initially done by taking a moving average, so say averaging the last 7 days of data and using that as the current day’s value. More complicated schemes weight that average, so that the older data contributes less.

At the recent PHP UK Conference 2012 I had the opportunity to chat about machine learning and IR with a bunch of very smart people. One of the conversations included the always enlightening Rowan Merewood, and was around ranking Twitter friends. It’s reasonably well known that Google used to use a variant of PageRank based on who-follows-who to rank it’s Twitter search results (back when it had them). The question is, could the same kind of thing work over a much smaller set - say using it to rank the influence users I follow, in order, perhaps, to prioritise tweets?

I had a great time at the recent PHP Benelux Conference in Belgium. There was a real mix of very interesting people to talk to, and I came away from it buzzing with new ideas (and a ridiculously long todo list). Some of the conversations I had during the weekend were around technical presenting at conferences and usergroups, so I thought I’d collect a handful of the tips that were discussed into a post, and use a few of my favourite speakers at the event to illustrate them.

A lot of interesting techniques involve taking statistical samples, and using those to predict what we’ll see in the future. Usually this works pretty well, but when we’re dealing with a lot of options or if we have some options that are very rare that approach can go pretty wrong. If we go down the street and note down how many men and women we see, we’ll probably be able to use that to predict the chance of the next person we see being male or female pretty well. However, if we were counting all the species of animals we encounter, and trying to use that to predict what we’ll see in the future, we’d likely run in to a couple of problems.

In the last post we had a simple stepping algorithm, and a gradient descent implementation, for fitting a line to a set of points with one variable and one ‘outcome’. As I mentioned though, it’s fairly straightforward to extend that to multiple variables, and even to curves, rather than just straight lines.

I’ve had a couple of emails recently about the excellent Stanford Machine Learning and AI online classes, so I thought I’d put up the odd post or two on some of the techniques they cover, and what they might look like in PHP.

For my own benefit, if nothing else, since I keep seeming to need this snippet of code, I thought I’d encapsulate a Xapian More Like This/Find Similar example in a very brief blog post.

Benfords Law is not an exciting new John Nettles based detective show, but an interesting observation about the distribution of the first digit in sets of numbers originating from various processes. It says, roughly, that in a big collection of data you should expect to see a number starting with 1 about 30% of the time, but starting with 9 only about 5% of the time. Precisely, the proportion for a given digit can be worked out as:

Monte Carlo simulations are a handy tool for looking at situations that have some aspect of uncertainty, by modelling them with a pseudo-random element and conducting a large number of trials. There isn’t a hard and fast Monte Carlo algorithm, but the process generally goes: start with a situation you wish to model, write a program to describe it that includes a random input, run that program many times, and look at the results.