Back at Makerere working on the air pollution monitoring project with Engineer Bainomugisha.
One of my favourite things at Makerere is sitting at a table outside the guest house, with a cup of “African Spiced Tea”, watching the Marabou storks.
I’ve finally got differential privacy for Gaussian processes on pip.
pip install dp4gp
Details and notes are in the development repo, and the paper it is based on is here. Although since then I’ve introduced inducing inputs, which appears to massively improve the results (see also presentation). The figures below demonstrate the scale of the DP noise added without and with inducing inputs.
A quick post to link to a jupyter notebook demonstrating how to download a whole dataset. Here’s the code. It simply hops 7999 entries at a time, downloading all the records that fall between the two ends of each step.
import json, requests from datetime import datetime, timedelta apiurl = 'http://thingspeak.com/channels/241694' nextid = 1 result = None alldata =  endtime = None while result != '-1': print nextid result = json.loads(requests.post(apiurl+'/feeds/entry/%d.json' % nextid).content) starttime = endtime if result == '-1': endtime = datetime.now() else: endtime = datetime.strptime(result['created_at'],'%Y-%m-%dT%H:%M:%SZ') if (nextid==1): starttime = endtime else: start = datetime.strftime(starttime,'%Y-%m-%dT%H:%M:%SZ') end = datetime.strftime(endtime-timedelta(seconds=1),'%Y-%m-%dT%H:%M:%SZ') data = json.loads(requests.post(apiurl+'/feeds.json?start=%s&end=%s' % (start,end)).content) print nextid, len(data['feeds']) alldata.extend(data['feeds']) nextid += 7999 #thought download was 8000 fields, but it's 8000 records. 8000/len(result)
Using AWS: The cloud is where things are these days, learning to use their interface and the API. Maybe could approach AWS to see if they’ll provide free credits to allow the course to use their servers? They do have a free tariff that students could use maybe?
Mobile development: something that’s now ubiquitous, so really should be in a course maybe at Makerere? Could make it one week.
Microprocessor development: the arduino and atTiny
Yesterday I took part in the department of Development Informatics big data event at the University of Manchester.
Really interesting discussion. Made me think more carefully about what effect the blind-spots in my data will have, and how collecting data can make these blind spots worse.
At the end of the day we had a bit of a discussion. Our group was particularly concerned by the effect on power shifts or concentrations with increased data aggregation (similar to Neil Lawrence’s Digital Oligarchy).
We started with the question “What will be done with the data?” and then “How does it become information?”
This led to the obvious point that it depends on who can use it, which then reinforces the power-shift that we started from. The outcome of “How does it become information” leads to the question “How can using data actually foster development? (and avoid inequality)”. We also had as concerns around the transformation from data to information that it is often bias or focused on inanimate or simple things (measuring the water pump rather than the people).
We finally looked at how to change or stop the shift or concentration in power. Two options presented themselves, either to stop using the data, and halt the path to large scale big-data analyses. This seems implausible, given the path we are on. A second option was “Can [the power shift] be mitigated by giving everyone access?”, in other words, will open data save us from the digital oligarchy?
This was again criticised; how can an illiterate farmer or boda-boda driver engage or use large data sets?
My own view is that we need layers of intermediary; from the machine learning/analysis experts who can combine and use the data, and visualise it in clear ways, to journalists and civil society who effectively ‘represent’ the citizen. Our concern is that the machine learning expert is a very particular part of society: usually white, highly educated, young and male. We can go some-way to mitigate that by investing in and supporting MSc and PhD level education in developing countries… however, I’m aware the students at Makerere (for example) were not a ‘typical’ sample of the Ugandan population. Most of the population is rural, with a good proportion unable to read or write. I suspect that the Ugandan students will represent their country-men and women little better than a muzungu. However, it is a start down the path, towards some form of democratic or universal access to the power provided by machine learning and big-data.
Regarding our new paper: Differentially Private Gaussian Processes.
In layman terms Gaussian Processes (GPs) are, usually, used to describe a dataset, to allow predictions to be made. E.g. given previous patient data, what is the best treatment a new patient should receive? It’s a nice framework as it incorporates assumptions clearly, and, because of its probabilistic underpinnings gives estimates of the confidence for a particular estimate.
Differential Privacy is a method that’s recently started to go main-stream. I’ve written a brief introduction presentation here. The general idea is to add noise to any query result to mask the value of an individual row in a database, but still allow inference to be done on the whole database.
My research areas cover both Gaussian Processes and Differential Privacy, so it seemed to make sense to see if I could apply one to the other. In our latest paper we look at two ways to do this:
For the former I developed a new kernel (a way of describing how data is correlated or structured) for binned or ‘histogram’ data. See this ipython notebook for examples. This hopefully is useful for many applications (outside the DP field). For example any inference using binned datasets. At the moment I’ve only applied it to the RBF kernel.
For the latter I used the results of  to determine the noise required to be added to the mean function of my GP. I found that we could considerably reduce the noise scale by using inducing (pseudo) inputs.
Both methods can be further improved, but it’s still early days for Differential Privacy. We need to look at how to apply DP to as many methods as possible, and start to incorporate it into more domains. I’ll be looking at how to apply it to the MND database. Finally, we need an easy “introduction to DP for practitioners”. Although I don’t know if the field is sufficiently mature for this yet.
 Hall, R., Rinaldo, A., & Wasserman, L. (2013). Differential privacy for functions and functional data. The Journal of Machine Learning Research, 14 (1), 703–727.
A few months ago I took part in the NBN R hack-evenings at the University of Sheffield. Unlike everyone else in the room I was coding in python.
import pynbn c = pynbn.connect('lionfish_pynbn','password'); sp = c.get_tvk(query='Bombus terrestris') #get the tvk (the "taxon version key" for buff tails) keys =  for res in sp['results']: k = res['ptaxonVersionKey'] keys.append(str(k)) print "%d species match this query string" % len(keys) print keys tvk = keys print "We'll use the first key (%s)" % tvk #we usually take the first item from this list (advice from the NBN hackday) obs = c.get_observations(tvks=[tvk], start_year=1990, end_year=2010) #get observations print "There are %d records for B. terrestris between 1990 and 2010" % len(obs)
I’ll shortly be adding this to pip, so it can be installed with,
pip install pynbn
When I lived in Uganda, I used to cycle to work every day, and was aware of the pollution building as I cycled down the hill into the city centre.
I wanted to measure this pollution, so using the (very cheap) shinyei sensor and a mobile phone investigated what we could measure, and with what accuracy. I’ve not got very far yet, but this funding app gives and idea about what we’ve done and what we hope to achieve.