Follow the data!

Last week I was down in Wellington for the VUW NZCER NZAMT16 Mathematics & Statistics Education Research Symposium, as well as for the NZAMT16 teacher conference. It was a huge privilege to be one of the keynote speakers and my keynote focused on teaching data science at the school level. I used the example of following music data from the New Zealand Top 40 charts to explore what new ways of thinking about data our students would need to learn (I use “new” here to mean “not currently taught/emphasised”).

It was awesome to be back in Wellington, as not only did I complete a BMus/BSc double degree at Victoria University, I actually taught music at Hutt Valley High School (the venue for the conference) while I was training to become a high school teacher (in maths/stats and music). I didn’t talk much in my keynote about the relationship between music and data analysis, but I did describe my thoughts a few years ago (see below):

All music has some sort of structure sitting behind it, but the beauty of music is in the variation. When you learn music, you learn about key ideas and structures, but then you get to hear how these same key ideas and structures can be used to produce so many different-sounding works of art. This is how I think we need to help students learn statistics – minimal structure, optimal transfer, maximal experience. Imagine how boring it would be if students learning music only ever listened to Bach.

https://www.stat.auckland.ac.nz/en/about/news-and-events-5/news/news-2017/2017/08/the-art-of-teaching-statistics-to-teenagers.html

Due to some unforeseen factors, I ended up ZOOMing my slides from one laptop at the front of the hall to another laptop in the back room which was connected to the data projector. Since I was using ZOOM, I decided to record my talk. However, the recording is not super awesome due to not really thinking about the audio side of things (ironically). If you want to try watching the video, I’ve embedded it below:

You can also view the slides here: bit.ly/followthedataNZAMT. I’m not sure they make a whole lot of sense by themselves, so here’s a quick summary of some of what I talked about:

  • Currently, we pretty much choose data to match the type of analysis we want to teach, and then “back fit” the investigative problem to this analysis. This is not totally a bad thing, we do it in the hope that when students are out there in the real world, they think about all the analytical methods they’ve learned and choose the one that makes sense for the thing they don’t know and the data they have to learn from. But, there’s a whole lot of data out there that we don’t currently teach students about how to learn from, which comes from the computational world our students live in. If we “follow the data” that students are interacting with, what “new” ways of thinking will our students need to make sense of this data?
  • Album covers are a form of data, but how do we take something we can see visually and turn this into “data”. For the album covers I used from one week of 1975 and one week of 2019, we can see that the album covers from 1975 are not as bright and vibrant as those from 2019, similarly we can see that people’s faces feature more in the 1975 album covers. We could use the image data for each album cover, extract some overall measure of colour and use this to compare 1975 and 2019. But what measure should we use? What is luminosity, saturation, hue, etc.? How could we overfit a model to predict the year of an album cover by creating lots of super specific rules? What pre-trained models can we use for detecting faces? How are they developed? How well do they work? What’s this thing called a “confusion matrix”?
  • An intended theme across my talk was to compare what humans can do (and to start with this), with what we could try to get computers to do, and also to emphasise how important human thinking is. I showed a video of Joy Buolamwini talking about her Gender Shades project and algorithmic bias: https://www.youtube.com/watch?v=TWWsW1w-BVo and tried to emphasise that we can’t teach about fun things we can do with machine learning etc. without talking about bias, data ethics, data ownership, data privacy and data responsibility. In her video, Joy uses faces of members of parliament – did she need permission to use these people’s faces for her research project since they were already public on websites? What if our students start using photos of our faces for their data projects?
  • I played the song that was number one the week I was born (tragedy!) as a way to highlight the calendar feature of the nztop40 website – as long as you were born after 1975, you can look up your song too. Getting students to notice the URL and how it changes as you navigate a web page is a useful skill – in this case, if you navigate to different chart weeks, you can notice that the “chart id” number changes. We could “hack” the URL to get the chart data for different weeks of the years available. If the website terms and conditions allow us, we could also use “web scraping” to automate the collection of chart data from across a number of weeks. We could also set up a “scheduler” to copy the chart data as it appears each week. But then we need to think about what each row in our super data set represents and what visualisations might make sense to communicate trends, features, patterns etc. I gave an example of a visualisation of all the singles that reached number one during 2018, and we discussed things I had decided to do (e.g. reversing the y axis scale) and how the visualisation could be improved [data visualisation could be a whole talk in itself!!!]
  • There are common ways we analyse music – things like key signature, time signature, tempo (speed), genre/style, instrumentation etc. – but I used one that I thought would not be too hard to teach during the talk: whether a song is in the major or minor key. However, listening to music first was really just a fun “gateway” to learn more about how the Spotify API provides “audio features” about songs in its database, in particular supervised machine learning. According to Spotify, the Ed Sheeran song Beautiful people is in the minor key, but me and guitar chords published online think that it’s in the major key. What’s the lesson here? We can’t just take data that comes from a model as being the truth.
  • I also wanted to talk more about how songs make us feel, to extend thinking about the modality of the song (major = happy, minor = sad), to the lyrics used in the song as well. How can we take a set of lyrics for a song and analyse these in terms of overall sentiment – positive or negative? There’s lots of approaches, but a common one is to treat each word independently (“bag of words”) and to use a pre-existing lexicon. The slides show the different ways I introduce this type of analysis, but the important point is how common it is to transfer a model trained within one data context (for the bing lexicon, customer reviews online) and use it for a different data context (in this case, music lyrics). There might just be some issues with doing this though!
  • Overall, what I tried to do in this talk was not to showcase computer programming (coding) and mathematics, since often we make these things the “star attraction” in talks about data science education. The talk I gave was totally “powered by code” but do we need to start with code in our teaching? When I teach statistics, I don’t start with pulling out my calculator! We start with the data context. I wanted to give real examples of ways that I have engaged and supported all students to participate in learning data science: by focusing on what humans think, feel and see in the modern world first, then bringing in (new) ways of thinking statistically and computationally, and then teaching the new skills/knowledge needed to support this thinking.
  • We have an opportunity to introduce data science in a real and meaningful way at the school level, and we HAVE to do this in a way that allows ALL students to participate – not just those in enrichment/extension classes, coding clubs, and schools with access to flash technology and gadgets. While my focus is the senior levels (Years 11 to 13), the modern world of data gives so many opportunities for integrating statistical and computational thinking to learn from data across all levels. We need teachers who are confident with exploring and learning from modern data, and we need new pedagogical approaches that build on the effective ones crafted for statistics education. We need to introduce computational thinking and computer programming/coding (which are not the same things!) in ways that support and enrich statistical thinking.

If you are a NZ-based teacher, and you are interested in learning more about teaching data science, then please use the “sign-up” form at undercoverdata.science (the “password” is datascience4everyone). I’ll be sending out some emails soon, probably starting with learning more about APIs (for an API in action, check out learning.statistics-is-awesome.org/popularity_contest/ ).

Upcoming workshop: Using R to explore and exploit features of images

If you’ve been keeping track of my various talks & workshops over the last year or so, you will have noticed that I’ve become a little obsessed with analysing images (see power of pixels and/or read more here).  As part of my PhD research, I’ve been using images to broaden students’ awareness of what is data, and data science, and it’s been so much fun!

If you’re in the Auckland area next week, you could come along to a workshop I’m running for R-Ladies and have some fun yourself using the statistical programming language R to explore images. The details for the workshop and how to sign up are here: https://www.meetup.com/rladies-auckland/events/255112995/

The power of pixels: Using R to explore and exploit features of images

Thursday, Oct 18, 2018, 6:00 PM

G15, Science Building 303, University of Auckland
38 Princes Street Auckland, NZ

30 Members Attending

Kia ora koutou Anna Fergusson, one of our R-ladies Auckland co-organisers, will be the speaker at this meetup. We’ll explore a range of techniques and R packages for working with images, all at an introductory level. Time: 6:00 arrival for a 6:30pm start. What to bring: Laptops with R installed, arrive early if you are a beginner and would like hel…

Check out this Meetup →

This is not a teaching-focused workshop, it’s more about learning fun and cool things you can do with images, like making GIFs like the one below….

Of course it’s a cat!

…. and other cool things, like classifying photos as cats or dogs, or finding the most similar drawing of a duck!

Duck duck stats!

It will be at an introductory level,  and you don’t need to be a “lady” to come along, just supportive of gender diversity in the R community (or more broadly, data science)! If you’ve never used R before, don’t worry – just bring yourself along with a laptop and we’ll look after you 🙂

Cat and whisker plots – sampling from the Quick, Draw! dataset

Last night, I saw a tweet announcing that Google had made data available on over 50 million drawings from the game Quick, Draw! I had never played the game before, but it is pretty cool. The idea behind the game is whether a neural network can learn to recognize doodling – watch the video below for more about this (with an example about cats of course!)

For each game, you are challenged to draw a certain object within 20 secs, and you get to see if the algorithm can classify your drawing correctly or not. See my attempt below to draw a trumpet, and the neural network correctly identifying that I was drawing a trumpet.

 

Since I am clearly obsessed with cats at the moment, I went straight to the drawings of cats. You can see ALL the drawings made for cats (hundred of thousands) and can see variation in particular features of these drawings. I thought it would be cool to be able to take a random sample from all the drawings for a particular category, so after some coding I set up this page: learning.statistics-is-awesome.org/draw/. I’ve included below each drawing the other data provided in the following order:

  • the word the user was told to draw
  • the two letter country code
  • the timestamp
  • whether the drawing was correctly classified
  • number of individual strokes made for the drawing

[Update: There are now more variables available – see this post for more details]

So, on average, how many whiskers do Quick, Draw! players draw on their cats?

I took a random sample of 50 drawings from those under the cat category using the sampling tool on learning.statistics-is-awesome.org/draw/. Below are the drawings selected 🙂

Counting how many individual whiskers were drawn was not super easy but, according to my interpretation of the drawings, here is my sample data (csv file). Using the awesome iNZight VIT bootstrapping module (and the handy option to add the csv file directly to the URL e.g. https://www.stat.auckland.ac.nz/~wild/VITonline/bootstrap/bootstrap.html?file=http://learning.statistics-is-awesome.org/draw/cat-and-whisker-plots.csv), I constructed a bootstrap confidence interval for the mean number of whiskers on cat drawings made by Quick, Draw! players.

 

So, turns out it’s a fairly safe bet that the mean number of whiskers per cat drawing made by Quick, Draw! players is somewhere between 2.2 and 3.5 whiskers. Of course, these are the drawings that have been moderated (I’m assuming for appropriateness/decency). When you look at the drawings, with that 20 second limit on drawing time, you can see that many players went for other features of cats like their ears, possibly running out of time to draw the whiskers. In that respect, it would be interesting to see if there is something going on with whether the drawing was correctly classified as being a cat or not – are whiskers a defining feature of cat drawings?

I reckon there are a tonne of cool things to explore with this dataset, and with the ability to randomly sample from the hundreds and hundreds of thousands of drawings available under each category, a good reason to use statistical inference 🙂 I like that students can develop their own measures based on features of the drawings, based on what they are interested in exploring.

After I published this post, I took a look at the drawings for octopus and then for octagon, a fascinating comparison.

I wonder if players of Quick, Draw! are more likely to draw eight sides for an octagon or eight legs for an octopus? I wonder if the mean number of sides drawn for an octagon is higher than the mean number of legs draw for an octopus?