Unearthing gold in Toronto

I just got home from Toronto, the mining capital of the world, after an awesome weekend hacking with Diego Castañeda, a recent PhD grad in astrophysics that is working with us) and Anneya Golob (another astrophysicist and Diego's partner). Given how much I bang on about hackathons, it might surprise you to know that this was the first hackathon I have properly participated in, without having to order tacos or run out for more beer every couple of hours.

PArticipants being briefed by one of the problem sponsors on the first evening.

PArticipants being briefed by one of the problem sponsors on the first evening.

What on earth is Unearthed?

The event (read about it) was part of a global series of hackathons organized by Unearthed Solutions, a deservedly well-funded non-profit based in Australia that is seeking to disrupt every single thing in the natural resources sector. This was their fourteenth event, but their first in Canada. Remarkably, they got 60 or 70 hackers together for the event, which I know from my experience organizing events takes a substantial amount of work. Avid readers might remember us mentioning them before, especially in a guest post by Jelena Markov and Tom Horrocks in 2014.

A key part of Unearthed's strategy is to engage operating companies in the events. Going far beyond mere sponsorship, Barrick Gold sent several mentors to the event, the Chief Innovation Officer Michelle Ash, as well as two judges, Ed Humphries (head of digital transformation) and Iain Allen (head of digital mining). Barrick provided the chellenge themes, as well as data and vivid descriptions of operational challenges. The company was incredibly candid with the participants, and should be applauded for its support of what must have felt like a pretty wild idea. 

Team Auger Effect: Diego and Anneya hacking away on Day 2.

Team Auger Effect: Diego and Anneya hacking away on Day 2.

What went down?

It's hard to describe a hackathon to someone who hasn't been to one. It's like trying to describe the Grand Canyon, ice climbing, or a 1985 Viña Tondonia Rioja. It's always fun to see and hear the reactions of the judges and other observers that come for the demos in the last hours of the event: disbelief at what small groups of humans can do in a weekend, for little tangible reward. It flies in the face of everything you think you know about creativity, productivity, motivation, and collaboration. Not to mention intellectual property.

As the fifteen (!) teams made their final 5-minute pitches, it was clear that every single one of them had created something unique and useful. The judges seemed genuinely blown away by the level of accomplishment. It's hard to capture the variety, but I'll have a go with a non-comprehensive list. First, there was a challenge around learning from geoscience data:

  • BGC Engineering, one of the few pro teams and First Place winner, produced an impressive set of tools for scraping and analysing public geoscience data. I think it was a suite of desktop tools rather than a web application.
  • Mango (winners of the Young Innovators award), Smart Miner (second place overall), Crater Crew, Aureka, and Notifyer and others presented map-based browsers for public mining data, with assistance from varying degrees of machine intelligence.
  • Auger Effect (me, Diego, and Anneya) built a three-component system consisting of a browser plugin, an AI pipeline, and a social web app, for gathering, geolocating, and organizing data sources from people as they research.

The other challenge was around predictive maintenance:

  • Tyrelyze, recognizing that two people a year are killed by tyre failures, created a concept for laser scanning haul truck tyres during operations. These guys build laser scanners for core, and definitely knew what they were doing.
  • Decelerator (winners of the People's Choice award) created a concept for monitoring haul truck driving behaviour, to flag potentially expensive driving habits.
  • Snapfix.io looked at inventory management for mine equipment maintenance shops.
  • Arcana, Leo & Zhao, and others looked at various other ways of capturing maintenance and performace data from mining equipment, and used various strategies to try to predict 

I will try to write some more about the thing we built... and maybe try to get it working again! The event was immensely fun, and I'm so glad we went. We learned a huge amount about mining too, which was eye-opening. Massive thanks to Unearthed and to Barrick on all fronts. We'll be back!

Brad BEchtold of Cisco (left) presenting the Young Innovator award for under-25s to Team Mango.

The winners of the People's Choice Award, Team Decelerate.

The winners of the contest component of the event, BGC Engineering, with Ed Humphries of Barrick (left).

UPDATE  View all the results and submissions from the event.

Wish there was a hackathon just for geoscientists and subsurface engineers?
You're in luck! Join us in Paris for the Subsurface Hackathon — sponsored by Dell EMC, Total E&P, NVIDIA, Teradata, and Sandstone. The theme is machine learning, and registration is open. There's even a bootcamp for anyone who'd like to pick up some skills before the hack.

No secret codes: announcing the winners

The SEG / Agile / Enthought Machine Learning Contest ended on Tuesday at midnight UTC. We set readers of The Leading Edge the challenge of beating the lithology prediction in October's tutorial by Brendon Hall. Forty teams, mostly of 1 or 2 people, entered the contest, submitting several hundred entries between them. Deadlines are so interesting: it took a month to get the first entry, and I received 4 in the second month. Then I got 83 in the last twenty-four hours of the contest.

How it ended

Team F1 Algorithm Language Solution
1 LA_Team (Mosser, de la Fuente) 0.6388 Boosted trees Python Notebook
2 PA Team (PetroAnalytix) 0.6250 Boosted trees Python Notebook
3 ispl (Bestagini, Tuparo, Lipari) 0.6231 Boosted trees Python Notebook
4 esaTeam (Earth Analytics) 0.6225 Boosted trees Python Notebook

The winners are a pair of graduate petroelum engineers, Lukas Mosser (Imperial College, London) and Alfredo de la Fuente (Wolfram Research, Peru). Not coincidentally, they were also one of the more, er, energetic teams — it's say to say that they explored a good deal of the solution space. They were also very much part of the discussion about the contest on GitHub.com and on the Software Underground Slack chat group, aka Swung (you're in there, right?).

I will be sending Raspberry Shakes to the winners, along with some other swag from Enthought and Agile. The second-place team will receive books from SEG (thank you SEG Book Mart!), and the third-place team will have to content themselves with swag. That team, led by Paolo Bestagini of the Politecnico di Milano, deserves special mention — their feature engineering approach was very influential, being used by most of the top-ranking teams.

Coincidentally Gram and I talked to Lukas on Undersampled Radio this week:

Back up a sec, what the heck is a machine learning contest?

To enter, a team had to predict the lithologies in two wells, given wireline logs and other data. They had complete data, including lithologies, in nine other wells — the 'training' data. Teams trained a wide variety of models — from simple nearest neighbour models and support vector machines, to sophisticated deep neural networks and random forests. These met with varying success, with accuracies ranging between about 0.4 and 0.65 (i.e., error rates from 60% to 35%). Here's one of the best realizations from the winning model:

One twist that made the contest especially interesting was that teams could not just submit their predictions — they had to submit the code that made the prediction, in the open, for all their fellow competitors to see. As a result, others were quickly able to adopt successful strategies, and I'm certain the final result was better than it would have been with secret code.

I spent most of yesterday scoring the top entries by generating 100 realizations of the models. This was suggested by the competitors themselves as a way to deal with model variance. This was made a little easier by the fact that all of the top-ranked teams used the same language — Python — and the same type of model: extreme gradient boosted trees. (It's possible that the homogeneity of the top entries was a negative consequence of the open format of the contest... or maybe it just worked better than anything else.)

What now?

There will be more like this. It will have something to do with seismic data. I hope I have something to announce soon.

I (or, preferably, someone else) could write an entire thesis on learnings from this contest. I am busy writing a short article for next month's Leading Edge, so if you're interested in reading more, stay tuned for that. And I'm sure there wil be others.

If you took part in the contest, please leave a comment telling about your experience of it or, better yet, write a blog post somewhere and point us to it.

Seismic inception

A month ago, some engineers at Google blogged about how they had turned a deep learning network in on itself and produced some fascinating and/or disturbing images:

One of the images produced by the team at Google. Click to see a larger version. Read more. CC-BY.

The basic recipe, which Google later open sourced, involves training a deep learning network (basically a multi-layer neural network) on some labeled images, animals maybe, then searching for matching patterns in a target image, like these clouds. If it finds something, it emphasizes it — given the data, it tries to construct an animal. Then do it again.

Or, here's how a Google programmer puts it (one of my favourite sentences ever)...

Making the "dream" images is very simple. Essentially it is just a gradient ascent process that tries to maximize the L2 norm of activations of a particular DNN layer. 

That's all! Anyway, the point is that you get utter weirdness:

OK, cool... what happens if you feed it seismic?

That was my first thought, I'm sure it was yours too. The second thing I thought, and the third, and the fourth, was: wow, this software is hard to compile. I spent an unreasonable amount of time getting caffe, the Berkeley Vision & Learning Centre's deep learning software, working. But on Friday I cracked it, so today I got to satisfy my curiosity.

The short answer is: reptiles. These weirdos were 8 levels down, which takes about 20 minutes to reach on my iMac.

Seismic data from the Virtual Seismic Atlas, courtesy of Fugro. 


Er, right... what's the point in all this?

That's a good question. It's just a bit of fun really. But it makes you wonder:

  • What if we train the network on seismic facies? I think this could be very interesting.
  • Better yet, what if we train it on geology? Probably spurious: seismic is not geology.
  • Does this mean learning networks are just dumb machines, or can they see more than us? Tough one — human vision is highly fallible. There are endless illusions to prove this. But computers only do what we tell them, at least for now. I think if we're careful what we ask for, we can use these highly non-linear data-crunching algorithms for good.
  • Are we out of a job? Definitely not. How do you think machines will know what to learn? The challenge here is to make this work, and then figure out how it can help change, or at least accelerate, our understanding of the subsurface.

This deep learning stuff — of which the University of Toronto was a major pioneer during its emergence in about 2010 — is part of the machine learning revolution that you are, like it or not, experiencing. It will take time, and it will make awful mistakes, but the indications are that machine learning will eat every analytical method for breakfast. Customer behaviour prediction, computer vision, natural language processing, all this stuff is reeling from the relatively sudden and widespread availability of inexpensive computer intelligence. 

So what are we going to do with that?

           Okay, one more. from  Paige Bailey's Twitter feed .

           Okay, one more. from Paige Bailey's Twitter feed.

Six comic books about science

Ever since reading my dad's old Tintin books late into the night as a kid, I've loved comics and graphic novels. I've never been into the usual Marvel and DC stuff — superheroes aren't my thing. But I often re-read Tintin, I think I've read every Astérix, and since moving to Canada I've been a big fan of Seth and Chester Brown.

Last year in France I bought an album of Léonard, an amusing imagining of da Vinci's exploits as an inventor... Almost but not quite about science. These six books, on the other hand, show meticulous research and a love of natural philosophy. Enjoy!

The Thrilling Adventures of Lovelace and Babbage

Sydney Padua, 2015. New York, USA: Pantheon. List price USD 28.95.

I just finished devouring this terrific book by Padua, a young Canadian animator. It's an amazing mish-mash of writing and drawing, science and story, computing and history, fiction and non-fiction. This book has gone straight into my top 10 favourite books ever. It's really, really good.

Author — Amazon — Google — Pantheon

T-Minus: The Race to the Moon

Jim Ottaviani, Zander Cannon, Kevin Cannon, 2009. GT Labs. List price USD 15.99.

Who doesn't love books about space exploration? This is a relatively short exposition, aimed primarily at kids, but is thoroughly researched and suspenseful enough for anyone. The black and white artwork bounces between the USA and USSR, visualizing this unique time in history.

Amazon — GoogleGT Labs


Jim Ottaviani, Leland Myrick, 2011. First Second Books. List price USD 19.99.

A 248-page colour biography of the great physicist, whose personality was almost as remarkable as his work. The book covers the period 1923 to 1986 — almost birth to death — and is neither overly critical of Feynman's flaws, nor hero-worshipping. Just well-researched, and skillfully told.

AmazonGoogleFirst Second.

A Wrinkle in Time

Hope Larson, Madeleine L'Engle, 2012. New York, USA: Farrar, Straus & Giroux. List price USD 19.99

A graphic adaptation of L'Engle's young adult novel, first published in 1963. The story is pretty wacky, and the science is far from literal, so perhaps not for all tastes — but if you or your kids enjoy Doctor Who and Red Dwarf, then I predict you'll enjoy this. Warning: sentimental in places.

Amazon — MacmillanAuthor 

Destination Moon and Explorers on the Moon

Hergé, 1953, 1954. Tournai, Belgium: Casterman (English: 1959, Methuen). List price USD 24.95.

These remarkable books show what Hergé was capable of imagining — and drawing — at his peak. The iconic ligne claire artwork depicts space travel and lunar exploration over a decade before Apollo. There is the usual espionage subplot and Thom(p)son-based humour, but it's the story that thrills.


What about you? Have you read anything good lately?

Coding to tell stories

Last week, I was in Calgary on family business, but I took an afternoon to host a 'private beta' for a short course that I am creating for geoscience computing. I invited about twelve familiar faces who would be provide gentle and constuctive feedback. In the end, thirteen geophysicists turned up, seven of whom I hadn't met before. So much for familiarity.

I spent about two and half hours stepping through the basics of the Python programming language, which I consider essential material — getting set up with Python via Enthought Canopy, basic syntax, and so on. In the last hour of the afternoon, I steamed through a number of geoscientific examples to showcase exercises for this would-be course. 

Here are three that went over well. Next week, I'll reveal the code for making these images. I might even have a go at converting some of my teaching materials from IPython Notebook to HTML:

To plot a wavelet

The Ricker wavelet is a simple analytic function that is used throughout seismology. This curvaceous waveform is easily described by a single variable, the dominant frequency of its many contituents frequencies. Every geophysicist and their cat should know how to plot one: 

To make a wedge

Once you can build a wavelet, the next step is to make that wavelet interact with the earth. The convolution of the wavelet with this 3-layer impedance model yields a synthetic seismogram suitable for calibrating seismic signals to subtle stratigraphic geometries. Every interpreter should know how to build a wedge, with site-specific estimates of wavelet shape and impedance contrasts. Wedge models are important in all instances of dipping and truncated layers at or below the limit of seismic resolution. So basically they are useful all of the time. 

To make a 3D viewer

The capacity of Python to create stunning graphical displays with merely a few (thoughtful) lines of code seemed to resonate with people. But make no mistake, it is not easy to wade through the hundreds of function arguments to access this power and richness. It takes practice. It appears to me that practicing and training to search for and then read documentation, is the bridge that carries people from the mundane to the empowered.

This dry-run suggested to me that there are at least two markets for training here. One is a place for showing what's possible — "Here's what we can do, now let’s go and build it". The other, more arduous path is the coaching, support, and resources to motivate students through the hard graft that follows. The former is centered on problem solving, the latter is on problem finding, where the work and creativity and sweat is. 

Would you take this course? What would you want to learn? What problem would you bring to solve?