Machine Learning & Data Science

Linear Digressions - Old to New

Elle run

mornings

Linear Digressions

Machine learning and particle physics go together like peanut butter and jelly--but this is a relatively new development.  

For many decades, physicists looked through their fairly large datasets using the laws of physics to guide their exploration; that tradition continues today, but as ever-larger datasets get made, machine learning becomes a more tractable way to deal with the deluge.  

With this in mind, ATLAS (one of the major experiments at CERN, the European Center for Nuclear Research and home laboratory of the recently discovered Higgs boson) ran a machine learning contest over the summer, to see what advances could be found from opening up the dataset to non-physicists.  

The results were impressive--physicists are smart folks, but there’s clearly lots of advances yet to make as machine learning and physics learn from one another.  And who knows--maybe more Nobel prizes to win as well!

https://www.kaggle.com/c/higgs-boson

Hunting for the Higgs

In the 1850s, there were a lot of things we didn’t know yet: how to create an airplane, how to split an atom, or how to control the spread of a common but deadly disease: cholera.  

When a cholera outbreak in London killed scores of people, a doctor named John Snow used it as a chance to study whether the cause might be very small organisms that were spreading through the water supply (the prevailing theory at the time was miasma, or “bad air”).  By tracing the geography of all the deaths from the outbreak, Snow was practicing elementary data science--and stumbled upon one of history’s most famous outliers.  

In this episode, we’ll tell you more about this single data point, a case of cholera that cracked the case wide open for Snow and provided critical validation for the germ theory of disease.

http://en.wikipedia.org/wiki/1854_Broad_Street_cholera_outbreak

How Outliers Helped Defeat Cholera

Computers are capable of many impressive feats, but making you laugh is usually not one of them. Or could it be? This episode will talk about a custom-built machine learning algorithm that searches through text and writes jokes based on what it finds. 

The jokes are formulaic: they're all of the form "I like my X like I like my Y: Z" where X and Y are nouns, and Z is an adjective that can describe both X and Y. For (dumb) example, "I like my men like I like my coffee: steaming hot." The joke is funny when ZX and ZY are both very common phrases, but X and Y are rarely seen together. 

So, given a large enough corpus of text, the algorithm looks for triplets of words that fit this description and writes jokes based on them. Are the jokes funny? You be the judge...

http://homepages.inf.ed.ac.uk/s0894589/petrovic13unsupervised.pdf

Computers Try to Tell Jokes

Baseball is characterized by a high level of equality between teams; even the best teams might only have 55% win percentages (contrast this with college football, where teams go undefeated pretty regularly). In this regime, where 2 outcomes (Giants win/Giants lose) are approximately equally likely, we can model the win/loss chances with a binomial distribution. 

Using the binomial distribution, we can calculate an interesting little result: what's the chance of the world series going to only 4 games? 5? 6? All the way to 7? Then we can compare to decades' worth of world series data, to see how well the data follows the binomial assumption. 

The result tells us a lot about sports psychology--if each game is independent of the others, 4/5/6/7 game series are equally likely. The data shows a different trend: 4 and 7 game series are significantly more likely than 5 or 6. There's a powerful psychological effect at play--everybody loves the 7th game of the world series, or a good sweep. And it turns out that the baseball teams, whether they intend it or not, oblige our love of short (4) and long (7) world series!

http://blog.philbirnbaum.com/2007/06/winning-world-series-in-x-games.html

Stats of World Series Streaks

A true classic topic in ML: Facial recognition is very high-dimensional, meaning that each picture can have millions of pixels, each of which can be a single feature. It's computationally expensive to deal with all these features, and invites overfitting problems. PCA (principal components analysis) is a classic dimensionality reduction tool that compresses these many dimensions into the few that contain the most variation in the data, and those principal components are often then fed into a classic ML algorithm like and SVM. 

One of the best thing about eigenfaces is the great example code that you can find in sklearn--you can distinguish pictures of world leaders yourself in just a few minutes!

http://scikit-learn.org/stable/auto_examples/applications/face_recognition.html

Facial Recognition with Eigenfaces

Ever tried to visualize a cluster of data points in 40 dimensions?  Or even 4, for that matter?  We prefer to stick to 2, or maybe 3 if we're feeling well-caffeinated.  The t-SNE algorithm is one of the best tools on the market for doing dimensionality reduction when you have clustering in mind.

Relevant links:
https://www.youtube.com/watch?v=RJVL80Gg3lA

t-SNE: Reduce Your Dimensions, Keep Your Clusters

This week, we're doing a crash course in recurrent neural networks--what the structural pieces are that make a neural net recurrent, how that structure helps RNNs solve certain time series problems, and the importance of forgetfulness in RNNs.  

Relevant links: 
http://colah.github.io/posts/2015-08-Understanding-LSTMs/

Recurrent Neural Nets

This week: everybody's favorite WWII-era classifier metric!  But it's not just for winning wars, it's a fantastic go-to metric for all your classifier quality needs.

Rock the ROC Curve

It seems like every day there's more and more machine learning problems that involve learning on text data, but text itself makes for fairly lousy inputs to machine learning algorithms.  That's why there are text vectorization algorithms, which re-format text data so it's ready for using for machine learning.  In this episode, we'll go over some of the most common and useful ways to preprocess text data for machine learning.

Feature Processing for Text Analytics

Word2Vec is probably the go-to algorithm for vectorizing text data these days.  Which makes sense, because it is wicked cool.  Word2Vec has it all: neural networks, skip-grams and bag-of-words implementations, a multiclass classifier that gets swapped out for a binary classifier, made-up dummy words, and a model that isn't actually used to predict anything (usually).  And all that's before we get to the part about how Word2Vec allows you to do algebra with text.  Seriously, this stuff is cool.

Word2Vec

As more and more data gets collected seemingly every day, and data scientists use that data for modeling, the technical limits associated with machine learning on big datasets keep getting pushed back.  This week is a first-hand case study in using scikit-learn (a popular python machine learning library) on multi-terabyte datasets, which is something that Katie does a lot for her day job at Civis Analytics.  There are a lot of considerations for doing something like this--cloud computing, artful use of parallelization, considerations of model complexity, and the computational demands of training vs. prediction, to name just a few.

Things You Learn When Building Models for Big Data

Kullback Leibler divergence, or KL divergence, is a measure of information loss when you try to approximate one distribution with another distribution.  It comes to us originally from information theory, but today underpins other, more machine-learning-focused algorithms like t-SNE.  And boy oh boy can it be tough to explain.  But we're trying our hardest in this episode!

KL Divergence

Neural networks are complex models with many parameters and can be prone to overfitting.  There's a surprisingly simple way to guard against this: randomly destroy connections between hidden units, also known as dropout.  It seems counterintuitive that undermining the structural integrity of the neural net makes it robust against overfitting, but in the world of neural nets, weirdness is just how things go sometimes.

Relevant links: https://www.cs.toronto.edu/~hinton/absps/JMLRdropout.pdf

Neural Net Dropout

Open source software is ubiquitous throughout data science, and enables the work of nearly every data scientist in some way or another. Open source projects, however, are disproportionately maintained by a small number of individuals, some of whom are institutionally supported, but many of whom do this maintenance on a purely volunteer basis. The health of the data science ecosystem depends on the support of open source projects, on an individual and institutional level.

https://hdsr.mitpress.mit.edu/pub/xsrt4zs2/release/2

The Data Science Open Source Ecosystem

This is a re-release of an episode that originally ran in October 2019.

If you’re trying to manage a project that serves up analytics data for a few very distinct uses, you’d be wise to consider having custom solutions for each use case that are optimized for the needs and constraints of that use cases. You also wouldn’t be YouTube, which found themselves with this problem (gigantic data needs and several very different use cases of what they needed to do with that data) and went a different way: they built one analytics data system to serve them all. Procella, the system they built, is the topic of our episode today: by deconstructing the system, we dig into the four motivating uses of this system, the complexity they had to introduce to service all four uses simultaneously, and the impressive engineering that has to go into building something that “just works.”

Procella: YouTube's super-system for analytics data storage

A few weeks ago, we put out a call for data scientists interested in issues of race and racism, or people studying how those topics can be studied with data science methods, should get in touch to come talk to our audience about their work. This week we’re excited to bring on Todd Hendricks, Bay Area data scientist and a volunteer who reached out to tell us about his studies with the Stanford Open Policing dataset.

A Data Science Take on Open Policing Data

The data science and artificial intelligence community has made amazing strides in the past few years to algorithmically automate portions of the healthcare process. This episode looks at two computer vision algorithms, one that diagnoses diabetic retinopathy and another that classifies liver cancer, and asks the question—are patients now getting better care, and achieving better outcomes, with these algorithms in the mix? The answer isn’t no, exactly, but it’s not a resounding yes, because these algorithms interact with a very complex system (the healthcare system) and other shortcomings of that system are proving hard to automate away. Getting a faster diagnosis from an image might not be an improvement if the image is now harder to capture (because of strict data quality requirements associated with the algorithm that wouldn’t stop a human doing the same job). Likewise, an algorithm getting a prediction mostly correct might not be an overall benefit if it introduces more dramatic failures when the prediction happens to be wrong. For every data scientist whose work is deployed into some kind of product, and is being used to solve real-world problems, these papers underscore how important and difficult it is to consider all the context around those problems.

A Reality Check on AI-Driven Medical Assistants

All good things must come to an end, including this podcast. This is the last episode we plan to release, and it doesn’t cover data science—it’s mostly reminiscing, thanking our wonderful audience (that’s you!), and marveling at how this thing that started out as a side project grew into a huge part of our lives for over 5 years.

It’s been a ride, and a real pleasure and privilege to talk to you each week. Thanks, best wishes, and good night!

—Katie and Ben