Wolfram Blog News, views, & ideas from the front lines at Wolfram Research 2014-11-26T22:44:08Z http://blog.wolfram.com/feed/atom/ WordPress Rita Crook <![CDATA[Benedict Cumberbatch Can Charm Humans, but Can He Fool a Computer?]]> http://blog.internal.wolfram.com/?p=23036 2014-11-26T20:13:52Z 2014-11-26T15:42:00Z The Imitation Game, a movie portraying Alan Turing’s life (who would have celebrated his 100th birthday on Mathematica‘s 23rd birthday—read our blog post), was released this week, which we’ve been looking forward to. Turing machines were one of the focal points of the movie, and we launched a prize in 2007 to determine whether the 2,3 Turing machine was universal.

So of course, Cumberbatch’s promotional video where he impersonates other beloved actors reached us as well, which got me wondering, could Mathematica‘s machine learning capabilities recognize his voice, or could he fool a computer too?

I personally can’t stop myself from chuckling uncontrollably while watching his impressions, however, I wanted to look beyond the entertainment factor.

So I started wondering: Is he actually good at doing these impressions? Or are we all just charmed by his persona?

Is my psyche just being fooled by the meta-language, perhaps? If we take the data of pure voices, does he actually cut the mustard in matching these?

In order to determine the answer, 10 years ago we would have needed to stroll the streets and play audio snippets to 300 people from the James Bond movies, The Shining, Batman, and Cumberbatch’s impression snippets—then survey whether those people were fooled.

But no need, if you have your Mathematica handy!

With Mathematica‘s machine learning capabilities, it’s possible to classify sample voice snippets easily, which means we can determine whether Benedict’s impressions would be able to fool a computer. So I set myself the challenge of building a decent enough database of voice samples, plus I took snippets from each of Benedict’s impression attempts, and I let Mathematica do its magic.

We built a path to each person’s snippet database, which Mathematica exported for analysis:

Classify sample voice snippets

We imported all of the real voices:

Import real voices

The classifier was trained simply by providing the associated real voices to Classify; in the interest of speed, a pre-trained ClassifierFunction was loaded from cfActorWDX.wdx:

Classifier was trained simply by providing the associated real voices to Classify

My audio database needed to include snippets of Benedict’s own voice, snippets of the impersonated actors’ own voices, and the impressions from Cumberbatch. The sources for the training were the following: Alan Rickman, Christopher Walken, Jack Nicholson, John Malkovich, Michael Caine, Owen Wilson, Sean Connery, Tom Hiddleston, and Benedict Cumberbatch. I used a total of 560 snippets, but of course, the more data used, the more reliable the results. The snippets needed to be as “clean” as possible (no laughter, music, chatter, etc. in the background).

These all needed to be exactly the same length (3.00 seconds), and we made sure all snippets were the same length by using this function in the Wolfram Language:

Making sure snippets are all same length

Some weren’t single-channel audio files, so we needed to exclude this factor as an additional feature to optimize our results during the export stage:

Excluding single-channel audio files

Thanks go to Martin Hadley and Jon McLoone for the code.

Drum-roll… time for the verdict!

I have to break everyone’s heart now, and I’m not sure I want to be the one to do it… so I will “blame” Mathematica, because machine learning could indeed mostly tell the difference between the actors’ real voices and the impressions (bar two).

As the results below reveal, Mathematica provides 97–100% confidence on the impressions tested:

Mathematica provides 97-100% confidence on the impressions tested

For most impressions, there is a very small reported probability of any classification other than Benedict Cumberbatch or Alan Rickman.



It might be worth noting that Rickman, Connery, and Wilson all have a slow rhythm to their speech, with many pauses (especially noticeable in the snippets I used), which could have confused the algorithm.

Sad Benedict Cumberbatch

Now it’s time to be grown up about this, and not hold it against Benedict. He is still a beloved charmer, after all.

My admiration for him lives on, and I look forward to seeing him in The Imitation Game!

Download the accompanying code for this blog post as a Computable Document Format (CDF) file.

Wolfram Blog http://blog.wolfram.com/ <![CDATA[Deck the Halls with Lines of Coding]]> http://blog.internal.wolfram.com/?p=22970 2014-11-24T18:56:40Z 2014-11-24T18:56:40Z Cyber savings header

Thanksgiving is just around the corner, and that means you only have five weeks left to knock out your holiday shopping. Never fear, Wolfram is delivering amazing deals to customers across the globe, including North and South America, Australia, and parts of Asia and Africa to inspire a whole new year of computational creativity.

Give the perfect gift to your high school or college student with 20% off of Mathematica Student Edition, or treat yourself to the same discount on Mathematica Home Edition.

Mathematica discounts

Or, for the engineering hobbyist or recreational system designer on your list, get 20% off of SystemModeler Student and Home Editions.

SystemModeler discount

To make things extra merry, we are offering a Cyber Monday deal. US and Canada shoppers will see these discounts increase from 20% to 25% off, for even greater savings!

This holiday season, you can also get an extra three months of Wolfram|Alpha Pro free with a one-year subscription! View Step-by-step solutions for your math and chemistry queries, upload and analyze your own data and files, get extended computation time, and interact with plots and graphs—as well as receive access to Wolfram Problem Generator, where you’ll have unlimited practice problems with Step-by-step solutions.

Wolfram|Alpha Pro discount

Want to give the gift of W|A Pro to a special someone else instead? Send it in the form of an electronic gift card.

These special offers are available until December 31. If after all that you still have more items on your holiday list to check off, visit our online store for exclusive Wolfram merchandise! Happy Holidays!

Matthias Odisio <![CDATA[Removing Haze from a Color Photo Image Using the Near Infrared with the Wolfram Language]]> http://blog.internal.wolfram.com/?p=22862 2014-11-25T22:11:16Z 2014-11-21T19:31:34Z For most of us, taking bad pictures is incredibly easy. Band-Aid or remedy, digital post-processing can involve altering the photographed scene itself. Say you’re trekking through the mountains taking photos of the horizon, or you’re walking down the street and catch a beautiful perspective of the city, or it’s finally the right time to put the new, expensive phone camera to good use and capture the magic of this riverside… Just why do all the pictures look so bad? They’re all foggy! It’s not that you’re a bad photographer—OK, maybe you are—but that you’ve stumbled on a characteristic problem in outdoor photography: haze.

What is haze? Technically, haze is scattered light, photons bumped around by the molecules in the air and deprived of their original color, which they got by bouncing off the objects you are trying to see. The problem gets worse with distance: the more the light has to travel, the more it gets scattered around, and the more the scene takes that foggy appearance.

What can we do? What can possibly help our poor photographer? Science, of course.

Wolfram recently attended and sponsored the 2014 IEEE International Conference on Image Processing (ICIP), which ended October 30 in Paris. It was a good occasion to review the previous years’ best papers at the conference, and we noticed an interesting take on the haze problem proposed by Chen Feng, Shaojie Zhuo, Xiaopeng Zhang, Liang Shen, and Sabine Süsstrunk [1]. Let’s give their method a try and implement their “dehazing” algorithm.

The core idea behind the paper is to leverage the different susceptibilities of the light being scattered, which depend on the wavelength of the light. Light with a larger wavelength, such as red light, is more likely to travel around the dust, the smog, and all the other particles present in the air than shorter wavelength colors, like green or blue. Therefore, the red channel in an image carries better information about the non-hazy content of the scene.

But what if we could go even further? What prevents us from using the part of the spectrum slightly beyond the visible light? Nothing really—save for the fact we need an infrared camera.

Provided we are well equipped, we can then use the four channels of data (near infrared, red, green, and blue) to estimate the haze color and distribution and proceed to remove it from our image.

RGB, IR removal

In order to get some sensible assessments, we need a sound model of how an image is formed. In a general haze model, the content of each pixel is composed of two parts:

  • The light reflected by the objects in the scene (which will be called J)
  • The light scattered by the sky (A)

It is a good approximation to say that the “color of the air” A is constant for a specific place and time, while the “real color” J is different for each pixel. Depending on the amount of air the light had to travel through, a fraction (t) of the real color is transmitted to the camera, and the remaining portion (1-t) is replaced by scattered light.

We can summarize these concepts in a single haze equation:

Single haze equation

We need to determine J, t, and A. Let’s first estimate the global air-light color A. For a moment we will assume that portions of the image are extremely hazed (no transmission, i.e. t = 0). Then we can estimate the color A simply from the pixel values of those extremely hazed regions.

On the image below, a mouse click yields A = HTML box #425261.

Mouse-click color yield

However, our assumption that the transmission is zero in the haziest regions is clearly not verified, as we can always distinguish distant objects through the haze. This means that for images where haze is never intense, it is not possible to pick A with a click of the mouse, and we have to resort to some image processing to see how we can produce a solid estimation for images with all types of haze.

Let’s say first that it has proven difficult to obtain good dehazing results on our example images when reproducing the ICIP paper’s method for estimating the air-light color. As an alternative method, we estimate the air light color using the concept of dark channel.

The so-called dark channel prior is based on the observation that among natural images, it is almost always the case that within the vicinity of each pixel, one of the three channels (red, green, or blue) is much darker than the others, mainly because of the presence of shadows, dark surfaces, and colorful objects.

If for every pixel at least one channel must be naturally dark, we can assume that where this condition does not hold is due to the presence of scattered light—that is, the hazed region we’re looking for. So we look for a good estimation for A intersecting the brightest pixels of our images (maximum haze or illumination) within the region defined by a high value in the dark channel (highest haze).

Highest value in the dark channel

We extract the positions of the brightest pixels in the dark channel images, extract the corresponding pixel values in the hazed image, and finally cluster these pixel values:

Cluster pixel values

The selected pixels marked in red below will be clustered; here they all belong to a single region, but it may not be the case on other images:

Single region

We are looking for the cluster with the highest average luminance:

Cluster with highest average luminance

This is our estimate of the air-light color:

Estimate of air-light color

Looking once more at the equation (1), we’ve made some progress, because we are only left with computing the transmission t and the haze-free pixel value J for each pixel:

Computing the transmission t and the haze-free pixel value J

Since we choose an optimization approach to solve this problem, we first compute coarse estimates, t0 and J0, that will serve as initial conditions for our optimization system.

On to finding a coarse estimate for the transmission t0. Here’s the trick and an assumption: If we assume the transmission does not change too much within a small region of the image (that we are calling Ω), we can think of t0 to be locally constant. Dividing both sides of equation (1) by A and applying the local minimum operator min both on the color channels and the pixels in each region Ω yields:

Coarse estimate for the transmission t0

But Formula in code is exactly the definition of the dark channel of the haze-free image J and, since Ak is a positive number, we infer that this term of the equation is practically zero everywhere, given our prior assumption that natural images have at least one almost zero channel in the pixels of any region. Using this simplification yields:

Yield of simplification

This is the t0 image. The darker the image area, the hazier it is assumed to be:

The darker the image area, the hazier it is assumed to be

Now the real transmission map cannot be that “blocky.” We’ll take care of that in a second. In the ICIP 2013 paper, there is another clever process to make sure we keep a small amount of haze so that the dehazed image still looks natural. This step involves information from the near-infrared image; we describe this step in the companion notebook that you can download at the bottom of this post. Here is an updated transmission map estimate after this step:

Updated transmission map estimate

To further refine this estimate by removing these unwanted block artifacts, we apply a technique named guided filtering. It is beyond the scope of the blog post to describe the details of a guided filter. Let’s just say that here, the guided filtering of the transmission map t0 using the original hazed image as a guide allows us, by jointly processing both the filtered image and the guided image, to realign the gradient of t0 with the gradient of the hazed image—a desired property that was lost due to the blocking artifacts. The function ImageGuidedFilter is defined in the companion notebook a the end of this post.

Guided filtering

As too much dehazing would not look realistic, and too little dehazing would look too, well, hazed, we adjust the transmission map t0 by stretching it to run from 0.1 to 0.95:

Update transmission map

Thanks to our estimates for the air-light color A and the transmission map t0, another manipulation of equation (1) gives us the estimate for the dehazed image J0:

Estimate for dehazed image J0

Estimate for the dehazed image J0
Estimate for the dehazed image J0

You can compare with the original image just by positioning your mouse on top of the graphic:

It’s a good start, but a flat subtraction may be too harsh for certain areas of the image or introduce some undesirable artifacts. In this last part, we will use some optimization techniques to try to reconcile this gap and ask for the help of the infrared image to keep a higher level of detail even in the most hazed region.

The key is in the always useful Bayes’ rule for inference. The question we are asking ourselves here is which pair of t and J is the most likely to produce the observed images IRGB and INIR (the near-infrared image).

In the language of probability, we want to calculate the joint distribution
Joint distribution

Using the Bayes’ theorem, we rewrite it as:


And simplify it assuming that the transmission map t and the reflectance map J are uncorrelated, so their joint probability is simply the product of their individual ones:

Joint probability is simply the product of their individual ones

In order to write this in a form that can be optimized, we now assume that each probability term is distributed according to:

Probability formula

That is, it peaks in correspondence with the “best candidate” x-tilde. This allows us to exploit one of the properties of the exponential function—e-ae-be-c… = e-(a+b+c+…)—and, provided that the addends in the exponent are all positive, to move from the maximization of a probability to the minimization of a polynomial.

We are now left with the task of finding the “best candidate” for each term, so let’s dig a bit into their individual meaning guided by our knowledge of the problem.

The first term is the probability to have a given RGB image given specific t and J. As we are working within the framework of equation (1)—the haze model IRGB = Jt + A(1 – t)—the natural choice is to pick:

||IRGBJt + A(1 – t) ||

The second term relates the color image to the infrared image. We want to leverage the infrared channel for details about the underlying structure, because it is in the infrared image that the small variations are less susceptible to being hidden by haze. We do this by establishing a relationship between the gradients (the 2D derivatives) of the infrared image and the reconstructed image:

||▽J – ▽INIR||

This relation should take into account the distance between the scene element and the camera—being more important for higher distances. Therefore we multiply it by a coefficient inversely related to the transmission:

Multiply it by a coefficient inversely related to the transmission

The last two terms are the transmission and reflection map prior probabilities. This corresponds to what we expect to be the most likely values for each pixel before any observation. Since we don’t have any information in this regard, a safe bet is to assume them equal to a constant, and since we don’t care about which constant, we just say that their derivative is zero everywhere, so the corresponding terms are simply:




Putting all these terms together brings us to the final minimization problem:

Final minimization problem

Where the regularization coefficients λ1,2,3 and the exponents α and β are taken from the ICTP paper.

To resolve this problem, we can insert the initial condition t0 and J0, move a bit around, and see if we are doing better. If that is the case, we can then use the new images (let’s call them t1 and J1) for a second step and calculate t2 and J2. After many iterations, when we feel the new images are not much better than those of the previous step, we stop and extract the final result.

This new image J tends to be slightly darker than the original one; in the paper, a technique called tone mapping is applied to correct for this effect, where the channel values are rescaled in a nonlinear fashion to adjust the illumination:

V’ = Vγ

During our experiments, we found instead that we were better off applying the tone mapping first, as it helped during the optimization.

To help us find the correct value for the exponent γ, we can look at the difference between the low haze—that is, high transmission—parts of the original image IRGB and the reflectance map J0. We want to find a value for γ that makes this difference the smallest possible and adjust J0 accordingly:

Value for gamma
Value for gamma

We now implement a simplified version of the steepest descent algorithm to solve the optimization problem of equation (6). The function IterativeSolver is defined in the companion notebook a the end of this post.


When that optimization is done, our final best guess for the amount of haze in the image is:

Best guess amount of haze

And finally, you can see the unhazed result below. To compare it with the original, hazy image, just position your mouse on top of the graphics:

Unhazed result

We encourage you to download the companion CDF notebook to engage deeper in dehazing experiments.

Let’s now leave the peaceful mountains and the award-winning dehazing method from ICIP 2013 and move to Paris, where ICIP 2014 just took place. Wolfram colleagues staffing our booth at the conference confirmed that dehazing (and air pollution) is still an active research topic. Attending such conferences has proven to be excellent opportunities to demonstrate how the Wolfram Language and Mathematica 10 can facilitate research in image processing, from investigation and prototyping to deployment. And we love to interact with experts so we can continue to develop the Wolfram Language in the right direction.

Download this post as a Computable Document Format (CDF) file.


[1] C. Feng, S. Zhuo, X. Zhang, L. Shen, and S. Süsstrunk. “Near-Infrared Guided Color Image Dehazing,” IEEE International Conference on Image Processing, Melbourne, Australia, September 2013 (ICIP 2013).

[2] K. He, J. Sun, X. Tang. “Single Image Haze Removal Using Dark Channel Prior,” IEEE Conference on Computer Vision and Pattern Recognition, Miami, Florida, June 2009 (CVPR’09).

Images taken from:

[3] L. Schaul, C. Fredembach, and S. Süsstrunk. “Color Image Dehazing Using the Near-Infrared,” IEEE International Conference on Image Processing, Cairo, Egypt, November 2009 (ICIP’09).

Emily Suess <![CDATA[2014 Wolfram Innovator Award Winners]]> http://blog.internal.wolfram.com/?p=22822 2014-11-20T16:54:48Z 2014-11-20T16:54:48Z Now in its fourth year, the Wolfram Innovator Awards are an established tradition and one of our favorite parts of the annual Wolfram Technology Conference. This year, Stephen Wolfram presented seven individuals with the award. Join us in celebrating the innovative ways the winners are using Wolfram technologies to advance their industries and fields of research.

Wolfram Innovator Award
Candidates are nominated by Wolfram employees and evaluated by a panel of experts to determine the winners. We are excited to announce the US recipients of the 2014 Innovator Awards:

Prof. Richard J. GaylordProf. Richard J. Gaylord
Professor of Materials Science and Engineering Emeritus, University of Illinois


Mark KotanchekMark Kotanchek
CEO, Evolved Analytics LLC


John MichopoulosJohn Michopoulos
Head of Computational Multiphysics Systems Laboratory, Naval Research Laboratory


Rodrigo MurtaRodrigo Murta
Retail Intelligence Manager, St Marche Supermercados


Bruce TorrenceBruce Torrence
Professor of Mathematics, Randolph-Macon College


Yves PapegayYves Papegay
Research Scientist, French National Institute for Research in Computer Science and Control


Chad SlaughterChad Slaughter
System Architect, Enova Financial

Earlier this year, European Innovator Award winners were announced at the European Wolfram Technology Conference in Frankfurt, Germany:

Dr. János KarsaiDr. János Karsai
Associate Professor, Department of Analysis, University of Szeged

Frank ScherbaumFrank Scherbaum
Professor, Institute of Earth and Environmental Sciences, University of Potsdam

Congratulations to all of our 2014 Wolfram Innovator Award winners! Read more about our deserving recipients and their accomplishments.

Wolfram Blog http://blog.wolfram.com/ <![CDATA[Fractal Fun: Tweet-a-Program Mandelbrot Code Challenge]]> http://blog.internal.wolfram.com/?p=22780 2014-11-17T16:55:24Z 2014-11-17T16:55:24Z This week Wolfram will be celebrating Benoit Mandelbrot‘s birthday and his contributions to mathematics by holding a Tweet-a-Program challenge. In honor of Mandelbrot, tweet us your favorite fractal-themed lines of Wolfram Language code. Then, as with our other challenges, we’ll use the Wolfram Language to randomly select winning tweets (along with a few of our favorites) to pin, retweet, and share with our followers. If you win, we’ll send you a free Wolfram T-shirt!

In Tweet-a-Program’s first few exciting months, we’ve already seen a number of awesome fractal examples like these:

First fractal image

Second fractal image

Third fractal image

To win, tweet your submissions to @WolframTaP by the end of the week (11:59pm PDT on Sunday, November 23). So that you don’t waste valuable code space, we don’t require a hashtag with your submissions. However, we do want you to share your code with your friends by retweeting your results with hashtag #MandelbrotWL.

We can’t wait to see what you come up with!

Christopher Carlson <![CDATA[Announcing the Winners of the 2014 One-Liner Competition]]> http://blog.internal.wolfram.com/?p=22685 2014-11-10T19:57:52Z 2014-11-10T18:56:37Z This year’s Wolfram Technology Conference once again included the One-Liner Competition, an opportunity for some of the world’s most talented Wolfram Language developers to show us the amazing things you can do with tiny pieces of Wolfram Language code.

In previous years, One-Liner submissions were allowed 140 characters and 2D typesetting constructs. This year, in the spirit of Tweet-a-Program, we limited entries to 128-character, tweetable Wolfram Language programs. That’s right: we challenged them to write a useful or entertaining program that fits in a single tweet.

And the participants rose to the occasion. Entries were blind-judged by a panel of Wolfram Research developers, who awarded two honorable mentions and first, second, and third prizes.

One honorable mention went to Michael Sollami for his “Mariner Valley Flyby,” which takes you on a flight through the terrain of the Mariner Valley on Mars. The judges were greatly impressed by the idea and the effect. Unfortunately, a small glitch in the program is visible at the start of the output, due to an error in the code. Since Michael’s submission is right up against the 128-character limit, it would have taken some clever tweaking to fix it.

Mariner Valley Flyby

Mariner Valley Flyby

An honorable mention also went to Filip Novotný for a program that rolls a “marble” in the direction that you tilt your laptop. Yeah, yeah, we’ve all seen that before; every laptop has an accelerometer these days. But… Filip’s code doesn’t use the accelerometer. Instead, it tracks the view seen by the laptop camera and infers from it the laptop’s tilt. All in 128 characters.

Filip’s entry was also awarded a dishonorable mention for the impressively dense syntax form a@@@b@c@d@e~f~g@h, which kept his entry under the 128-character limit, and kept the judges busy trying to decipher it.


Third place went to Michael Sollami for “Gödel, Escher, Batman,” an intriguing bat-like 3D form, as well as a nice play on the title of Douglas Hofstadter’s book. See if you can figure out how the 3D form derives from the intersections of three orthogonal Batman curve regions.

Gödel, Escher, Batman

Gödel, Escher, Batman

Second place went to Jesse Friedman for his “Spoonerism Generator.” Each time you evaluate his code, you get a different wacky rendering of Poe’s poem The Raven, where the first letters of words that begin with consonants are scrambled.

Incredibly, Jesse claimed one of the few 6-letter Internationalized Resource Identifiers in existence just for the competition, by finding an unusual character (R) that had not yet been snapped up. At 13 years old, Jesse is the youngest prize winner by far in the One-Liner Competition.

Spoonerism Generator

First place went to Alex Hirsbrunner for his “Boeing 767 Flight Range,” a One-Liner entry that actually does something useful. His code makes a world map showing how far a Boeing 767 can fly from the conference location. The judges were impressed by his combined use of SemanticInterpretation, to get at obscure information (the range of the aircraft), and GeoGraphics, to make a nice presentation of it. He even had enough characters to spare to include a title in the graphic, so the whole thing is self-explanatory.

Boeing 767

Thanks to all participants for entertaining us with their abundant creativity. If you have thoughts of attending the Wolfram Technology Conference in 2015, get started now honing your Wolfram Language skills for next year’s One-Liner Competition.

You can see all of this year’s One-Liner entries by downloading this notebook.

Leonardo Laguna Ruiz <![CDATA[Using Arduinos as SystemModeler Components]]> http://blog.internal.wolfram.com/?p=22374 2014-11-07T16:01:00Z 2014-11-07T16:01:00Z With the new, free ModelPlug library for Wolfram SystemModeler, you can connect Arduino boards to simulations in SystemModeler. Arduinos interface easily with input and output components, so you can integrate them into SystemModeler models, for example, to operate lights, run servos, and monitor sensors, switches, and potentiometers. With the ModelPlug library, you can freely mix hardware and software components in your simulations and use the Arduino as a data acquisition board.

If you want to follow along, you can download a trial of SystemModeler. It’s also available with a student license, or you can buy a home-use license. All hardware used in this blog post can be bought for less than $50.

Set of icons

After downloading the library from the SystemModeler Library Store, installing it as as easy as double-clicking the package and accepting the license agreement. With the library, you can connect any Firmata-capable board to SystemModeler. This includes all Arduino boards.

I’m using an Arduino Uno board in this blog post. The easiest example I can think of is blinking the LED that’s on the board.

Arduino board
Arduino Uno board with internal LED highlighted

To do this, I construct a simple model with a Boolean signal that’s transmitted to the digital pin 13 on my board, where the LED is. The LED will blink with the same duration as the pulse I send in.

Blinking an LED
The model in SystemModeler using the ModelPlug components digital pin and the Arduino board

Next up I can place an LED on a breadboard and connect it to pin 9 on my Arduino:

Placing LED on breadboard
Schematic of connecting an external LED to port 9 on an Arduino Uno Board

When I connect a sine function to the analog output on my board, the real-valued signal is converted to the voltage needed at the pin where the LED is connected.

Dimming LED
A sine wave connected to a pin dimming an LED

I can now see how the LED varies between full light and no light with the sine wave. Without a single line of “Arduino coding” on my part!

If we take a step back and look at the big picture, there are basically four different scenarios we can image: one where we connect simulated input to simulated components, one where we connect simulated input to real hardware (as I did in the previous two examples), one where we connect hardware input to simulated components, and a fourth scenario where everything is in hardware.

Simulated vs. real

Below I structured the four scenarios in a grid.

Different scenarios where the ModelPlug library can be useful

The first case, where both input and components are modeled, is readily available in an out-of-the-box installation of SystemModeler, and for many uses this is all that is needed. That’s why I didn’t highlight it in the grid. When you are connecting hardware input to hardware components, it can be controlled and facilitated from SystemModeler, and signals can be filtered and processed in SystemModeler. The most interesting scenarios, in my mind, are where some parts are simulated and some parts exist in hardware. Let’s look at some examples where the components are simulated and the input comes from hardware.

Input devices and simulated model

Here I set up my Uno board to read the values from pin 14. There I’ve connected a light-dependent resistor that will read light levels.

Light-dependent resistor
Uno board with photoresitor connected to the analog input

In my model in SystemModeler, I now connect this analog signal from pin 14 to another component.

Connect analog signal to another component

This component will take the values from the light-dependent resistor and compare them to a threshold value. If the value is greater than 0.02 and the simulation time is larger than 5, it will terminate the simulation.

Compare the light value to a threshold and see if the simulation should be terminated

If I cover the light-dependent resistor with my hand for a short while, it will get the dark, the values will be lower than 0.02, and the simulation will end.

Here’s an example where both inputs and outputs are hardware: instead of using the signal from the light-sensitive resistor to terminate the simulation, I can filter it with a lowpass filter in SystemModeler. I then scale it to turn a servo between -90° and 90°. Whenever I hold my hand over the light-dependent resistor, the servo will turn, and when I release it, the servo will turn back.

Inputs and outputs are hardware
Analog signal connected via software components to a servo

This page requires that JavaScript be enabled in your browser. Learn how »

The previous example shows you that not only can you connect “hardware to hardware” via ModelPlug, but also how to realize functionality in components defined by equations. Instead of connecting resistors and capacitors in a circuit to lowpass filter the signal, or write a program that filters the signal, I use model components to do the filtering. This enables you to prototype very quickly.

To take this further, I’m going to use a bigger software model and connect it to analog input values. I’ll use a model of an inverted pendulum pushed on a cart. The complete system, including motor, gear, and 3D components for pendulum and the cart, are software models. To this model I’ll connect signals generated by an accelerometer connected to the Arduino. In the model below, the analog signal will enter from the left. The first highpass filter will filter out stationary trends in the signal, and the lowpass filter will smooth out the signal.

Inverted pendulum system with filters

At the tip of the pendulum, I’ll connect a force component that will convert the analog signal to a force pushing the top of the pendulum.

Mechanical system with force attached
Mechanical system with force attached

Finally I connect an accelerometer to an analog pin.

Accelerometer connected to inverted pendulum
Accelerometer connected to inverted pendulum

Now when I quickly move my accelerometer, a disturbance is generated and the control system for the pendulum will have to try to adapt. Can you move it fast enough to knock the pendulum down?

This page requires that JavaScript be enabled in your browser. Learn how »

The ModelPlug library is free and available for download from the SystemModeler Library Store. It works on Mac, Windows, and Linux. Try it out and let us know what you think, in the comments or in the SystemModeler group on Wolfram Community.

Download the SystemModeler Modelica (.mo) file via compressed (.zip) file.

Vitaliy Kaurov <![CDATA[Modeling a Pandemic like Ebola with the Wolfram Language]]> http://blog.internal.wolfram.com/?p=22253 2014-11-04T20:06:40Z 2014-11-04T19:00:23Z Data is critical for an objective outlook, but bare data is not a forecast. Scientific models are necessary to predict pandemics, terrorist attacks, natural disasters, market crashes, and other complex aspects of our world. One of the tools for combating the ongoing and tragic Ebola outbreak is to make computer models of the virus’s possible spread. By understanding where and how quickly the outbreak is likely to appear, policy makers can put into place effective measures to slow transmissions and ultimately bring the epidemic to a halt. Our goal here is to show how to set up a mathematical model that depicts a global spread of a pandemic, using real-world data. The model would apply to any pandemic, but we will sometimes mention and use current Ebola outbreak data to put the simulation into perspective. The results should not be taken as a realistic quantitative projection of current Ebola pandemic.

Ebola animation

To guide us through the computational science of pandemics, I have reached out to Dr. Marco Thiel, who was already describing various Ebola models on Wolfram Community (where readers could join the open discussion). We have worked with him to code the global pandemic model below, a task made considerably easier by many of the new features recently added to the Wolfram Language. Marco is an applied mathematician with training in theoretical physics and dynamical systems. His research was featured on BBC News, and due to its applied mathematical nature, concerns very diverse subjects, from the stability of our solar system to patterns in the mating behavior of fireflies to forensic mathematics, and much more. Dealing with this diversity of real-world problems, Marco and his colleagues and students at the University of Aberdeen have made Wolfram technologies part of their daily lives. For example, the core of code for this blog entry was written by India Bruckner, a very bright young pupil from Aberdeen’s St Margaret’s School for Girls, with whom Marco had a summer project.

The current Ebola outbreak “is the deadliest, eclipsing an outbreak in 1976, the year the virus was discovered,” according to The New York Times. Its data summary as of October 27, 2014, states that there are at least 18 Ebola patients who have been treated or are being treated in Europe and America, mostly health and aid workers who contracted the virus in West Africa and traveled to their home countries for treatment. The C.D.C. reported in September that a worst-case scenario could exceed a million Ebola cases in four months. There are no FDA-approved drugs or vaccines to defend against the virus, which is fatal in 60 to 90 percent of cases and spreads via contact with infected bodily fluids. Here is the current situation in West Africa in the pandemic locus, according to the numbers from The New York Times:



Data Source: The New York Times

Vitaliy: Marco, do you think mathematical modeling can help stop pandemics?

Marco: The recent outbreak of the Ebola virus disease (EVD) has shown how quickly diseases can spread in human populations. This threat is, of course, not limited to EVD; there are many pathogens, such as various types of influenza (H5N1, H7N9, etc.) with the potential to cause a pandemic. Therefore, mathematical modeling of the transmission pathways becomes ever more important. Health officials need to make decisions as to how to counter the threat. There are a large number of scientific publications on the subject, such as the recent Science publication by Dirk Brockmann, which is available here. Professor Brockmann also produced videos to illustrate the research, which can be found on YouTube (video1, video2, video3). It would be interesting to reproduce some of the results from that paper and generally explore the subject with Mathematica.

Vitaliy: How does one set up a computational model of a spreading disease?

Marco: Detailed online models, such as GLEAMviz, are available and can be run by everyone interested in the subject. That particular model contains, just like many other similar models, three main layers: (1) an epidemic model that describes the transmission of the disease in a hypothetical, homogeneous population; (2) population data, that is, distribution of people and population densities; and (3) a mobility layer that describes how people move. I used a similar model that uses the powerful algorithms of Mathematica, its built-in databases, and its powerful data import capabilities. Also, Mathematica‘s visualization algorithms allowed for developing a new model for the spreading of diseases. Some advantages of a DIY model are that we fully control the program and can amend it to our requirements. Mathematica‘s curated data is a very useful modeling starting point and can be paired with flexible data import from external web sources. This is one application of Conrad Wolfram’s idea of publicly available computable data. Using the current capabilities of Mathematica, we can get a glimpse of what will be possible when governmental/public data becomes computable.

There are many different types of epidemic models. In what follows, we will mainly deal with the so-called Susceptible Infected Recovered (SIR) model. It models a population that consists of three compartments: susceptibles can become infected upon contact with the infected; the infected do recover at a certain rate.

Susceptible Infected Recovered (SIR) model

To model the outbreak with the Wolfram Language, we need equations describing the number of people in each of these categories as functions of time. We will first use time-discrete equations. If we suppose first that there are only three categories and no interaction between them, we could get the following:

Category 1

This means that the number, actually the percentage, of Susceptibles/Infected/Recovered at time t+1 is the same as at time t. Let’s assume that a random contact of an infected with a susceptible leads to a new infection with probability b; the probability of a random encounter is proportional to the number of susceptibles (Sus) and also to the number of infected (Inf). This assumption means that people are taken out of the compartment of the susceptibles and go into the infected category.

Category 2

Next, we assume that people recover with a probability c; the recovery is proportional to the sick people; that is, the more who are sick, the more who recover.

Category 3

We also need initial values for the percentages of people in the respective categories. Note that the “interaction terms” on the right-hand side always add up to zero, so that the overall population size does not change. If we start at initial conditions that add up to one, the population size will always stay one. This is an important feature of the model. Every person has to stay in one of the three compartments; we will take great care to make sure that this is also true for the SIR model on the network that we describe later! There is, however, some flexibility of how we can interpret the three compartments. In our final example we will, for example, consider deaths. It might seem logical to think that these “leave” our population. In order to keep our population constant, which is important for our model, we will then use a simple trick: we will interpret the last group, the Recovered (Rec), as a set that contains the truly recovered and the dead. It is a reasonable assumption that neither the dead nor the recovered infect other people, so they are inert to our model. Our simple assumption will be that a fixed percentage of people of the Rec group will be alive and the remainder will be dead. Hence, we include dead people in our model—so that they don’t actually leave the groups—and we do not consider births. This results in a constant population size.

This is a naive implementation of the SIR model, which allows you to change the parameters:

Naive implementation

We use vectors Sus, Inf, and Rec and iterate them. We will later develop a more direct implementation. Note that the parameters b and c “parametrize” many effects that are at this stage not directly modeled. For example, the infection rate b does describe the risk of infection and therefore models things like population density (high density might lead to more infections) and behavior of people (if there are many mass events, that might increase the infection probability—so does schooling!). The recovery rate c might describe things like quality of the health care system, availability of physicians, and so on. Later we will try to model some of these effects more directly.

The SIR model might not be the most suitable to describe an Ebola outbreak. It is, however, not too far off either. People get infected by contact; the Recovered category might be interpreted as holding the percentage of people who have either survived or died, if we assume that reinfection is unlikely. In different countries/circumstances, the recovery/death rate might vary substantially—something we will model later explicitly.

A more systematic way of looking at the overall behavior of the SIR model is to study the so-called parameter space. We can represent how different characteristics, like the highest number of infected or the total number of people who get infected in the course of the outbreak, depend on the parameters. The axes of the following diagram show the infection and recovery rates, and the percentage of people who contract the disease during the outbreak is color-coded:

Infection and recovery rates

This shows that for small recovery rates and large infection rates, more than 90% of people contract the disease, whereas for large recovery rates and low infection rates, the total percentage of infected is about 5%, which, in fact, equals the initial percentage of infected.

Vitaliy: To go from pure mathematical to real-world simulations, we would need data, such as populations and their geographic locations. How could data be accessed?

Marco: We will later couple different subpopulations (e.g. airports, cities, countries, etc.) and study the spreading of the disease among these. Each subpopulation is described by an SIR model. When we start coupling the subpopulations, their individual sizes will play a crucial role. Population data, like many other types of data, is built right into Mathematica, so it is quite easy to use that for our modeling.

We will use built-in data to improve our model toward the end, but for a start we could use the international network of all airports to model the transport of the disease. We first need a list of all airports and all flight connections. On the website openflights.org you will find all the data we need. I saved the file “airports.dat” and the file “routes.dat.”

Vitaliy: We could use the latest Semantic Data Import feature to interface with external data. SemanticImport can import a file semantically to return a Dataset object. Dataset and Association are new functions and represent a structured database based on a hierarchy of lists and associations. Dataset can represent not only full rectangular multidimensional arrays of data, but also arbitrary tree structures, corresponding to data with arbitrary hierarchy. It is very easy to access and process data with Dataset, and we will make use of it. A tiny fraction of airports.dat was cleaned up for more precise semantic import. All data files used here are attached at the end of this post, together with the Mathematica notebook.

Marco: Yes, indeed. SemanticImport is very powerful. In my first modeling attempt I used Import and then the new Interpreter function, both of which are very powerful, too. I then needed to do some more “cleaning up” of the data; a multistep procedure that is a common problem when you use external non-computable data. But thanks to your suggestion to use SemanticImport, I could make the code much more concise and readable:

More concise

Vitaliy: Yellow-framed entries are semantically processed as Entity:

Semantically processed

So we notice that SemanticImport automatically classified the third and fourth columns as cities and countries and converted them to Entity, which is the built-in data representation in the Wolfram Language.

Marco: We can now plot all airports worldwide.

Vitaliy: Indeed, with the new functionality GeoGraphics and its numerous options such as GeoBackground, GeoProjection, and GeoRange, we can tune up the image to a balanced representation of a massive amount of data:

The fifth column in airports is a three-letter IATA airport code. We will need this airport identification code because it identifies connecting routes between airports in the second dataset. Not all data entries have it; for example, here are the last 100 cases:

Last 100 cases

Some of these are also false because they have numbers. We will clean the data by removing entries with no IATA valid code. Here are the original entries:

Original entries

We will retain cleaned-up rows in the total amount that follows:

Cleaned up rows

Marco: Next, we create a list of rules for all airport IDs and their coordinates:

List of rules

Vitaliy: We used the Dispatch function, which generates an optimized dispatch table representation of a list of rules and will never affect results that are obtained, but may make the application of long lists of rules much faster.

Marco: Now we can calculate the connections:

Calculate connections

Not every IATA code has geo coordinates. Let’s clean out those with missing data:

Clean out missing data

Out of a total of 67,210, we will plot just 15,000 random routes, which reflects well on the full picture:

Vitaliy: Once we have the data, how can it be integrated with mathematical models?

Marco: We need to describe the mobility pattern of the population. We will use the global air transport network to build a first model of a pandemic. We think of the flights as connections between different areas. These areas could be interpreted as “catchment areas” of the respective airports.

Vitaliy: We could make use of the Graph function to generate the network quickly and efficiently. There are many different routes between the same airports, and this would correspond to multigraph, a new Wolfram Language feature. For the sake of simplicity of a starting model, we will consider only the fact of connection between two airports, drawing a single edge if there is at least one route. We will use multigraphs later in the improved model:

Starting model

In the resulting graph, vertices are given by IATA codes.

As we can see below, there are several disconnected components that are negligible due to relatively small size. We will discard them, as they will not significantly alter any dynamics:

From our graph, we can construct the adjacency matrix:

Adjacency matrix

Marco: The (i,j)th entry of the matrix is zero if the airports i and j are not connected, and one otherwise. The coupling matrix is automatically converted into the SparseArray form to make computations and memory use more efficient. We can also use a “population size” term for each airport, that is, its catchment area. We will later try to make more reasonable guesses at the population size, but for now we will simply assume that it is the same for each airport:

Catchment area

Now comes a tricky step. We need to define the parameters for our model. The problem is that each of them “parametrizes” many effects and situations. For our model, we need:

  • The probability of infection. This is a factor that determines in our model how likely it is that a contact leads to an infection. This factor will certainly depend on many things: population density, local behavior of people, health education, and so on. To get our modeling started, we will choose the following for all airports:

Probability of infection

  • The rate of recovery. The rate of recovery will very much depend on the type of disease and the quality of the health system. It will also depend on whether everyone has access to health insurance. In countries where only a fraction of the population has access to high-quality health care, diseases will generally spread faster. For our initial modeling, we will set the following for all airports:

Rate of recovery

With these two parameters, we have the epidemic model determined. But we still need one more parameter.

  • Migration factor. It is a factor of proportionality that describes the propensity of a certain population (in the catchment area of an airport) to travel. In this model, we have taken it to be constant, but it would certainly also depend on the financial situation in that country and other factors. It describes, roughly speaking, the percentage of people in a catchment area/country who travel. We do not use a multi-agent model where the movement of individuals is described. We use a compartmental population model where, in fact, we describe percentages of the population who travel. We do (at least later in the post, for the country-based simulation) take the different population size into consideration, and in the form of the multigraph, also how many flights there are from country to country. Using the coupling matrix, we will not introduce a migration of individuals between different airports. We will first use a general migration factor (same for all), which we set to the following:

Migration factor

This is a strong assumption. We will use—at first—the same migration factor for all three categories: susceptibles, infected, and recovered. In reality, the infected, particularly in the infectious phase where they might have developed symptoms, will probably have a different mobility pattern. Also, the mobility will be different in different countries, and it will depend on the distance traveled as well. We will later choose more realistic parameters.

We next initialize our model and assume that at first there are only susceptibles in all cities and no infected or recovered:

Initialize our model

Now we should introduce 5% infected to the catchment area of the original airport.

Vitaliy: The outbreak began in Guinea in December 2013, then spread to Liberia and Sierra Leone. According to The New York Times, the very first case treated outside West Africa was a nurse from Spain infected in September while treating a missionary who had been infected in Sierra Leone. She was then flown to a hospital in Madrid. There are a few airports in Guinea:

Airports in Guinea

We choose the CKY code for the Conakry International Airport in the capital and the largest city and compute its index:

Compute index

Marco: Before we write down the SIR equations with the coupling terms, we introduce two objects:

Introduce objects

This is the total number of (potential) passengers at each airport. We condense the coupling matrix into a smaller list that for each airport only contains a list of connected airports:

Condense coupling matrix

This is very useful, because the coupling matrix is sparse and sumind will speed up the calculation dramatically. Now we can write down the SIR equations:

Write down SIR equations

The coupling terms are highlighted in orange. Basically, we calculate a weighted average over the number of people in each compartment for all neighboring airports. There are many other types of coupling that one could choose.

Next we iterate the equations and calculate the time course:

Iterate equations

To get a first impression, we can plot the S, I, and R curves for some 200 airports:

Plotting curves for 200 airports

Next, we calculate the maximal number of sick people at any of the airports:

Maximal number of sick people at airport

We will generate a list of airport coordinates ordered as vertices of our graph:

List of airport coordinates

We can now look at a couple of frames:

Time progresses by columns from top to bottom and between the columns from left to right. Here and below in a similar simulation graphic, the color codes the number of infected people.

Note that there are three main regions: Europe, which gets infected first, then the Americas and Asia. This kind of spreading is related to the network structure. We can try to visualize this with Graph.

Vitaliy: The clustering algorithms of Graph can help us see which continents are major facilitators in the spread of the pandemic via airways. We’ll make use of built-in data to get the list of continents:

Make use of built-in data

We’ll discard Antarctica and build a database denoting which airport codes belong to which continent:

This is a function that can tell what continent a particular code belongs to:

Code and continent

For instance:

Find continent

There are many differently sized communities in our network:

Communities in network

Communities are clusters of many flights joining airports of the same community compared to few flights joining airports of different communities. Not to overcrowd the plot with labels, let’s label only those communities whose size is greater than 60, based on the largest fraction of airport codes belonging to the labeling continent:

Length greater than 60

Now we can visualize the network structure and see how major hubs enable transportation to smaller ones. In this particular plot, colors just differentiate between different communities and do not represent infected people:

Marco: The three dominating communities become apparent. The graphic also shows via which of the main communities smaller groups are infected. This could have implications for decisions about preventative measures. We can now generate a graph that is similar to one presented in Dirk Brockmann’s paper:

Generate graph

Vitaliy: Again, time progresses by columns from top to bottom and between the columns from left to right. We used “RadialEmbedding” for GraphLayout and set “RootVertex” to airport. I see that the pandemic proceeds from the vertex that is central outward through the hierarchy layers. Marco, could you please explain what this means?

Marco: In the center of the representation, we find the airport where the outbreak started. The layer around it represents all the airports that can be reached with direct connections; they are the first to be hit. The next layer is the airports that can be reached with one connection flight; they are the next to be infected, and so on. This shows that the structure of the network, rather than geographical distance, is important.

Vitaliy: How can we make this model a bit more realistic?

Marco: We have studied some simple, prototypical cases of a model that locally describes the outbreak of a disease with an SIR model and then couples many SIR models based on connections of some transport network. However, there are many problems that still would have to be addressed:

  1. The probability of transmission will depend on many factors, for example, the health system and the population density in a region.
  2. The recovery rate will also depend on many factors, for example, the health system.
  3. Not all possible links (roads/flight trajectories) will be taken with equal probability. Published papers suggest that there is a power law: the longer the distance, the less likely someone is to travel.
  4. The migration/traveling rate will depend on the categories susceptibles, infected, recovered; sick people are less likely to travel.

Also we might want to be able to model different attempts by governments to control the outbreak. So let’s try to address some of these issues and generate another, global model of an Ebola outbreak. If we wanted to model all cities and all airports worldwide, that would probably be asking too much from an ordinary laptop. Based on the sections above, it should be clear how to extend the model, but I want to highlight some further approaches that might be useful.

Our model will represent all (or most) countries worldwide. The connections among them will be modeled based on the respective flight connections. To start with, we will collect some data for our model.

As before, we will import the airport data and the flight connections:

Import airport data

The SemanticImport function directly recognizes the countries that the airports belong to. We can easily construct a list of airports-to-country data:

Airports-to-country data

This time we construct the graph using the Graph command. As we want to study the connections between countries, we substitute the airports by the respective countries:

Substitute airports

Vitaliy: Are you saying that the graph connections indicate country adjacencies rather than airport connections?

Marco: The connections are generated by the flights; that is, the flights’ paths are the edges. They go from airport to airport, but we are actually interested in modeling the connections between countries. Therefore, we substitute (identify) the airports with their respective countries. Formally, you can think of this as constructing the network of all airport connections and then identifying all airports (nodes) that correspond to the same country. We could say that the mathematical term for that is vertex identification.

It turns out that data on some airports is missing; most of the missing data is from very small airports. We delete all airports for which the country is not known:

Delete all airports for which country is unknown

Last but not least, we construct the coupling/adjacency matrix:

Coupling matrix

We can also get the list of all countries that will form part of our model:

List of countries

It is clear that if the population density is higher, that might lead to a higher infection rate; this is, of course, quite an assumption, because what is important is mostly local population densities. If the country is very large, but everybody lives in one city, the effective density is much higher.

To take the population into consideration, we will need data on the population size and density for all countries:

Data on population

Now we will build a simple “model” for how the population density might (partially) determine the infection rate. In the first model we used an infection rate of ρ=0.2, and that gave “reasonable” results, in the sense that we saw a pattern of the spreading that was as expected. We want to extend the model, without completely modifying the parameter range. So it is “reasonable”—as a first guess—to choose the parameters for the extended model in the same range. As a matter of fact, we observe that the crucial thing is the ratio of λ and ρ. So basically we are saying that we want to start from about the same ratio as in the first simulation.

To modify the infection rate with respect to the population density, we look first at the histogram of the population densities:

Histogram of population densities

We can also calculate the median:

Calculate median

We will make the assumption that the infection rate will increase with the population density. For the “median population density” we wish to obtain an infection rate of 0.2. We will make the bold assumption that the relationship is given by the following chart:

Infection rate vs. population density

We can calculate the infection rates for all countries:

Infection rates

Of course, this “model” for the dependence of the infection rate on the population density is very crude. I invite the reader to improve that model!

Vitaliy: It would be really interesting to know some typical laws, based on real data, about how infection rate varies with population density. It is great to have this law as a changeable condition so that readers can try their versions of it and observe change in the simulations. Percolation phenomena perhaps should be considered, where there is a sudden steep change in infection rate when population density reaches some critical value. There could also be some saturation behavior due to a city’s infrastructure. But I am just guessing here. Would you please share your own take on this?

Marco: If there are more people in a confined space, the infection rate should increase monotonically. It also appears to make sense that the infection rate does not increase linearly, but that its derivative decreases (perhaps even saturates). The infection rate should somewhat depend on the distance to the next neighbor, that is, go with the square root of the density, which would lead to an exponent of 0.5. But then the movement of individual people would be blocked by their neighbors, effectively decreasing the slope, so we would need an exponent smaller than 0.5. Well that was at least my thought. The article “The Scaling of Contact Rates with Population Density for the Infectious Disease Models” by Hu et al., 2013, shows that our assumptions are reasonable—at least for this crude model.

To estimate the recovery rate, we will make another very daring assumption. We will assume that the health system is better in countries where the life expectancy is higher. To build this submodel, we will make use of the life expectancy data built into Mathematica:

Built-in life expectancy data

We can visualize the distribution of life expectancies for all countries in a histogram:

Visualize distribution of life expectancies

Note that there are a few countries that are lacking this data. We set their life expectancies to a typical 70; this will not influence the results, because these countries (very small countries, i.e. islands, and Antarctica) will not play a role in our model.

The median life expectancy:

Median life expectancy

Now, like before, we will formulate our assumption that life expectancy is a proxy (approximate indicator) for the quality of the health system, which is a proxy for the recovery rate. This assumption is again quite crude, because cases in relatively rich countries like Spain and the US suggest that recovery rates might not be substantially higher in wealthy countries:

Recovery rate vs. life expectancy

As with the infection rates, we calculate the recovery rates for all countries:

Calculate recovery rates

We can also represent what that means for all countries; the median values are marked in white and blue, while the parity of the infection rate and the recovery rate is indicated by the black dashed line. The background color represents the percentage of people who contract the disease at any time during the outbreak, as derived from the SIR model (see above):

SIR model

Vitaliy: A typical modeling process entails running the simulation many times, trying to understand the system behavior and the influence of parameter values. Our particular values will result in some common-sense final outcome for what we can imagine will happen fighting a very tough but survivable pandemic. Countries with weak economies will suffer the highest damage, while their counterparts should count smaller losses. What is important here, I think, is to understand how that contrast depends on mobility network topology, demographics, and other real-world factors. We also should expect that these complex components and nonlinearity could result in counterintuitive behavior sometimes favoring the weak and damaging the strong.

Marco, could you please explain in greater detail the plot above?

Marco: The plot above shows the main two parameters of our model: infection versus recovery rate. Each point denotes a country—hovering over them will display the name. The white and green lines indicate the median values. They cut the diagram into four areas. The upper-left box (high infection rate/low recovery rate) contains countries that have the most challenging conditions. There are countries such as Sierra Leone, Nigeria, and Bangladesh in it. In the lower right (low infection rate/high recovery rate) are the countries that have the best prospects of containing the disease: United States, Canada, and Sweden. The black dashed line indicates a critical separator: above the line there are countries where the infection rate outpaces the recovery rate. In those countries an outbreak is very likely. Below the line the recovery rate dominates the infection rate, and the chances are that the disease can be contained. Note that these are just “base parameters”; they do not take the network into consideration yet. Note that countries below the dashed line can block the spread of the disease. If there are enough of them, that again will decrease the number of casualties.

Vitaliy: As we have mentioned before, our simulation is significantly exaggerated to amplify and see clearly various effects of a pandemic. This is why we intentionally shifted the system above the black dashed line. Our readers are welcome to consider more realistic parameter values.

Marco: To address another flaw of our first models, we will assume that susceptibles, infected, and recovered travel at different rates. There is one additional point to consider. If we set the migration/movement rates too high, regional differences will average out! If susceptibles, infected, and recovered all travel (fast enough), the populations of different countries mix over the periods of the simulation, so that we just see an average. In reality, the percentages of people traveling will be relatively low over the simulation period, so they will nearly be zero. In order to see the disease spreading, we will set relatively high rates for the movement of the infected; we will ignore the traveling of the healthy and recovered. This is not a bad assumption, because they do not contribute to the spreading of the disease anyway:

Setting rates

Let’s clear the variables to make sure that we start with a clean slate:

Clear variables

As before, we initiate all countries to have purely susceptible populations:

Purely susceptible populations

The recent outbreak started in Sierra Leone, so we will need to find out which vertex models that country:

Vertex modeling Sierra Leone

Next we set a low number of infected people in that country:

Lower number infected

We then iterate as before:

Iterate as before

Now we can actually iterate:

Actual iteration

This is a first test of whether the simulation worked—we plot time series for all countries:

Plot time series for all countries

Note that at the end of the simulation, most of the lines have saturated, indicating that the outbreak has come to a standstill. Here is also a time course of the total sick and deaths over the course of the outbreak:

Time course

We can now visualize the global spread of Ebola with the help of a few time frames:

Visualize global spread

Note that the number of dead people is extremely high in all of our simulations. About three billion dead people at the end would be an absolutely devastating outbreak, most likely much, much larger than what we are currently observing. This might be due to the choice of parameters; the ratio of the infection rate to the recovery rate might be different. On the other hand, we model an outbreak where the systems are relatively “inert,” meaning that they do not take efficient countermeasures. If we were to model that, we would need to decrease the infection rate over time, due to the actions of the governments.

The model is indeed just a conceptual model at this point. The reader can change certain features, that is, close down airports, and see what the effect is. He/she can change infection and recovery rates and see how that influences the outcome. There are many things you can play with. (A really bad infection rate/recovery rate ratio might be important if Ebola should mutate and become airborne, for example.) We have not (yet) tried to choose the parameters so that they optimally describe the current outbreak.

Vitaliy: I run our model many times for different parameter values. For example, in an opposite to the above situation, when recovery is greater than infection rate and the initial values are smaller:

Recovery greater than infection

We get much lower infected and sick numbers:

Total sick vs. total deaths

Also note different peak behavior, which tells us that pandemics can relapse—and with greater potential casualties. The New York Times data shows such relapses in the section “How Does This Compare to Past Outbreaks?” Also the C.D.C. projection and current stage of pandemic are probably an approximate exponential increase matching the very beginning (left side) of peaks in the graphs.

Can we calibrate our model by real data, for instance, time scale and absolute values of infection and recovery rates?

Marco: There are many papers that study effects of, say, population density on infection rates, such as the one mentioned above by Hu et al. We can use models or observed data to calibrate our model. Mathematica offers many great tools to fit parametric curves (e.g. from models) to data. This can help to improve our model. Regarding the time scale, we need to stress that we have not yet attached a time unit to each iteration step. We can, however, use observed data, for example, for the current outbreak, and try to find parameters ρ, λ, and μ so that the number of casualties/infected and its change over time are correctly described by the model.

Vitaliy: The ListLinePlot and GeoRegionValuePlot both show behavior different from the simplified original model we started from. Now we have countries that get outbreaks and others that seem resistant to the pandemic. This looks more realistic to me. Could you please share your thoughts about this and perhaps compare these results to Dirk Brockmann’s paper?

Marco: The most important factor is that the high migration rates in the first models lead to a mixing of the populations so that at the end, all populations have the same rates of people in each of the three categories. Just like in Dirk Brockmann’s paper, we see that the network is very important for the spreading of the disease. The distance on the graph is important. This is particularly so because countries for which the recovery rate is higher than the infection rate basically block the disease from spreading.

Vitaliy: Marco, I thank you very much for your inspiring insights and time. Do you have some ideas how we could improve the model further or any other concluding remarks for our readers?

Marco: It is very easy to come up with more ideas about how to improve the model: we could include further transport mechanisms, that is, streets/cars, boats, train networks, etc. Each would have an effect on how people move. We could use mobile phone data or other data to better model how people actually move. Also, the SIR model does not take into consideration the incubation time, that is, the time from infection until you show the first symptoms. Our model describes populations in compartments; numbers of infected etc. are given in percentages. This is only a valid approximation if the number of infected is large enough (like a concentration given as a real number is a good idea if there are lots of molecules!), but particularly at the beginning of an outbreak, when there are few infected, other types of model such as a multi-agent model might be more realistic. The model we have presented should only be taken as a first step; more effects can be included and their relevance can be studied. By trial and error, we can try to describe the real outbreak better and better. But we also need to be careful: the more effects we include in our model, the more parameters we will get. That can lead to all sorts of problems when we need to estimate them.

We may note that the simulation does not correctly describe the cases in the US and Europe. Currently we know that 18 people were brought there on purpose. They are not at all part of the natural propagation of the disease. Also, we are speaking about “percents or more” of a population, we are not even discussing individual cases. Epidemiologically speaking, there is nothing going on in the US or Spain at the moment regarding Ebola. Sierra Leone and other countries in Africa do get into a regime where the model is “more valid.”

The model is not specific to the Ebola epidemic. There is a lot of estimation and guesswork, and we choose the parameters so that the epidemic spreading becomes clear, but the model shows a potential order of the spreading: first mainly western/central African countries, then Europe, then the US. This is very reasonable and coincides with much more complicated models. Of course, there are many other effects, such as that the spreading between neighboring countries in Africa will probably not be via flights, but rather via very local transport, that is, people crossing borders by foot/car or similar. Our model is certainly conceptual in so far as it only considers one way of transport, namely flights, which is not the full picture, of course, but it does show that there are dramatic differences between countries to be expected. In fact, we would expect a much lower percentage of casualties in Europe and the US than in the countries in Africa were everything originated. The model shows that the highest probability of spreading is between neighboring African countries, which is what larger models predict as well. It also says that certain countries in Europe, such as Germany/UK/France etc. are more at risk than others because of their flight connections. The US would be less at risk than the European countries, that is, it would get significant numbers of infected later. All of that seems to be qualitatively quite correct. Australia and Greenland would get the disease very late, or not at all, again in agreement with our model (well, at least if rho ~ lambda ~ 0.1).

In that sense I suppose that it is more realistic to chose rho equal or smaller than lambda. That will mostly limit the spreading to some African countries with low probabilities of infections in Europe and the US. Even though the “global average” infection rate could be even lower than the recovery rate, at least in some countries, due to population density and (lack of) quality of health systems, the local infection rate might be higher than the recovery rate. Also, the number of infected will obviously be lower than what our model predicts, as our model does not take countermeasures by governments or the WHO into consideration. That is easy to model by reducing the infection rate over time, more in rich countries, less in poor countries.

The predictions have to be seen as probabilities or risk that certain countries experience large natural outbreaks, that is, our model does not consider individual cases or cases that are transported to hospitals on purpose. Also, the infection rate is certainly very different for people who work in hospitals with patients. That means that the probability for a nurse contracting Ebola is much higher than for the average person, and of course we do not model that. Also the precise propagation of an epidemic depends on many “random” factors, which cannot be reasonably taken into account in this or any other model. One can introduce random factors and run the model various times to make predictions about likely scenarios, or probabilities. In that sense, our model performs actually qualitatively very well.

The conceptual nature of the model allows us to look at different scenarios: What if infection rate versus recovery rate changes? What if there is more mobility, that is, mu changes? What if we also use local transport etc.? We did not try to fit the parameters/network etc. to optimally describe the Ebola outbreak; rather, we provide the basic model to develop several scenarios and to understand the basic ingredients for this type of modeling.

Vitaliy: I once again thank Marco and our readers for being part of this interview and invite everyone to participate in the related Wolfram Community thread where Marco and others discuss and share related ideas, code, and in-depth studies.

Download this post in a compressed (.zip) file composed of the CDF (.cdf) and accompanying data (.dat) files.

Wolfram Blog http://blog.wolfram.com/ <![CDATA[Wolfram Technology Conference 2014]]> http://blog.internal.wolfram.com/?p=22349 2014-10-31T15:49:35Z 2014-10-31T15:49:35Z Wolfram technology users from around the world gathered in Champaign, Illinois, our headquarters, October 21–24, for another successful Wolfram Technology Conference. Attendees got access to the latest information about our emerging technologies and gained insights from colleagues who shared innovative ways of using Wolfram technologies.


The conference kicked off with a keynote by Stephen Wolfram, and then rolled right into the other 125 scheduled talks. Also featured were a connected devices playground, “Meet-Ups,” social/networking events, small group meetings, roundtable discussions, and tours of Wolfram and the Blue Waters supercomputer at the University of Illinois.

Keynote by Stephen Wolfram

Attendees came from 20 countries spanning 6 continents, and 80 of them were at the conference for the first time. Topics covered every industry/specialty, including, but not limited to, engineering, finance, computer science, physics, astronomy, mathematics, image processing, robotics, and even quilting.

Wolfram developers and attendees alike gave talks and workshops, illustrating not only how the technology was built, but also some of the diverse applications our users have implemented in their professional fields. Their inspiring and innovative projects ranged from a corporate search engine using the Wolfram Language to stitch-coding and movie color maps; they demonstrated everything from integrating Mathematica and the Unity Game Engine and creating online courses to high-frequency training, connected devices, and embedding code.

Numerous possibilities with our technology were showcased, demonstrating how devices can be integrated with Mathematica, Wolfram|Alpha, and the Wolfram Language to perform real-time data analysis. In the connected devices playground, multiple devices were set up for attendees to interact with, including two Raspberry Pis with breadboards and LEDs, an 8Cube from hypnocube, two Tinkerforge Weather Stations hooked up to Raspberry Pis, a Sphero 2.0, an Intel Edison, an Arduino, and Electric Imps connected to a temperature/humidity sensor and a heart rate monitor.


We also brought back a favorite—the One-Liner competition. Attendees were tasked with thinking up “amazing things” using just one tweet of Wolfram Language code. Stay tuned for the announcement of the results and the winners of the 2014 Innovator Awards!

Wolfram Blog http://blog.wolfram.com/ <![CDATA[Calling All Goblins: Tweet-a-Program Halloween Code Challenge]]> http://blog.internal.wolfram.com/?p=22220 2014-10-27T18:25:33Z 2014-10-27T18:25:33Z Halloween is quickly approaching, and to help you gear up for trick-or-treating, costume parties, and pumpkin carving, we’re issuing another Tweet-a-Program Code Challenge! This time, instead of spaceships and planets, we want you to tweet us your spookiest Halloween-themed lines of Wolfram Language code. We’ll then use the Wolfram Language to randomly select three winning tweets (and a few of our favorites) to pin, retweet, and share with our followers. Winners will also be awarded a free Wolfram T-shirt!

Take some inspiration from these examples, while you come up with your creepy codes:

Tweet-a-Program skull

Tweet-a-Program Halloween

In order to win, your Halloween-themed submissions must be tweeted to us before the clock strikes midnight, Pacific time, on All Hallows’ Eve (11:59pm PDT, Friday, October 31). So you don’t waste needed code space, no hashtag is required with your original submission, but we encourage you to share your results by retweeting them with hashtag #SpookyCode.

We’re excited about the possibilities—just keep an eye on your creation and make sure that it doesn’t… come alive!