latest news

Distractions and amusements, with a sandwich and coffee.

Poetry is just the evidence of life. If your life is burning well, poetry is just the ash
•
• burn something
• more quotes

Country flags are pretty colorful and some are even pretty.

Instead of drawing the flag in a traditional way (yawn...), I wanted to draw it purely based on the color proportions in the flag (yay!). There are lots of ways to do this, such as stacked bars, but I decided to go with concentric circles. A few examples are shown below.

Once flags are drawn this way, they can be grouped by similarity in the color proportions.

Check out the posters or read about the method below.

Or, download my country flag color catalog to run your own analysis.

To determine the proportions of colors in each flag, I started with the collection of all country flags in SVG from Wikipedia. The flags are conveniently named using the countries' ISO 3166-2 code. At the time of this project (21 Mar 2017), this repository contained 312 flags, of which I used 256.

I originally wanted to use the flag-icon-css collection, but ran into problems with it. It had flags in only either 1 × 1 or 4 × 3 aspect ratio, which distorted and clipped many flags. Many flags were also inaccurately drawn and had inconsistent use of colors. For example, in Turkey's flag the red inside the white crescent was slightly different than elsewhere in the flag.

I converted the SVG files to high resolution PNG (2,560 pixels in width) and sampled the colors in each flag, keeping only those colors that occupied at least 0.01% of the flag. I apply this cutoff to avoid blends between colors due to anti-aliasing applied in the conversion. When drawing the flags as circles, I only use colors that occupy at least 1% of the flag—this impacts flags that have detailed emblems, such as Belize. I apply some rounding off of the proportions and colors with the same proportion are ordered so that lighter colors (by Lab luminance) are in the center of the circle.

There are various ways to represent the proportions of the flag colors as concentric rings—in other words, to use symbols of different size to encode area.

The accurate way is to have the area of the ring be proportional to the area of the color on the map. The inaccurate way is to encode the area by the the width of the ring. These two cases are the `k=0.5` and `k=1` columns in the figure below, where `k` is the power in `r = a^k` by which the radius of the ring, `r`, is scaled relative to the area, `a`. A perceptual mapping using `k=0.57` has been suggested by some.

My goal here is not to encode the proportions so that they can be read off quantitatively. To find a value of `k`, I drew some flags and looked at their concentric ring representation. For example, with `k=0.57` the Nigerian flag's white center is too large for my eye while for `k=1` it is definitely too small. I liked the proportions for `k=1/\sqrt{2}` but wasn't happy with the fact that flags like France's, which have colors in equal areas, didn't have equal width rings.

In the end I decided on a hybrid approach in which the out radius of color `i` whose area is `a_i` is `r_i = a_i^k + \sum_{j=0}^{i-1} a_j^k` where the colors are sorted so that `a_{i-1} \le a_i`. If I use `k=0.25`, I manage to have flags like France have equal width rings but flags like Nigeria in which the proportions are not equal are closer to the encoding with `k=1/\sqrt{2}`. In this hybrid approach smaller areas, such as the white in the map of Turkey, are exaggerated. Notice that here `k` plays a slightly different role—it's used as the power for each color individually, `\sum a^k`, rather than their sum, `\left({\sum a}\right)^k`.

For the purists this choice of encoding might appear as the crime of the worst sort, representing neither correct (`k=0.5`) nor the conventionally incorrect encoding associated with `k=1`. Think of it this way—I know what rule I'm breaking.

The similarity between two flags is calculated by forming an intersection between the radii positions of the concentric rings of the flags.

For each intersection, the similarity of colors is determined using `\Delta E`, which is the Euclidian distance of the colors in LCH space. I placed less emphasis on luminance and chroma in the similarity calculation by fist transforming the coordinates to `(\sqrt L,\sqrt C, H)`) before calculating color differences. The similarity score is $$ S = \sum \frac{\Delta r}{\sqrt{\Delta E}} $$

Color pairs with `\Delta E < \Delta E_{min} = 5` are considered the same and have an effective `\Delta E = 1`.

I explored different cutoffs and combinations of transforming the color coordinates. This process was informed based on how the order of the flags looked to me.

I decided to start the order with Tonga, since it had the highest average similarity score to all other flags in some of my trials. The flag that is most different from other flags, as measured by the average similarity score, is Israel.

I couldn't find a list of colors in the flags of countries, so I provide my analysis here. Every country's SVG flag was converted into a 2,560 × 1,920 PNG file (4,915,200 pixels). Colors that occupied at least 0.01% of the pixels are listed in their HEX format, followed by the number of pixels they occupy. The fraction of the flag covered by sampled colors is also shown.

DOWNLOAD #code img_pixels sampled_pixels fraction_sampled_pixels hex:pixels,hex:pixels,... ... cm 4366506 4364514 0.999544 FCD116:1513103,007A5E:1456071,CE1126:1395340 cn 4369920 4364756 0.998818 DE2910:4260992,FFDE00:103764 co 4364800 4364800 1.000000 FCD116:2183680,003893:1090560,CE1126:1090560 ...

DOWNLOAD #code1 code2 similarity_score ad ae 0.0108360578506763 ad af 0.0288161214840692 ad ag 0.0510922121861494 ad ai 0.42746294322472 ... zw ye 0.473278765746989 zw yt 0.238101673130705 zw za 0.810589244643825 zw zm 0.573265751850587

Inference creates a mathematical model of the datageneration process to formalize understanding or test a hypothesis about how the system behaves. Prediction aims at forecasting unobserved outcomes or future behavior. Typically we want to do both and know how biological processes work and what will happen next. Inference and ML are complementary in pointing us to biologically meaningful conclusions.

Statistics asks us to choose a model that incorporates our knowledge of the system, and ML requires us to choose a predictive algorithm by relying on its empirical capabilities. Justification for an inference model typically rests on whether we feel it adequately captures the essence of the system. The choice of pattern-learning algorithms often depends on measures of past performance in similar scenarios.

Bzdok, D., Krzywinski, M. & Altman, N. (2018) Points of Significance: Statistics vs machine learning. Nature Methods 15:233–234.

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: supervised methods. Nature Methods 15:5–6.

Celebrate `\pi` Day (March 14th) and go to brand new places. Together with Jake Lever, this year we shrink the world and play with road maps.

Streets are seamlessly streets from across the world. Finally, a halva shop on the same block!

Intriguing and personal patterns of urban development for each city appear in the Boonies, Burbs and Boutiques series.

No color—just lines. Lines from Marrakesh, Prague, Istanbul, Nice and other destinations for the mind and the heart.

The art is featured in the Pi City on the Scientific American SA Visual blog.

Check out art from previous years: 2013 `\pi` Day and 2014 `\pi` Day, 2015 `\pi` Day, 2016 `\pi` Day and 2017 `\pi` Day.

We examine two very common supervised machine learning methods: linear support vector machines (SVM) and k-nearest neighbors (kNN).

SVM is often less computationally demanding than kNN and is easier to interpret, but it can identify only a limited set of patterns. On the other hand, kNN can find very complex patterns, but its output is more challenging to interpret.

We illustrate SVM using a data set in which points fall into two categories, which are separated in SVM by a straight line "margin". SVM can be tuned using a parameter that influences the width and location of the margin, permitting points to fall within the margin or on the wrong side of the margin. We then show how kNN relaxes explicit boundary definitions, such as the straight line in SVM, and how kNN too can be tuned to create more robust classification.

Bzdok, D., Krzywinski, M. & Altman, N. (2018) Points of Significance: Machine learning: a primer. Nature Methods 15:5–6.

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.

In a Nature graphics blog article, I present my process behind designing the stark black-and-white Nature 10 cover.

Nature 10, 18 December 2017

In this primer, we focus on essential ML principles— a modeling strategy to let the data speak for themselves, to the extent possible.

The benefits of ML arise from its use of a large number of tuning parameters or weights, which control the algorithm’s complexity and are estimated from the data using numerical optimization. Often ML algorithms are motivated by heuristics such as models of interacting neurons or natural evolution—even if the underlying mechanism of the biological system being studied is substantially different. The utility of ML algorithms is typically assessed empirically by how well extracted patterns generalize to new observations.

We present a data scenario in which we fit to a model with 5 predictors using polynomials and show what to expect from ML when noise and sample size vary. We also demonstrate the consequences of excluding an important predictor or including a spurious one.

Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.

Just in time for the season, I've simulated a snow-pile of snowflakes based on the Gravner-Griffeath model.

The work is described as a wintertime tale in In Silico Flurries: Computing a world of snow and co-authored with Jake Lever in the Scientific American SA Blog.

Gravner, J. & Griffeath, D. (2007) Modeling Snow Crystal Growth II: A mesoscopic lattice map with plausible dynamics.