view updates

Distractions and amusements, with a sandwich and coffee.

listen; there's a hell of a good universe next door: let's go.
•
• go there
• more quotes

Some of the images in this writeup are part of Ana Swanson's Wonk Blot post How a dog sees a rainbow, and 12 other images that explain how we see color at the Washington Post.

In an audience of 8 men and 8 women, chances are 50% that at least one has some degree of color blindness^{1,2}. When encoding information or designing content, use colors that is color-blind safe.

^{1}About 8% of males and 0.5% of females are affected with some kind of color blindness in populations of European descent (wikipedia, Worldwide prevalence of red-green color deficiency, JOSAA). The rate for other races is lower Asians and Africans is lower (Caucasian Boys Show Highest Prevalence of Color Blindness Among Preschoolers, AAO).

^{2}The probability that among `N=8` men and `N=8` women at least one person is affected by color blindness is `P(men,women) = P(8,8) = 1 - (1-0.08)^8 * (1-0.005)^8 = 0.51`. For `N=34` (i.e., 68 people in total), this probability is `P(34,34)=0.95`. Because the rate of color blindness in women is so low, for most groups of mixed gender we can approximate the probability by only counting the men. For example, in a group of 17 women the probability that at least one of them is color blind is `P(0,17) = 0.082`, which is the same probability as for 1 man, `P(1,0)`.

Color Oracle is a good and free color blindness simulator for Windows, Mac and Linux.

You can download the RGB transformation table for deuteranopia, protanopia and tritanopia. It is available for all (r,g,b) colors in steps of 5 in each of the channels. The mapping for all other RGB colors can be interpolated.

Transformation for *all* 16.8 million RGB colors (interpolated from the table above) are also available independently for each type of color blindness:
deuteranopia,
protanopia,
and
tritanopia.

The normal human eye is a 3-channel color detector^{3}. There are three types of photoreceptors, each sensitive to a different part of the spectrum. Their combined response to a given wavelength produces a unique response that is the basis of the perception of color.

^{3}Compared to hearing, the color vision is a primitive detector. While we can hear thousands of distinct frequencies and process them simultaneously, we have only three independent color inputs. While the ear can distinguish pure tones from complex sounds that have multiple frequencies the eye is relatively unsophisticated in separating a color sensation into its three constituent primary stimuli.

People with color blindness have one of the photo receptor groups either reduced in number or entirely missing. With only two groups of photoreceptors, the perception of hue is drastically altered.

For example, in *deuteranopia*, the most common type of color blindness, the medium (M) wavelength photoreceptors are reduced in number or missing. This results in the loss of perceived difference between reds and greens because only one group of photoreceptors (L) are sensitive to the wavelengths of these colors. The spectrum appears to be split into two hues along the blue-green boundary (see figure below).

Visible light is in the range of 390-700 nm. The exact definition of the upper limit varies, with some sources giving as high as 760 nm. Shorter wavelengths are absorbed by the cornea (<295nm) and lens (315-390nm). Some near infrared light also reaches the retina (760-1400nm).

The opposite condition to color blindness exists too—tetrachromacy. In this case, an individual has an extra type of color receptor which improves discrimination in the red part of the spectrum. While the anatomy of their retina can be described, how true tetrachromats subjectively perceive color is unknown. And, perhaps, even unknowable.

Tetrachromacy is common in other animals, such as fish (e.g. goldfish, zebrafish) and birds (e.g. finch, starling). The dimensionality of the perceived color space isn't necessarily proportional to the number of different receptors. If the signal from 3 color receptors are combined by the brain and each processor has a weighted response to a broad range of wavelengths, then a color can be modeled by a point in 3-dimensional space, in which the receptors are the axes. This system can perceive a large number of colors.

In the extreme case where the receptors respond to a very narrow range, of which none overlap with the other, a color is one of three points in a 1-dimensional space. This sytem can perceive only 3 colors.

For example, although the mantis shrimp has 12 different color receptors, the receptors work independently, their color discrimination is poorer than ours.

If you use Color Oracle to transform your screen colors to simulate color blindness, you can see that none of the equivalent swatches in one kind of color blindness are equivalent in another. This is particularly interesting when applied to a duotone image which is drawn using equivalent colors. In the figure below^{4}, a row of Mr. Spocks disappears (or is difficult to see) to people with color blindness.

^{4}In tribute to Leonard Nimoy, 1931–2015

To people with color blindness, some colors appear the same. This equivalence can be used to identify distinct colors which are unique to those with normal and color blind vision.

The seven colors (and black) in the figure below are perceived as reasonably distinct by both normal and color blind individuals. The table on the left is reproduced from Nature Method's Points of View: Color blindness by Bang Wong.

For more tips about designing with color blindness in mind, see Color Universal Design (CUD) — How to make figures and presentations that are friendly to Colorblind people.

The figure below shows the mapping of different colors to six different grades of each of the two hues seen by deuteranopes. It offers more distinct options than the 7-color palette above.

Even more color choices for color blindess, including colors that map onto greys.

If you're looking to encode quantitative information, I suggest using the subset of Brewer palettes that are safe for color blindess (e.g. pink-yellow-green, brown-blue-green).

*With four parameters I can fit an elephant and with five I can make him wiggle his trunk. —John von Neumann.*

By increasing the complexity of a model, it is easy to make it fit to data perfectly. Does this mean that the model is perfectly suitable? No.

When a model has a relatively large number of parameters, it is likely to be influenced by the noise in the data, which varies across observations, as much as any underlying trend, which remains the same. Such a model is overfitted—it matches training data well but does not generalize to new observations.

We discuss the use of training, validation and testing data sets and how they can be used, with methods such as cross-validation, to avoid overfitting.

Altman, N. & Krzywinski, M. (2016) Points of Significance: Model Selection and Overfitting. *Nature Methods* **13**:703-704.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Classifier evaluation. *Nature Methods* **13**:603-604.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. *Nature Methods* **13**:541-542.

*It is important to understand both what a classification metric expresses and what it hides.*

We examine various metrics use to assess the performance of a classifier. We show that a single metric is insufficient to capture performance—for any metric, a variety of scenarios yield the same value.

We also discuss ROC and AUC curves and how their interpretation changes based on class balance.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Classifier evaluation. *Nature Methods* **13**:603-604.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. *Nature Methods* **13**:541-542.

Today is the day and it's hardly an approximation. In fact, `22/7` is 20% more accurate of a representation of `\pi` than `3.14`!

Time to celebrate, graphically. This year I do so with perfect packing of circles that embody the approximation.

By warping the circle by 8% along one axis, we can create a shape whose ratio of circumference to diameter, taken as twice the average radius, is 22/7.

If you prefer something more accurate, check out art from previous `\pi` days: 2013 `\pi` Day and 2014 `\pi` Day, 2015 `\pi` Day, and 2016 `\pi` Day.

*Regression can be used on categorical responses to estimate probabilities and to classify.*

The next column in our series on regression deals with how to classify categorical data.

We show how linear regression can be used for classification and demonstrate that it can be unreliable in the presence of outliers. Using a logistic regression, which fits a linear model to the log odds ratio, improves robustness.

Logistic regression is solved numerically and in most cases, the maximum-likelihood estimates are unique and optimal. However, when the classes are perfectly separable, the numerical approach fails because there is an infinite number of solutions.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. *Nature Methods* **13**:541-542.

Altman, N. & Krzywinski, M. (2016) Points of Significance: Regression diagnostics? *Nature Methods* **13**:385-386.

Altman, N. & Krzywinski, M. (2015) Points of Significance: Multiple Linear Regression *Nature Methods* **12**:1103-1104.

Altman, N. & Krzywinski, M. (2015) Points of significance: Simple Linear Regression *Nature Methods* **12**:999-1000.