Mad about you, orchestrally.feel the vibe, feel the terror, feel the painmore quotes

# pi day: beautiful

UCD Computational and Molecular Biology Symposium, Dublin, Ireland. 1-2 Dec 2016.

# visualization + design

The 2016 Pi Day art imagines the digits of Pi as physical masses collapsing under gravity and is featured in the articles The Gravity of Pi and The Boundless Beauty of Pi at the Scientific American SA Visual blog.

# $\pi$ Approximation Day Art Posters

2016 $\pi$ approximation day
2016 $\pi$ day
2015 $\pi$ day
2014 $\pi$ approx day
2014 $\pi$ day
2014 $\pi$ day
2013 $\pi$ day
Circular $\pi$ art

The never-repeating digits of $\pi$ can be approximated by $22/7 = 3.142857$ to within 0.04%. These pages artistically and mathematically explore rational approximations to $\pi$. This 22/7 ratio is celebrated each year on July 22nd. If you like hand waving or back-of-envelope mathematics, this day is for you: $\pi$ approximation day!

Want more math + art? Discover the Accidental Similarity Number. Find humor in my poster of the first 2,000 4s of $\pi$.

## getting it mostly right

Curiously, the 22/7 rational approximation of $\pi$ is more accurate (to within 0.04%) than using the first three digits $3.14$, which are accurate to 0.05%.

It seems that $\pi$ Approximation Day is 20% more accurate (verify on Wolfram Alpha)! And therefore definitely worth celebrating. $$\frac{(\pi-3.14)-(22/7-\pi)}{\pi-3.14} = 0.206$$

## art of $\pi$ rational approximation

The poster shows the accuracy of 10,000 rational approximations of $\pi$ for each $m/n$ and $m=1...10000$. Read about the details of the method.

Accuracy of 10,000 rational approximations of $\pi$ for each $m/n$ and $m=1...10000$. (zoom, BUY ARTWORK)

## art of $\pi$-approximate packing

These posters show warped circles, which embody the 22/7 approximation of $\pi$, using a retro 1970's color scheme. Read about the details of the method.

Packing of warped circles that embody the 22/7 approximation of $\pi$. Here warped circles are clipped by perfect circles. The color scheme is 1970's retro. (zoom, BUY ARTWORK)
VIEW ALL

# Classifier Evaluation

Fri 05-08-2016

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.

Nature Methods Points of Significance column: Classifier Evaluation (read)

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

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

# Happy 2016 $\pi$ Approximation, roughly speaking

Sun 24-07-2016

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.

# Logistic Regression

Wed 13-07-2016

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.

Nature Methods Points of Significance column: Logistic regression? (read)

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.

Altman, N. & Krzywinski, M. (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.

# Visualizing Clonal Evolution in Cancer

Thu 02-06-2016

Genomic instability is one of the defining characteristics of cancer and within a tumor, which is an ever-evolving population of cells, there are many genomes. Mutations accumulate and propagate to create subpopulations and these groups of cells, called clones, may respond differently to treatment.

It is now possible to sequence individual cells within a tumor to create a profile of genomes. This profile changes with time, both in the kinds of mutation that are found and in their proportion in the overall population.

Ways to present temporal and phylogenetic evolution of clones in cancer. M Krzywinski (2016) Molecular Cell 62:652-656. (read)

Clone evolution diagrams visualize these data. These diagrams can be qualitative, showing only trends, or quantitative, showing temporal and population changes to scale. In this Molecular Cell forum article I provide guidelines for drawing these diagrams, focusing with how to use color and navigational elements, such as grids, to clarify the relationships between clones.

How to draw clone evolution diagrams better. M Krzywinski (2016) Molecular Cell xxx:xxx-xxx. (read)

I'd like to thank Maia Smith and Cydney Nielsen for assistance in preparing some of the figures in the paper.

Krzywinski, M. (2016) Visualizing Clonal Evolution in Cancer. Mol Cell 62:652-656.

# Binning High-Resolution Data

Wed 01-06-2016

Limitations in print resolution and visual acuity impose limits on data density and detail.

Your printer can print at 1,200 or 2,400 dots per inch. At reading distance, your reader can resolve about 200–300 lines per inch. This large gap—how finely we can print and how well we can see—can create problems when we don't take visual acuity into account.

Nature Methods Points of View column: Binning high-resolution data. (read)

The column provides some guidelines—particularly relevant when showing whole-genome data, where the scale of elements of interest such as genes is below the visual acuity limit—for binning data so that they are represented by elements that can be comfortably discerned.

Krzywinski, M. (2016) Points of view: Binning high-resolution data. Nature Methods 13:463.

# Regression diagnostics

Wed 11-05-2016

Residual plots can be used to validate assumptions about the regression model.

Continuing with our series on regression, we look at how you can identify issues in your regression model.

The difference between the observed value and the model's predicted value is the residual, $r = y_i - \hat{y}_i$, a very useful quantity to identify the effects of outliers and trends in the data that might suggest your model is inadequate.

Nature Methods Points of Significance column: Regression diagnostics? (read)

We also discuss normal probability plots (or Q-Q plots) and show how these can be used to check that the residuals are normally distributed, which is one of the assumptions of regression (constant variance being another).