Tango is a sad thought that is danced.think & dancemore quotes

# science: beautiful

EMBO Practical Course: Bioinformatics and Genome Analysis, 5–17 June 2017.

# Writing

## Poetry

Inked Sadness is a collection of poems written in that difficult time between being no longer young, but not yet old.

Expressions and Conversations of Love are short heart-breakers.

## Scripts

Proverbial Man considers how our names and expectations limit our perceptions and freedoms. No answers are offered. I don't think I finished this.

The Surrogate echoes my own fears of fleeing from Poland in the early 80s. Or at least the fears I retroactively erected, to add drama. Count on the Polish for drama.

## Narrative

Thirty Reasons should be enough to decide a place isn't worth returning to.

## NeverEnder — Epic Space Poem

By my friend and colleague Emanuele Raffaele Ettore Libertini, the NeverEnder is a voyage through space and words.

VIEW ALL

# Essentials of Data Visualization—8-part video series

Mon 16-01-2017

In collaboration with the Phil Poronnik and Kim Bell-Anderson at the University of Sydney, I'm delighted to share with you our 8-part video series project about thinking about drawing data and communicating science.

Essentials of Data Visualization: Thinking about drawing data and communicating science.

We've created 8 videos, each focusing on a different essential idea in data visualization: encoding, shapes, color, uncertainty, design, drawing missing or unobserved data, labels and process.

The videos were designed as teaching materials. Each video comes with a slide deck and exercises.

# P values and the search for significance

Mon 16-01-2017
Little P value
What are you trying to say
Of significance?
—Steve Ziliak

We've written about P values before and warned readers about common misconceptions about them, which are so rife that the American Statistical Association itself has a long statement about them.

This month is our first of a two-part article about P values. Here we look at 'P value hacking' and 'data dredging', which are questionable practices that invalidate the correct interpretation of P values.

Nature Methods Points of Significance column: P values and the search for significance. (read)

We also illustrate how P values can lead us astray by asking "What is the smallest P value we can expect if the null hypothesis is true but we have done many tests, either explicitly or implicitly?"

Incidentally, this is our first column in which the standfirst is a haiku.

Altman, N. & Krzywinski, M. (2017) Points of Significance: P values and the search for significance. Nature Methods 14:3–4.

Krzywinski, M. & Altman, N. (2013) Points of significance: Significance, P values and t–tests. Nature Methods 10:1041–1042.

# Intuitive Design

Thu 03-11-2016

Appeal to intuition when designing with value judgments in mind.

Figure clarity and concision are improved when the selection of shapes and colors is grounded in the Gestalt principles, which describe how we visually perceive and organize information.

One of the Gestalt principles tells us that the magenta and green shapes will be perceived as as two groups, overriding the fact that the shapes within the group might be different. What the principle does not tell us is how the reader is likely to value each group. (read)

The Gestalt principles are value free. For example, they tell us how we group objects but do not speak to any meaning that we might intuitively infer from visual characteristics.

Nature Methods Points of View column: Intuitive Design. (read)

This month, we discuss how appealing to such intuitions—related to shapes, colors and spatial orientation— can help us add information to a figure as well as anticipate and encourage useful interpretations.

Krzywinski, M. (2016) Points of View: Intuitive Design. Nature Methods 13:895.

# Regularization

Fri 04-11-2016

Constraining the magnitude of parameters of a model can control its complexity.

This month we continue our discussion about model selection and evaluation and address how to choose a model that avoids both overfitting and underfitting.

Ideally, we want to avoid having either an underfitted model, which is usually a poor fit to the training data, or an overfitted model, which is a good fit to the training data but not to new data.

Nature Methods Points of Significance column: Regularization (read)

Regularization is a process that penalizes the magnitude of model parameters. This is done by not only minimizing the SSE, $\mathrm{SSE} = \sum_i (y_i - \hat{y}_i)^2$, as is done normally in a fit, but adding to this minimized quantity the sum of the mode's squared parameters, $\mathrm{SSE} + \lambda \sum_i \hat{\beta}^2_i$.

Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Regularization. Nature Methods 13:803-804.