curios
resources
presentations
lunch break
Distractions and amusements, with a sandwich and coffee.
Alex — Internet's Most Popular Rat
Poster Rat for Rat Genome Sequencing
|
The rat genome sequencing project at the Baylor College of Medicine Human Genome Sequencing Centre is complete. The genome has been analyzed and published. I'd like to introduce you one of the faces of the project: Alex, the genomics rat idol. Arguably, Alex is the most popular rat on the internet. For the justification of this strong statement, read on. 
▲ Alex, the rat. Rattus norvegicus on an ABI 3700 genome sequencer.
(zoom)
▲ Alex, the rat. Rattus norvegicus on an ABI 3700 genome sequencer.
(zoom)
Alex's BiographyAlex was born in May 2000. It's well known that a rat's cuteness reaches maximum at about 3-4 weeks. After this critical time, a pet store rat is less likely to be purchased and may be asked to act as snake food. In Alex's case, she was perilously close to her deadline. Luckily for her, we paid a ransom of $6.99 to the Noah's Ark pet shop in Vancouver. She was on her last cute leg. 
▲ Portrait of Alex, the genome rat (Rattus norvegicus). Here, she is seen in a forced portrait position
(zoom)
From May 2000 Alex spent most of her time hoarding food pellets and riding on shoulders. 
▲ Portrait of Alex, the genome rat (Rattus norvegicus). Riding on shoulder.
Alex liked to bite. And rats only bite hard — they don't nibble. Her contention for this unattractive behaviour was the uncanny similarity between a finger and a pellet of food. Other than unpredictable bouts of biting (by far the most exciting aspect of her personality), Alex lacked other distinguishing characteristics. Alex died of a seizure in late 2002. She was buried outside of the Museum of Anthropology. A ratty pair of underwear served as a burial shroud. And I hope you got that last pun. |
DOWNLOAD ALL PHOTOS.
Photos are for public use. Use, modification and distribution of these photos is unrestricted. Alex's PopularityDespite my best efforts at meaningful work, this web page continues to be the most popular of all my online offerings, making for a somewhat embarrassing achievement. Alex's images consistently show up first in Google's web search for 'rat', 'rat image' and image search for 'rat'. 
▲ Alex image is the first for Google's 'rat' search query (retrieved 16 Mar 2013).
(rat Google search)
▲ Alex image is the first for Google's 'rat image' search query (retrieved 16 Mar 2013).
(rat Google search)
Finally, Alex appears as the first entry in Google images for 'rat'. 
▲ Alex image is the first for Google's 'rat image' search query (retrieved 16 Mar 2013).
(rat Google search)
Alex's Public AppearancesAlex is neither without modesty nor public fame. Her first cover-ratgirl appearance was on the April 2004 issue of Genome Research. 
▲ Alex the rat appeared on the cover of Genome Research (April 2004).
(zoom)
More recently, she's appeared on the cover of Ethnologie Francaise (Jan-Mar 2009 issue). 
▲ Alex the rat appeared on the cover of Ethnologie Francaise (1/2009).
(zoom)
The topic of the issue was the relationship between animals and humans. It is fitting therefore to recount here the relationship I shared with Alex during her sojourn with us. |
upcoming
news + thoughts
Model Selection and Overfitting
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.
Background reading
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.
Classifier Evaluation
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.
Background reading
Lever, J., Krzywinski, M. & Altman, N. (2016) Points of Significance: Logistic regression. Nature Methods 13:541-542.
Happy 2016 `\pi` Approximation, roughly speaking
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.