Gilgamath

I recently saw a plot purporting to show the Rotten Tomatoes' 'freshness' rating of Robert De Niro movies over the years, with some chart fluff suggesting 'Bobby' has essentially been phoning it in since 2002, at age 59. Somehow I wrote and maintain a mirror of IMDb which would be well suited to explore questions of this kind. Since I am inherently a skeptical person, I decided to look for myself.

You talkin' to me?

First, we grab the 'acts in' table from the MariaDB via dplyr. I found that working with dplyr allowed me to very quickly switch between in-database processing and 'real' analysis in R, and I highly recommend it. Then we get information about De Niro, and join with information about his movies, and the votes for the same:

library(RMySQL)
library(dplyr)
library(knitr)
# get the connection and set to UTF-8 (probably not necessary here)
dbcon <- src_mysql(host='0.0.0.0',user='moe',password='movies4me',dbname='IMDB',port=23306)
capt <- dbGetQuery(dbcon$con,'SET NAMES utf8')
# acts in relation
acts_in <- tbl(dbcon,'cast_info') %>%
    inner_join(tbl(dbcon,'role_type') %>% 
        filter(role %regexp% 'actor|actress'),
        by='role_id')
# Robert De Niro, as a person:
bobby <- tbl(dbcon,'name') %>%
    filter(name %regexp% 'De Niro, Robert$') %>%
    select(name,gender,dob,person_id)
# all movies:
titles <- tbl(dbcon,'title') 
# his movies:
all_bobby_movies <- acts_in %>%
    inner_join(bobby,by='person_id') %>%
    left_join(titles,by='movie_id')
# genre information
movie_genres <- tbl(dbcon,'movie_info') %>%
    inner_join(tbl(dbcon,'info_type') %>% 
        filter(info %regexp% 'genres') %>%
        select(info_type_id),
        by='info_type_id') 
# get rid of _documentaries_ :
bobby_movies <- all_bobby_movies %>% 
    anti_join(movie_genres %>% 
        filter(info %regexp% 'Documentary'),by='movie_id')
# get votes for all movies:
vote_info <- tbl(dbcon,'movie_votes') %>% 
    select(movie_id,votes,vote_mean,vote_sd,vote_se)
# votes for De Niro movies:
bobby_votes <- bobby_movies %>%
    inner_join(vote_info,by='movie_id')
# now collect them:
bv <- bobby_votes %>% collect() 
# sort it
bv <- bv %>% 
    distinct(movie_id,.keep_all=TRUE …

Earlier this year, I participated in the Winton Stock Market Challenge on Kaggle. I wanted to explore the freely available tools in R for performing what I had routinely done in Matlab in my previous career, I was curious how a large investment management firm (and Kagglers) approached this problem, and I wanted to be eyewitness to a potential overfitting disaster, should one occur.

The setup should be familiar: for selected date, stock pairs you are given 25 state variables, the two previous days of returns, and the first 120 minutes of returns. You are to predict the remaining 60 minutes of returns of that day and the following two days of returns for the stock. The metric used to score your predictions is a weighted mean absolute error, where presumably higher volatility names are downweighted in the final error metric. The training data consist of 40K observations, while the test data consist of 120K rows, for which one had to produce 744K predictions. First prize was a cool $20K. In addition to the prizes, Winton was explicitly looking for resumes.

I suspected that this competition would provide valuable data in my study of human overfitting of trading strategies. Towards that end, let us gather the public and private leaderboards. Recall that the public leaderboard is what participants see of their submissions during the competition period, based on around one quarter of the test set data, while the private leaderboard is the score of predictions on the remaining part of the test data, and is published in a big reveal at the close of the competition. Let's gather the leaderboard data.
(Those of you who want to play along at home can download my cut of the data.)

library(dplyr)
library(rvest)

# a function to load and process a leaderboard …

If you're like me, you just blindly check boxes when submitting packages to CRAN. (The 'submit' button should be labeled 'yolo' as far as I'm concerned.) After getting burned yet again for not actually checking my package with the development build of R, I decided to be slightly less stupid in the future. Rather than install R-devel, I made a docker base image for CRAN checking.

As an example, to check my sadists package, I made essentially the following Dockerfile:

# preamble#
FROM shabbychef/crancheck
MAINTAINER Steven E. Pav, shabbychef@gmail.com

# tweak this to force re-install
ENV DOCKER_INSTALL_NONCE 97c22800_9f88_4830_806a_2614e06600f2

# rinstall somethings...
RUN /usr/local/bin/install2.r PDQutils hypergeo orthopolynom shiny testthat ggplot2 xtable knitr

It starts FROM the crancheck image on docker hub. The general recipe would be to install any system packages via apt-get, then any CRAN packages via install2.r, then any github packages via /usr/local/bin/installGithub.r. The base image 'does the right thing' with respect to the entrypoint and you give the package file as the command.

I built it via:

docker build --rm -t shabbychef/sadists-crancheck docker/

Once the image is built, checking a package is as 'simple' as attaching the local directory as /srv in the container via a volume, and giving the name of the package file. (That is, when the command to the container is sadists_0.2.2.5000.tar.gz, it will try to check, as CRAN, the file /srv/sadists_0.2.2.5000.tar.gz. You had better make sure it is available there, so attach this directory here containing the package to /srv in the container.) In summary, run it like this:

docker run -it --rm --volume $(pwd):/srv:ro shabbychef/sadists-crancheck sadists_0.2.2.5000.tar.gz 

You get output as follows:

* using log directory …

I recently released a package to CRAN called madness. The eponymous object supports 'multivariate' automatic differentiation by forward accumulation. By 'multivariate', I mean it allows you to track (and automatically computes) the derivative of a scalar, or vector, or matrix, or multidimensional array with respect to a scalar, vector, matrix or multidimensional array.

The primary use case in mind is the multivariate delta method, where one has an estimate of a population quantity and the variance-covariance of the same, and wants to perform inference on some transform of that population quantity. With the values stored in a madness object, one merely performs the transforms directly on the estimate, and the derivatives are computed automatically. A secondary use case would be for the automatic computation of gradients when optimizing some complex function, e.g. in the computation of the MLE of some quantity.

A madness object contains a value, val, as well as the derivative of val with respect to some \(X\), called dvdx. The derivative is stored as a matrix in 'numerator layout' convention: if val holds \(m\) values, and \(X\) holds \(n\) values, then dvdx is a \(m \times n\) matrix. This unfortunately means that a gradient is stored as a row vector. Numerator layout feels more natural (to me, at least) when propagating derivatives via the chain rule.

For convenience, one can also store the 'tags' of the value and \(X\), in vtag and xtag, respectively. The vtag will be modified when computations are performed, which can be useful for debugging. One can also store the variance-covariance matrix of \(X\) in varx.

Here is an example session showing the use of a madness object. Note that by default if one does not feed in dvdx, the object constructor assumes that the value is equal to \(X\), and so …