Month: May 2018

Julia Silge explains Principal Component Analysis and shows us an example using Stack Overflow data:

We have tidy data, both because that’s what I get when querying our databases and because it is useful for exploratory data analysis when preparing for a machine learning algorithm like PCA. To implement PCA, we need a matrix, and in this case a sparse matrix makes most sense. Most developers do not visit most technologies so there are lots of zeroes in our matrix. The tidytext package has a function cast_sparse() that takes tidy data and casts it to a sparse matrix.

sparse_tag_matrix <- tag_percents %>% tidytext::cast_sparse(User, Tag, Value)

Several of the implementations for PCA in R are not sparse matrix aware, such as prcomp(); the first thing it will do is coerce the BEAUTIFUL SPARSE MATRIX you just made into a regular matrix, and then you will be sitting there for one zillion years with no RAM left. (That is a precise and accurate estimate from my benchmarking, obviously.) One option that does take advantage of sparse matrices is the irlba package.

This is a great walkthrough of an important topic.

Joachim Zuckarelli walks us through the xplain package in R:

The above XML produces the following output (don’t worry too much about the call of xplain(), we will discuss later on in more detail how to work with the xplain() function):

library(car)
library(xplain)
xplain(call="lm(education ~ young + income + urban, data=Anscombe)",
xml="http://www.zuckarelli.de/xplain/example_lm_foreach.xml")

##
## Call:
## lm(formula = education ~ young + income + urban, data = Anscombe)
##
## Coefficients:
## (Intercept) young income urban
## -286.83876 0.81734 0.08065 -0.10581
##
##
## Interpreting the coefficients
## —————————–
## Your coefficient ‘(Intercept)’ is smaller than zero.
##
## Your coefficient ‘young’ is larger than zero. This means that the
## value of your dependent variable ‘education’ changes by 0.82 for
## any increase of 1 in your independent variable ‘young’.
##
## Your coefficient ‘income’ is larger than zero. This means that the
## value of your dependent variable ‘education’ changes by 0.081 for
## any increase of 1 in your independent variable ‘income’.
##
## Your coefficient ‘urban’ is smaller than zero. This means that the
## value of your dependent variable ‘education’ changes by -0.11 for
## any increase of 1 in your independent variable ‘urban’.

I’ll be interested in looking at this in more detail, though my first glance indication is that it’ll be useful mostly in large shops with different teams creating and using models.

Sara Locatelli analyzes the lyrics to Hotel California using tidytext:

Sentiment analysis is a method of natural language processing that involves classifying words in a document based on whether a word is positive or negative, or whether it is related to a set of basic human emotions; the exact results differ based on the sentiment analysis method selected. The tidytext R package has 4 different sentiment analysis methods:

• “AFINN” for Finn Årup Nielsen – which classifies words from -5 to +5 in terms of negative or positive valence
• “bing” for Bing Liu and colleagues – which classifies words as either positive or negative
• “loughran” for Loughran-McDonald – mostly for financial and nonfiction works, which classifies as positive or negative, as well as topics of uncertainty, litigious, modal, and constraining
• “nrc” for the NRC lexicon – which classifies words into eight basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) as well as positive or negative sentiment

Sentiment analysis works on unigrams – single words – but you can aggregate across multiple words to look at sentiment across a text.

To demonstrate sentiment analysis, I’ll use one of my favorite songs: “Hotel California” by the Eagles.

Read the whole thing, though you can’t check out afterward.