The events Package
The aim of events is to make life a bit easier for people who
analyse event data: the kind of thing that KEDS/TABARI generates as
output (see e.g. here).
There's nothing fancy in the package, just a hopefully logical interface to all the data massaging we do to event data before any actual analysis.
The philosophy of the package is that you will want to read in raw
event data and then apply a sequence of filters and aggregators for
actors and event types, each of which will result in an new event data set.
The final step is an application of
make_dyads to make a set of named
temporally regular directed dyad time series suitable for time series analysis.
All rather unix-ish, if you care for that sort of thing.
There's a fairly complete walkthrough in the package vignette. You should probably read that.
The package is ultimately intended to unify the existing software, e.g. the packages currently linked from the PSU event data pages.
The unification is certainly not complete, but we're getting there. In particular, extremely large data sets are probably going to be rather unwieldy in the current version.
The sections below provide a quick compare and contrast to the software available from PSU that works with event data output (not Factiva stuff or actor dictionaries).
This functionality is implemented as a standalone function
scrub_keds and as an option in the event data reading function
This is also a standalone function
one_a_day and available as an
option in the event data reading function
Most, but not all, of the functionality of KEDS_Count is implemented.
Some differences are noted below:
Aggregation is possible by day, week, month, quarter and year, but not biweekly.
read_kedsassumes that two digit year specifications 69 to 99 indicate years between 1969 to 1999, and 00 to 68 indicate 2000 to 2068.
This is R and the POSIX standard interpretation for this needlessly ambiguous date formulation. In contrast KEDS_Count treats 28 to 99 as in the twentieth century and 00 to 27 as in the twenty first century.
Wildcarding for actors and event codes is not implemented directly, so you have to use R's facilities. An example using
grepis provided in the vignette. Also, using "***" to mean 'all targets' is done by using
filter_actorswith a suitable second parameter.
KEDS_Count aggregates up to a temporal unit, e.g. a week, differently to
events: It tosses events that occur before the beginning of the first full unit.
make_dyadsdoes not. There is no warning about this behaviour, but then it doesn't eat your data either.
KEDS_Count only offers summed scores and provides the per time unit N to construct means. This package will use any aggregating function you give it, e.g.
medianand also automatically generate N.
Unlike Aggregator, events does not have a customizable aggregation period, but it is a lot more flexible about what the aggregation function is (see above).
Since I can't figure out what this program does, there are probably no functions to replicate it.
High-Volume Processing Suite
All this stuff is currently out of reach, but there are plans for accessing databases as sources of event data. That'll be how we scale.
- More test coverage.
make_dyadscoverage is still lacking
- Database connection for larger data sources