Week 20 notes

Week 20, 2016

Welcome to week 11 of 34…

thisweek <- 11
yikes <- c(thisweek, 34-thisweek) / 34
barplot(as.matrix(yikes), horiz=TRUE, beside=FALSE)

Last week we got a scoreboard deployed sorta, then I did some silly things on github and broke it, and realised I had also done some silly things with postgres backups. So, the normal two-steps-forward-one-back routine.

Quote of the week comes from SO, original source unknown:

    "git is like UNIX. User friendly but picky about it's friends."

Last week we also tossed a new wrench in the mix, discussing netCDF as the main format for new data acquisitions. I don’t care much what format people send data in, but if we’re going to automate this tool (which is to say make it outlast it’s creator, a goal all makers share) then we need to predict the weird variations we’re gonna see. And I’ve never seen a netCDF file.

This week I need to:

• deploy AWS instance of shiny server
• decide if the SOS database format is the way to go forward,
• which makes maintenance easier
• more durable
• bit trickier upfront !
• mysteries to solve
• R fundamentals I still need
• ggplot2
• facet()

This week so far I am also reading and writing:

Dr. Furusho et al (including Dr. Ramos, my thesis advisor) point out Flanders uses a mixture of ground- and surface-water for a drinking water resource.

A potentially-interesting question I’ll try to address will involve ‘scoring’ basins with a strong surface / groundwater connection. A hypothesis introduced here last month by Dr. Murray Peel from U. Melbourne involved looking at drought tolerance across certain basins in Australia,

Peel asked if the models’ poor performances in certain basins (based on observed “Millenium Drought” period 1997-2009) due to 1) model structure? or 2) parameter set?

Answer: some models do better than others, but after Pareto runs there is NO solution within the parameter set for a set of discontinuous basins. They seem to share these characteristics: drier (less mean anual precip); lower slopes; less woody vegetation.

• Writing:
• in LaTex (yikes) for first time in forever
• soutenance en mauvaise français
• an English version
• an update to my linkedin profile
• modified local db to match v2 specifics
• built interactive dataframe to postgres db,
• finished main score series viewer

This first-pass structure worked for simple data off all one datatype…

… but in fact we are scoring many variables which need to be tied together more explicitely. Hence, version 2:

We need a structure for the scores to import - currently receiving text files and 3D “cubes” depending on source… tidying takes time; should be automated so users may load / arrange their scores (like EVS).

Considering NetCDF for this … oldie but goodie? Opinions? http://www.unidata.ucar.edu/software/netcdf/docs/faq.html#How-do-I-convert-netCDF-data-to-ASCII-or-text http://www.unidata.ucar.edu/software/netcdf/examples/files.html Discussion(s) of handling time using netCDF: http://www.unidata.ucar.edu/software/netcdf/time/ http://www.unidata.ucar.edu/software/netcdf/time/recs.html

Some NetCDF files from our friends at ECMWF: http://apps.ecmwf.int/datasets/

Doing all date comparisons using posix-happy functions:

m <- as.POSIXlt(dateValue)\$mon

Working with “reactive” call today: https://gallery.shinyapps.io/003-reactivity/

O’Reilly always publishes goodness, will put this here to remember later: http://www.cookbook-r.com/Graphs/

Something to look into on my time – confidence Intervals discussed in different context: http://learnbayes.org/papers/confidenceIntervalsFallacy/introduction.html …with nifty Shiny app to illustrate Figs 1 - 5 from article: https://richarddmorey.shinyapps.io/confidenceFallacy/ http://learnbayes.org/papers/confidenceIntervalsFallacy/