Skip to content
sjsrey edited this page Jan 31, 2012 · 28 revisions

Dev Meeting 2012-01-31

Attendees

  • Serge

  • Xing

  • Dani

  • David

  • Jonnie

  • Steph

Notes from Data Structure Subgroup

  • How to organize spatial data up until visualization - different world

Selecting System

  • Keep structure the same - nxm array (features, time) - for every attribute

  • NO

  • Data frames of series

  • Assuming spatial units are fixed over time

  • How about discrete vs. continuous time?? - In literature: discrete

  • For this round - discrete

  • Do all attributes have the same number of time periods??

  • Want to forecast - which results in some tables having more columns

    • Add new forecasting [panel] variable(?) - have class for variable - attributes = time labels, and spatial labels
  • Dimensions - for time and space labels

  • Indices

  • Would allow us to do hierarchies (states - counties - etc.)

  • Hierarchies would mean different numbers of 'i's

  • Should we allow two simultaneous analyses? - take limiting factor (e.g. 10-year vs. annual data - pull out decades)

  • Provide some new flexibility (queries/sub-setting at front end - selection system)

  • Query multi-dimensional arrays (stick with bumpy or use other)

Plumbing

  • Takes analytical selection and visualization selection

Analytics (PySal)

  • Moran’s I

  • Choose something

  • Plumbing gets values, resubmits how you want to bin it

  • How to do dynamic Moran’s I

  • Not recalculating each time

  • Just refitting line to scatterplot

Visualization (PySal? - future)

  • Maps, box plots, etc.

  • What do the inputs need to be?

  • Two data structures - one on analytics and one on visualization

  • Have a middle layer (understands what's going on in each piece - map, scatterplot, etc.)

  • So we don't need to worry about what the inputs are (don't need to retain additional information)

  • Don't need to go back to access additional information

  • Add Charlie’s weights visualizer(?)

Interaction

  • Brushing, linking, etc.

  • Do we need to know the magnitude of the attribute at that point (just need to know that the point corresponds to a point/polygon in a different view)

Firewall between all spaces...?

  • Would this sacrifice speed?

Visualization Subgroup

Views from Old STARS

  • Way it's currently organized - single STARS gui could be improved

  • Different subplots be put in separate source files for the purpose of clarification

  • Instead of having a comprehensive starsgui.py which could make it hard to organize

NEW Want to have user define interactions after the fact

  • Need to be able to check that newly specified interactions conflict with current interactions

  • More burden on the user to define interactions

  • Have some middle ground - keep existing interactions

  • Menuing system - ctrl-click (rebind command)

  • These interactions exist, do you want them?

  • What else do you want - do you want to turn off other interactions?

  • "Advanced" button

  • This item only, or all of this type?

Other

  • Third correlation table that's continuously updating

  • Performance - how you walk user through importing the data and how it's organized/stored in data structures - not the problem of visualization

  • Improve performance of repository, etc. from old STARS

  • Yes on one module for the abstract view class

  • Concrete classes - map, scatterplot, etc. would borrow from this abstract class

  • Multiple directories instead of one (which is how STARS is now)

  • Also - moving away from WX only

New gui from Google

  • From GapMinder guy

  • Written in Flash(?)

  • In realm of adding more to visualization

  • Now we have snapshots - but this smooths it

  • Want to provide the option of showing a trail or not