323 College of Education. 9:30 AM-11:00 AM. About 3 dozen attendees from around campus, unknown number online.

Opening Remarks

Tara Sadler, AITS

These were mostly about the AITS hosted Tableau service, also mentioned the Tableau Webinars.

Deep Dive into Tableau Prep

Ryan Young, OSMA

Builds a lot of dashboards, works with data on ~37% of all UI Students.

Projects:

  • Nationwide comparison with all other US universities
  • OMSA student dashboard
  • Course enrollment dashboard
  • Graduation rates, persistence, intercollege transfers
  • End of semester surveys
  • etc.

Tableau Prep looks like a visual-programming environment for data cleaning and massaging. There's a flow view, which is either a tree or DAG of operations like import, join, filter, etc. There's also a column view, which looks like a quick bar chart summary of counts per column, and there's the row view which is a standard spreadsheet table view. 

Workflow Example

  • Data comes from a wide variety of sources in a wide variety of types:
  • Showed 3 different flows he uses
    1. Nothing but a long chain of unions for spreadsheets.
    2. Deceptively short, but takes 45 minutes to run.
    3. A more complex one that involved multiple different operations like joins and filters (didn't really go into details.)
  • His Tableau likes:
    • timesaver
    • intuitive
    • flexible
  • His Tableau disklikes
    • automated data cleaning isn't as great as advertised
    • resource hog
    • no cut-paste of sub-flows in flow view, no modularity, so each flow has to be hand-crafted

Q&A

  • One question about why he used Tableau for certain things. His main justification is because of the wide variety of data sources and formats he has to work with.
  • Multiple questions about data cleaning, in particular the "fuzzy clustering" it does. E.g., it can try to group different spellings like {M Vanmoer, M Van Moer, M VanMoer, M VanMoore}, etc.
  • Question about the resource requirements. His biggest data table is around 20,000,000 rows and processing that con require about 0.5TB of free diskspace. Apparently Tableau does a ton of swapping things in and out of memory.

Using Tableau for Interactive Dashboards

Sergio Contreras, NCSA/IDDS (grad student, soon full time)

Had previous QlikView experience, but apparently that is even more outrageously priced than Tableau. At NCSA he's worked on dashboards for XSEDE, Blue Waters, and Industrial Partners.

XSEDE Dashboard Example

Originally, XSEDE had been doing their KPI and metrics reporting as static, complex tables embedded in documents. The main drawbacks to this were it wasn't interactive and it didn't show any trends. Sergio worked on moving this to an interactive dashboard. They do quite a bit of pre-processing using SQL, Python, and Ruby, because they want the dashboard interaction to be fast. They connect live to the pre-processed database, which is only about 2,000 records. 

The dashboard itself was stacked bar charts with reference lines, stoplights, check boxes, etc. A user can select a time frame, a metric/KPI, see if it met the projection, what the trend was, etc. They also have a simplified dashboard for a higher level overview. 

He had some comments about having to wrestle with Tableau, e.g., adding a link in a hover textbox isn't a link, it's an "action."  He showed a few other webpages they use in the project, one for keeping track of KPIs being dropped, added; another that is actually a Ruby frontend for doing some data prep, IIUC.

Q&A

One comment giving another work around for the link-in-hover-text problem. One question about the particulars of their live setup. They  host this at AITS and can get away with it because they've done all the preprocessing to drop the size of their database.

Closing Remarks

Encouragement from Tara about sharing information and trying to grow the campus Tableau user community.

  • No labels