Tableau Tips and Tricks

Presenter: Jie Zhang

Link to Tableau answers workbook: https://github.com/jiezhang28/presentations/blob/master/tip_tricks_for_presentation_answers.twbx

Link to Tableau Level of Detail feature: https://stanford.box.com/s/bclwssox8ahipdxbg69u9wtzcqwv1mjw

Notes: Aayush Neupane

Location: Alway Building, M208

Time: 9:15 - 10:00

Operations Intelligence

• Work outside of stanford

• Pull data, analyze data to discover insight, present to businesses

Live Coding

Data Source Pivot

• Survey data

  • Usually wide table

  • Tableau creates a lot of measures, which might be useless

  • Also offers a “pivot” feature

  • Select the columns that you wouldn't need, right click -> pivot.

    • Using repeated columns and creating them into rows

• Convert text data to force to use “text”

• Created average of scores based on questions, divided by gender.

• Q - what are you trying to achieve using this format?

• A - This is helpful to analyze survey data

Right Click Drag

• Right click and dragging a field gives you a list of options of type of field you’d want to use

  • Useful for dates where default is year

Quick Histogram

• Useful for seeing distributions

• Click on a measure, go to showme, click histogram

  • Wait for data to load, and that will use the ranges in data

• Log transform

  • Log([measure], 2) -> gives you a traditional histogram

Pricing of houses by zip code

• column: longitude

• Row: latitude

• Marks: median price, zip code

• Using log(price) gives you a better color coding instead of a traditional histogram

Add to Context

• Row: id

• Marks: Max(Price)

• Used top 10 priciest houses

• Filtering by zip code originally clears all results

  • Need to run zip code filter before top10

  • Right click on filter -> Add to context

• Data source filter > Context Filter > Filter

Level Of Detail (LOD)

• House sell value data

  • If a house was sold multiple times, it will have multiple records.

  • Overall sell value was inflated by multiple records

• Row: Id (kc_house_data)

• Filter: sum(number of records)

• Fixed id:  max(price)

  • Force each house to have a single max price as its record

Donut Chart

• Pie charts are bland looking

• Create a calculated field

  • Value 0

• Drag that to row, twice

• First chart, increase size

• Second chart, remove color, add label to the middle

• Join both charts at the end, you’ll see a donut chart with label in the middle

Jitter

• Box plot of bottle of wine prices

• If theres a lot of data points are collected in same place, use jitter to separate data

• Create a new measure called index()

• Easier to see data that's not just on single dimension

Source control

• Havent tried it, but tableau is XML file, source control is possible

Drawing shapes on tableau

• Takes a while to prepare, but it is possible using x, y coordinates

Word Cloud

• Filter: id

• Row: Description

• Create a word cloud out of the description

  • Tableau needs in format as 1row per word

• Change the words to lowercase : lower(description)

• Remove all punctuations: regexp_replace(string, pattern, replacement)

• Figure out tokens within the descriptions

  • Create a calculated field called token

  • len(regexp_replace(text, ‘\w+’, replacement)) -> replace all words words with spaces and then count those spaces.

• Force a inner join with no matching constant

  • Cartesian join

• Create filter

  • Index <= token

• Create data source filter

  • Repeat rows that only have repeated words

• 2 more layers of transformation

• Break down the tokens using findnth() function