Wednesday, 24 October 2018

TC18 - Devs on Stage

Tableau have a 1,000 developers working across the product suite

Dashboards - Amanda Luthy
Transparent background on visualisations in dashboards. Transparent Layout Containers too (now called zones) 2018.3
Toggle zone visibility - like the former hamburger menu pop-out 2018.3
Vector tile maps - more responsive mapping.
New background maps - topographic and satellite styles to build on traditional Tableau mapping
Navigation buttons - no more workarounds. Work even with hidden tabs (a favourite annoyance of most users). Can choose custom images 2018.3
Export to PowerPoint (excuse me whilst I cry a little - not with glee) 2019.1

Collaboration- Alex Vertin
Public - 1.5 m vizzes, 1.5 billion views
Add Attribution to Tableau Public
New Alert side panel on Server - add yourself to an existing alert 2019.1
New mobile app - biometric authentication, project navigation, interactive offline previews (scrolling, highlighting and tooltips) 2019.1
Automatic phone layouts - can still edit yourself 2019.1

Data Model - Swati Srivastav
The challenge of Tableau forming single tables often makes for large extracts and slower extracts (2018.3)
Multiple Table extracts start to solve this. HyPer can hold tables separately.
Security - encryption at rest
Drag and drop file load in web edit, federated joins and custom sql
LinkedIn Sales Navigator

Developer - Keshia Rose
Extension gallery updates
Extension API updates - write back updates
Support for webhooks - push notifications for events on server (ie failed extracts), can send feeds in to Slack / Convo etc will include snapshot

Analytics - Denny Bromley
Filtered nested sorting (sorting in a single column or a single row but both of the sorts persist)
Parameter actions - parameters filled by any of the action types (creates a lot more dynamic analysis). Time calculations - more interactive and simple once parameter actions set up
Set actions - sets hold multiple values where as parameters just hold one. Add elements to sets by just clicking. Click on one element to drill further (ie just one section of the treemap or table to drill in to a hierarchy further)




Tuesday, 23 October 2018

UBS - Deploying Tableau at Scale

Paul Banoub

Runs a CoE (Centre of Excellence)
Self service model

Scalability
Ability to handle a growing amount of work
Scalability involves performance, maintenance and expenditure
14,000 unique monthly users on server now compared to starting this on a box under his desk

Infrastructure and Tableau Architecture
Online gives you scalability but you lose control of some elements (Postgres database)
48 core environment (v10.3)
Different environments: Production, DR, UAT, Engineering, alpha, beta and Tabjolt / tabmon/ Viz alerts
Build the relationships with the infrastructure departments to get the best of their support and facilities
Upgrades are great but increase require more hardware. You can disable if you need to.
Keep close to Postgres, you will be rewarded to do so
Use open source tools to help avoid harder challenges (like tabjolt / log shark / lumberjack etc). Tableau Replayer - can push historical logs back in to a server environment to show what has happened and why
Splunk to aggregate logs to then do the analysis and not be at risk of handling sensitive data

Service & Support
Use an appropriate service model - how much self-service vs governed?
Self service best for scalability as lowest central demand
Focus on talent recruitment and keep up with R&D
Beware of key dependencies
Efficiency - get the help requests to the right place - servicenow for tickets, Tableau Doctor for the harder stuff and forums for the questions
Users can not be trusted
Housekeeping - delete after 100 days, old subscriptions binned, long running extracts

Training and Community
Create a community with whichever social hub your company offers
Use Tableau champions as an extension of your team
Beware cultural challenges

Vendor & Costs
Customer success program pushing UBS to do more as much as UBS pushes Tableau
Understand your Total Cost of Ownership
Understand the other tools so you know whether they are good or not and why

TC18 - Opening Keynote

Tableau prep developments
Data roles - built in definitions for email address, URL (all within one click on the smart recommendation icon).
Custom data role - create your own rules
Recommendations for fixing data (filter or fix)
Highlighting the flow - highlights where changes have been made based on the field you select
R and Python scripts able to be run in Prep (ie sentiment scoring)
Tableau Prep Conductor - add-on to server to publish and administer flows. Sits in alongside the existing server.

Data Modelling
Choose source table and then choose ‘related tables’.
As long as Primary and Other kres are set-up then the rest of the dataset will be formed
Tableau will automatically work out which table to pick so it avoids duplication
Measures are part of the table of origin rather than just being pushed together
Will determine the join type to use (reducing the need to scaffold)

Natural Language
Ask Data coming in 2019.1
Type in what you want to analyse / just ask your question and Tableau will form the visualisation which can then ask further questions of

Thursday, 5 July 2018

TC18 London - Practical Tableau Tips

Ryan Sleeper

Fundamentals = Success
Ryan’s two most popular posts out of 200 is one on bars and one on lines

Bar chart tips
  1. Give your headers space
  2. But don’t make your bars too fat
  3. For only a few bars - use a direct label and hide the axis
  4. Heavy axis ruler bar to give the base a firmer zero line

Line chart tips
  1. Maximise Data Ink ratio (Tufte concept) To do this minimise the non-data elements
  2. Remove redundant data ink
  3. Make increments on the axis larger
  4. Remove obvious axis labels
  5. Highlight the key element you want to analyse

Psychological Schemas - we learn by patterns so if we disrupt the pattern, we struggle. Ie if green didn’t equal good and red wasn’t bad
  1. Don’t break these within your visualisation or you will confuse your users

Spatial Context
  1. Adding any spatial element to a view helps add additional context and spot weird outliers as we have preconceptions about how the pattern should be

Describe
CTRL+E - in Tableau this can tell us a lot
You can also click on a calc in a Calculation window and click ‘describe’ on the dialogue box and you can copy and paste your calculation

Filter in Use
Create a calculation If sum(num of red) <> attr({fixed: sum (num of records)}) then ‘Filter in use’ else “” end
Add to the dashboard to show when someone is filtering

Strategy
Descriptive analytics (basic dashboarding)
Prescritptive analytics - explain why something has happened and describe what to do about it
Add your own commentary by using a string parameter on another sheet and a three choice option as to whether it is positive, neutral or negative and then show the resulting values in the main dashboard.

TC18 London How Tableau solved a life crisis - Bethany Lyons

Bethany Lyons

Don’t try to be good at Tableau - try to be good at solving (and generalising) problems in Tableau

Meta Points
  1. Optimising performance
  2. Progressing your career
  3. Building teams
  4. Making software recommendations

The crisis
Spending 180 days in the UK to warrant the Permanent Residency status in the UK
From passport stamp data, you can work out days in the UK
Level of detail calcs can give you the days in then UK by trip ID
Bethany had 189 days abroad in 2014 (building the partner ecosystem in France)
Data captured on 7th Jan 2018 but showed 11 days for 2018. This showed the days abroad had been allocated twice

Bethany tried to work out how to split and allocate the years
Step 1. Identify the years
Step 2. Allocate the right number of days to the year (chunky IF statement to work out Start date to end of year, and end date from start of year)
Step 3. There is double counting as trips can have multiple legs
Step 4. Use a sequence value and only bring back max sequence value
...but there are easier ways to solve this

Don’t hard count your calcs for the data you have, think about the use cases which will break your hard coding

Issue - immigration years were rolling year based on your application date

Types of temporal data tables
  1. Event - a record represents a state change at a moment in time
  2. State - each record represents state during a period of time
  3. Calendar - dense time range bounded by event time domain
  4. Time - each record represents a state in a moment in time

Event table was what Bethany used for the original data
State table had a value for a property
Calendar - is a dense time range
Time table - for a dense range of time, a property of an entity has a value (ie for each day, there is a specific value)

Time tables are created by joining calendar date to state table (ie scaffolding as it is more commonly known)

Bethany needed to create a state as to whether she is in the UK or not. To do this she needed a start and end date - you can use aggregation in Prep to do this
Calendar date - use the idea that calendar is <= end date and >= start date (to not bring in any unnecessary date

By adding an appointment date parameter, Bethany could then create a rolling year bin. Take the integer of the date / 365 (minus 1) to work out which year it is

Bethany was fine but then she wants to release were there dates when she would have hit the 180 day limit?

If you want to have a record for everyday, you can join the calendar date to the Data source by using a calc join (1 = 1...see Bethany’s session from last year) to create the rolling year calc

Like all Bethany sessions - watching the recording to get all the value from the techniques but it will hopefully capture some concepts to go away and learn (or Google)!

Complexity arises from incrementally and reactively solving a problem (walk away from the desk before solving the question)
Get a clear understanding of what is, work out the semantics of your question and then it’s over to you as the analyst

Meta summary
Performance problems arise from complexity -Holistic thinking is the best solution. Iterate but then don’t be afraid to rebuild.
Progressing your career - visual design and data design is very different. There is a massive shortage of data designers.
Building teams - get both visual and data designers
Software recommendations - all principles discussed are unique to Tableau. Most companies look at features and price, this leads to examples in the first convoluted feature. Great software helps you escape these elongated builds as it will have features that compose well (VizQL is a language and you can communicate in many different ways)

Wednesday, 4 July 2018

TC18 - Advanced Mark Types

Going beyond Bars and Lines

How many chart types are there in Tableau?
There is no answer as really it’s unlimited depending on the techniques you use
1967 - Jacques Bertans wrote the ‘Semiology of Graphics’
Jock McKinley’s research
2003 - Chris Stolte creates VizQL
Tableau writes all of these elements together into one product

Heatmap Calendar
This isn’t a chart type - it’s taking date parts and placing on to shelves
Weekday on Columns, Week on Rows etc

Pareto chart
Is a line chart and a bar chart in combination

Donut Pie Chart
A dual axis pie chart with the pie in front, whited out. Mrunal used min(1) where as I use avg([Number of Records])

Hex tile map
Using a scaffold of the states and their X Y co-ordinates

Does this mean I should be using all different chart types?
Stop and think - challenge stakeholder requirements - ask why? Need vs Want
Discuss the analytical value
Build vs Analyse mentality - rather than just build the report, if you are asking questions of the data then you will create a better product

If you want the challenge to test your skills then there are lots of blogposts to help you work through the charts

Best way to learn is deconstruct and recreating it
This is what Tableau Public hugely helps with

TC18 - Questioning your marks

Neil Richards - HESA

Zen doesn’t mean technical impossibility

What constitutes a Viz?
Tableau means a picture of a scene - a nice way to look at the overview of a Viz
A framework, one or more visual encodings, annotations

What are the Marks?
The actual representations of your data - ie the Visual Vocabulary from the Financial Times
Removing annotations and axis titles makes most data visualisation useless

What is data Viz consumption?
Does the Viz need to be consumed in 2 seconds?
Really, the fast element is important but don’t ignore the ‘longer term learnings’ that your consumer should be able to be deduced from your work

How important are titles?
Remove them and find out!
Be careful with screenshots as people will often chop titles
Not just titles but also add short descriptions to add more depth or detail for new consumers

What is White Space?
Defined as the bit that isn’t showing data (it doesn’t have to be white)
Data to Ink ratio - what you are using to show the information vs all ink showed on the page. Ie remove any unnecessary elements

Is white space always your friend?
Remove obvious axis titles (years) and just leave the scale
Use colour in your title to remove colour legends
Reduce the darkness of non-data elements to push the detail in to the background

What makes a visualisation memorable?
This is changing over time - style and tools change over time
Intrigue leads to insight (Beaumont) - make people curious
Chart junk can add to memory as it can create a more unique style

How can you make best use of annotations and labels?
Annotations can add to pinpointing insight but also adding more clarity
Empty space can make strong, clear messages with the absence of data

How important is self review and feedback?
Very - especially when you are challenging the normal conventions