All presentations are available in the Data Visualization Society Youtube channel.
https://www.youtube.com/watch?v=0RONImrGtZE :
A good lesson, quite universal:
"It required me some matury to appreciate what people could offer me instead of wanting to have their knowledge by myself. I think we all experience at some time, probably at an early career stage, this temptation to learn how do things all by yourself, to take every courses, every trainings. In an attempt to fill all those missing spots in our skills that we would see and that would make us feel miserable because we would believe then that it makes us not legitimate in ours fields. [...] I learned to appreciate and see there is more in collaboration than in independence and self-sufficiency"
(Btw: Julie's slides are probably the most beautifully animated and designed slides I've seen.)
https://www.youtube.com/watch?v=EnBIwqTYkjE
A mini-experiment that deals with a subject I'm facing everyday : our personal time and energy/creativity managemen.
Plus we see a potential use-case of applied operational research + data science for creating a schedule that better distributes attention and energy :)
(This reminds me of an article we studied in a cognitive science class ... : "We find that the percentage of favorable rulings drops gradually from ≈65% to nearly zero within each decision session and returns abruptly to ≈65% after a break.")
https://www.youtube.com/watch?v=ms1v9TqpBP0
Stephanie gives us a quick overview of 4 python dataviz packages: Bokeh, Plotnine, Altair and Plotly (+ the "usuals", seaborn and matplotlib).
notes + my 2cts:
- I love ggplot and the concept of grammar of graphics, so I'm very interested in plotnine, I can't wait to try it in a future project (plotnine uses R/ggplot as backend so it may be quite "heavy" to use).
- Bokeh is great for interactivity, but limited in features (e.g.: no one-liner histogram?): Holoview, which is not mentioned, but based on bokeh, is also well worth a look.
- Seaborn has well-defined statistical visualizations but doesn't really have a grammar of graphics / isn't very customizable without going through the boring and akward matplotlib API.
- Altair has an attractive declarative syntax but doesn't seems to be usable on medium to large datasets without sampling.
- Plotly is the champion of interactivity, and can scale to large datasets, but the json-based API is complex to use and the express API is easy but limited in customization (no grammar / compositionality).
Best of winners from the presentation (quite subjective):
- Plotnine: best histograms
- Plain scatter: Bokeh
- Facetted scatter: Plotly express or Plotnine
- Timeseries: Bokeh
- 3D scatter plot: Plotly
https://www.youtube.com/watch?v=wxmqG_jxJiw
Nothing much applicable / efficient / approachable for my projects here, but a treat for the eyes!
(Craig uses Houdini for dataviz, wow!).
https://www.youtube.com/watch?v=vuFsgz4W0Pk
Human are really weird. We are hard to predict, we are hard to satisfy, we are really bad at knowning at what we actually want. If are building visualisations for other people, your up against this battle, you are a code breaker.
Lessons learned from eye tracking analysis of dashboard consumption:
"Building data visualization for others is a design act [...] you have to take this part as seriously as you take the dataprep and analysis step."
A few tips:
-
BANS: "If you have an important number make it a BAN" (Big Ass Numbers)
-
Reading: "You can expect reading behaviors. Your audience will likely visually work your dashboard like they would a page in a book: does your sequencing make sense, are you helping them find their way and glean information from your analysis, [etc.]. Ask someone to look at it for you, try to see through their eyes."
-
Context matters: When you can't train/predict/known your audience and context of consumption, then "build context into your dataviz: either by adding a title in high contrast, design elements that are meaningful within their context and priming in the develivery vehicule [ex: the email you sent along with the dashboard/dataviz]" (Here Priming is used as a way to control context by giving visual/written cues that facilitate the correct interpretation of the data visualisation).
[en] https://www.youtube.com/watch?v=DJwG-4MqnEQ [es] https://www.youtube.com/watch?v=rvyosKes1Jc
https://www.youtube.com/watch?v=BldwMGnwrZw
https://www.youtube.com/watch?v=f1eOh6ocnqg
https://www.youtube.com/watch?v=viX2NN3wjZU
https://www.youtube.com/watch?v=jnHTb8lUuqE
https://www.youtube.com/watch?v=mQabRqO6xRc
https://www.youtube.com/watch?v=H-XLunjxItw
https://www.youtube.com/watch?v=t9gLlOmnAmw
https://www.youtube.com/watch?v=8B8uQt9SVpk
