You see them a lot in biological network / pathway / association studies. I personally hate them, because they're hard to read and comprehend. Many times a heatmap conveys the exact same data in a more digestible way. Alas, chord diagrams can be pretty and make it seems like the analysis is much fancier than it really is.
Plotting directed acyclic graphs while displaying the edge weights and node sizes is a bit of an unsolved problem. I agree that in most cases chord diagrams are more confusing than illuminating, but so are the alternatives.
Since I got introduced to this simple but effective visualization (in my 30s!), I tend to use it everywhere.
Out of the “chart generation” apps listed in the catalog, only Vizzlo seems to support it: https://vizzlo.com/create/waterfall-chart (and Tableau, I suppose, but IDK if I'd count this as a simple chart generator). I did not check the listed charting libraries.
In my experience, this is the go-to chart for understanding financial drivers at the senior leader/board level.
Even the popular plotting libraries in python don’t seem to have an easily configurable version. Recently tried plotly and found it plotting the categories in reverse order from the intuitive way you’d expect to see.
I’ve been meaning to script my own custom plot since Excel’s waterfalls are tedious to create and you don’t get enough control over the labeling.
I'm sorry you had a bad experience with Plotly... I'd be curious to hear more if you're willing to share! Were you working in Python or Javascript or within our GUI?
It's super hard to put together because you have to first do a factor analysis to even decide what the component parts are. I've always brute forced it using the (admittedly difficult) Excel function and the underlying dataset is not pretty!
So some brief advice from spending about 5 years as a visualization engineer/designer/data scientist:
1. Get an intuitive understanding of the data
2. (via Alberto Cairo) Aim to allow the user to correlate, organize and compare this data
3. Come up with a few visual means to do no. 2, and get feedback and iterate.
4. Go to the catalogs and see what work has already been done on your version of 3, and then implement.
It is amazing how many "original" charts I have created which have already been named and studied from a perceptual science point of view. But it is still important to come to a clear understanding from your own point of view of how the data fits that representation.
>> "original" charts I have created which have already been named and studied
thank you! I am slightly embarrassed to say, I thought I had personally invented a certain version of the BoxPlot, and after vacation presented it our Department Head, whose wife has a Masters Degree in Statistics! I told a story about what I wanted to solve, and how I intended to solve it, and the BoxPlot part was buried in there.. so it wasnt that bad. Only I (and now you, the reader) noticed my internal error.
Just to chime in as somebody who's taught data viz for years: I didn't hear anything in your story for you to be embarrassed about! Accidentally re-inventing the box plot just means that you're thinking at the same level as Mary Eleanor Spear and John Tukey, and in my book that's something to feel pretty good about. :-)
These are all great to use for data exploration. But if you're going to show data to a non-analyst audience, stick to bar charts (not histograms), trend lines, pie charts, and choropleth maps.
I'm not trying to condescend. I've tried for years to introduce more chart types, bit always received pushback on how hard they were to read.
I worked for a company that made data visualization software a couple years back (basically a Tableau type thing, but specific to one industry).
Even though our end-users were analysts, we still had "trainers" who we paid to show up in-person, on-site, to teach the end-users how to read all the types of visualizations that our software could generate. The visualizations weren't tremendously complex, but when most real-world analysts still spend most of their time in Excel, a brief explanation is still required for the end-users to actually buy-in.
I've just moved to a data scientist job where most of the work is done through Scala.
I've loved working with Scala for all the type safety and compiler helpers, but the lack of a data visualisation library makes me feel like I've lost a limb whenever I'm doing exploratory data analysis. I wonder why one hasn't been developed when it seems like there's twenty different libraries for python!
I used to use d3 for all parts of data visualizations, but lately I find myself using d3 utilities just to manipulate data to prepare for visualization, but then using a front end framework like react to render to the dom. This feels a lot cleaner than using d3 for dom manipulation.
I’ve referred to this resource in a couple of talks I’ve given on data vis. My advice is always to learn about different kinds of standard charts that are out there, then when visualizing a dataset use the best combination of words, numbers, and pictures to convey your idea. This may Or may not fit into a standard chart type.
Edward Tufte’s books have a lot of great examples for static visualizations and Bret Victor has some great examples applying these concepts to software (http://worrydream.com/MagicInk/)
I'm researching graphs used to communicate I/O error. I'm using histograms and scatterplots.
I suspect there was no small amount of work to develop the top-down, ontological framework for this catalog.
I wish there was a search feature for use cases to access the catalog from the bottom up. I would like the chance to discover additional visualizations using terms specific to my use case.
I had so much trouble building a catalog for my data visualisation.
As data scientists and software engineers, Tristan Mayer, Daniel Velasquez, and I have spent hours trying to find the most relevant datasets to do our analysis. Once we found the right one, we couldn't understand how to use it, or if we could trust it. This is painful but unfortunately too common.
We interviewed 150 companies at the end of our studies to search for solutions. Every one of them faced the problem. We worked hard for 6 months to build a solution and released the first version of our product.
Go check it out www.castordoc.com and give us feedback!
28 comments
[ 3.4 ms ] story [ 83.4 ms ] threadSince I got introduced to this simple but effective visualization (in my 30s!), I tend to use it everywhere.
Out of the “chart generation” apps listed in the catalog, only Vizzlo seems to support it: https://vizzlo.com/create/waterfall-chart (and Tableau, I suppose, but IDK if I'd count this as a simple chart generator). I did not check the listed charting libraries.
Even the popular plotting libraries in python don’t seem to have an easily configurable version. Recently tried plotly and found it plotting the categories in reverse order from the intuitive way you’d expect to see.
I’ve been meaning to script my own custom plot since Excel’s waterfalls are tedious to create and you don’t get enough control over the labeling.
1. Get an intuitive understanding of the data
2. (via Alberto Cairo) Aim to allow the user to correlate, organize and compare this data
3. Come up with a few visual means to do no. 2, and get feedback and iterate.
4. Go to the catalogs and see what work has already been done on your version of 3, and then implement.
It is amazing how many "original" charts I have created which have already been named and studied from a perceptual science point of view. But it is still important to come to a clear understanding from your own point of view of how the data fits that representation.
thank you! I am slightly embarrassed to say, I thought I had personally invented a certain version of the BoxPlot, and after vacation presented it our Department Head, whose wife has a Masters Degree in Statistics! I told a story about what I wanted to solve, and how I intended to solve it, and the BoxPlot part was buried in there.. so it wasnt that bad. Only I (and now you, the reader) noticed my internal error.
I'm not trying to condescend. I've tried for years to introduce more chart types, bit always received pushback on how hard they were to read.
Even though our end-users were analysts, we still had "trainers" who we paid to show up in-person, on-site, to teach the end-users how to read all the types of visualizations that our software could generate. The visualizations weren't tremendously complex, but when most real-world analysts still spend most of their time in Excel, a brief explanation is still required for the end-users to actually buy-in.
Whatever it is you're trying to do, there's usually already a library or framework ready and waiting for you to immediately import and start using.
https://github.com/vegas-viz/Vegas
for instance, treemap: https://datavizcatalogue.com/methods/treemap.html
I don't see echarts treemaps.
Edward Tufte’s books have a lot of great examples for static visualizations and Bret Victor has some great examples applying these concepts to software (http://worrydream.com/MagicInk/)
I suspect there was no small amount of work to develop the top-down, ontological framework for this catalog.
I wish there was a search feature for use cases to access the catalog from the bottom up. I would like the chance to discover additional visualizations using terms specific to my use case.
As data scientists and software engineers, Tristan Mayer, Daniel Velasquez, and I have spent hours trying to find the most relevant datasets to do our analysis. Once we found the right one, we couldn't understand how to use it, or if we could trust it. This is painful but unfortunately too common.
We interviewed 150 companies at the end of our studies to search for solutions. Every one of them faced the problem. We worked hard for 6 months to build a solution and released the first version of our product.
Go check it out www.castordoc.com and give us feedback!