Cluster analysis in a base

[mademi]

Hello everyone,

I have been using airtable for a while now, to collect data about certain objects in a base with grid-view. Each line represents one object and I have different rows, characterizing the objects in different dimensions.
Most of the rows are “multiple select” options with a max of 25 options which are freely combinable.
So basically I have a base with around 120 instances and each instance is characterized in different dimensions.

My interest now is to find clusters in a certain dimension which means I want to learn which options in a dimension have often been selected together. Is there a way of getting this done in airtable? The standard “group by” feature is working for dimensions (rows) where not many options are selectable but for dimension with more options it does not work in a sufficient way.

I have difficulties in finding out if there is an app which could support my intentions. Also i have a free plan for now and can not just test around with different apps. Does anyone know if there are apps that could help me with that analysis in airtable? If I knew there is an app I would consider getting a pro plan.

Thanks in advance for your help!

[Jeremy_Oglesby]

Hi @mademi ,

I’m quite certain that what you are trying to achieve will be either impossible, or else extremely ugly and onerous, if you were to try to use just Airtable’s basic interface.

It should be possible, however, with one of three potential options utilizing apps (thus requiring a “Pro” plan):

1. Scripting App - this one I am the most sure about; you can definitely pass all your records to a Scripting block, and use JavaScript to analyze them and build a tabular output that would highlight your most frequent confluences; the output won’t be pretty, but it will be functional
2. Vega-lite App - I’m not as sure about this one, but I think it could be possible to build visuals, such as a bubble chart, that aggregate the confluences and display their frequency in a much more visually pleasing manner than the Scripting App output; I’m not that great at building these visualizations yet, though, so to me, this looks like a lot of heavy work utilizing some of the more complex aspects of Vega-lite’s options
3. Custom App - this one I’m also sure of, because the possibilities are wide open; however, this also requires an experienced React.js developer to build a fairly complex custom app

[augmented]

Back when custom blocks (now apps) first came out, I created one which imported a javascript ML library in order to easily do linear regression (as a quick POC) and create a plot. So, if you want to go that route of rolling your own, you can. There’s also github with tons of apps to look through. Maybe something will work!

github.com

Airtable/blocks

rk/dependents

Contribute to Airtable/blocks development by creating an account on GitHub.

[Bill_French]

Jeremy_Oglesby:

with one of three potential options

And possibly two of three. I have a hunch you might get some traction on this requirement by using the scripting block to shape a viz-ready data set (i.e., to the heavy lifting in javascript) and then render the results in Vega-Lite.