Idea: Bivariate analyses

[sebastian]

At TeamBank the data scientist said it would be radical if he could, when seeing the analysis/distribution of ages, select a second dimension (say gender) and see a breakdown per age.

If you do multi column analyses then this is likely data you already have available. We should just think about a way of meaningfully exposing it!

[dandanlen]

Now that we have multi-column correlations / joint probabilities, this is indeed closer to being a possibility. Enabling this functionality on-demand can be done in one of two ways, one more client-heavy, the other more explorer-heavy:

1. Package up and returning the joint probability matrices for all available column combinations and then using these client-side to build the 2-dimensional breakdown.

• 😄 Requires some client-side data analysis - API consumer will need to perform some non-trivial work to generate visualisation data from raw probabilities.
• 😄 Easy to implement in the explorer.
• 😞 Requires a lot of extra data to be returned through the API (for n columns there are on the order of n^2 pairs).

2. Storing the matrices in the explorer and exposing a new API endpoint to request a multi-column data summary.

• 😄 Reuses existing explorer logic / datatypes for data analysis.
• 😄 Minimises work on the client-side.
• 😞 Requires new infrastructure in the explorer to allow persisting exploration results for subsequent requests.

[sebastian]

I vote for providing more metrics through the API and letting the client make decisions based on the data.

[sebastian]

Let's ignore it for now. We can revisit given time.
What needs to be decided in that case is what shape the data should take to be somewhat consumable.