# DataDistribution internals?

**URL:** <https://community.wolfram.com/t/datadistribution-internals/16814>\
**Category:** Questions\
**Tags:** statistics-and-probability\
**Created:** [October 2, 2020, 12:49pm UTC](https://community.wolfram.com/t/datadistribution-internals/16814 "2020-10-02T12:49:29Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![mark](https://avatars.discourse-cdn.com/v4/letter/m/e9c0ed/32.png) [@mark](https://community.wolfram.com/u/mark)\
**Post date:** [October 2, 2020, 12:49pm UTC](https://community.wolfram.com/t/datadistribution-internals/16814/1 "2020-10-02T12:49:29Z")

</div>

I think I understand the internals of an “Empirical” DataDistribution (dd) with one exception: I don’t know what dd[[2,3]] (i.e., False) signifies. Can anyone tell me?

Here’s the backstory:

I’m modeling conditional categorical distributions as nested Associations. For my purposes this implementation seems to be much faster than using the built-in CategoricalDistribution. The extensive functionality that exists for Associations provides the tools to do a variety of manipulations and do them quickly.

I don’t need DataDistributions but I’m toying with the idea of using them as containers. The “weights” are then categorical distributions represented as Associations (or perhaps even DataDistributions themselves). I have no problem setting this up and I can easily convert back and forth between DataDistributions and Associations by dipping into the internals of DataDistribution. But I’m curious as to what I might be getting involved with.

BTW, I know it’s dangerous to rely on the internal structure of a built-in object. I’ve done that in the past and lived to regret it. Many years ago I relied on the internal structure of InterpolatingFunction to implement functionality that was absent at the time. If I remember correctly, the functionality was eventually added, but not before the internals were changed.

---

<div class="post-metadata">

**Author:** ![mark](https://avatars.discourse-cdn.com/v4/letter/m/e9c0ed/32.png) [@mark](https://community.wolfram.com/u/mark)\
**Post date:** [October 9, 2020, 7:35pm UTC](https://community.wolfram.com/t/datadistribution-internals/16814/2 "2020-10-09T19:35:01Z")

</div>

It looks like dd[[2,3]] is where the _bandwidth_ is stored for a “SmoothKernel” DataDistribution. So False makes sense for an “Empirical” DataDistribution, which has no bandwidth.