UMAP on large data set

[jovanaarsic]

Dear all,
I need a little bit of help. 🤗
The task I have to solve is the clustering task on a large data set (aka. 6 mil rows/20 cols) with a mixtured data types.

The primary idea was to apply UMAP->HDBSCAN.
In accordance with my understanding, I've applied UMAP on subset. With idea of transformation the rest of data set.
OOM popped up during the process of fitting the umap on 100K.

The set of questions related to UMAP would be:

1. Is there possibility of applying the UMAP on entire dataset?
2. Is this expected behaviour, or some parameters, metrics used or data itself are the core of the issue?
3. As I work on instances (Data Bricks), do you have recommendation in terms of memory performances to be able to fit on subsumple and transform the rest of the data in reasonable amount of time?

Additionally,
Does anyone have experience with a similar problem, solved with different approach?

All the advices, suggestions and experiences are welcoming. 🤗