Hi Ricardo,

Have you tried to re-express the Dataset object using Tabular, new in WL 14.2 ? Tabular is usually much more efficient in storage, because it uses special ways to encode the different types of columns, say dates or strings. Even if your Dataset object is deeper than two levels, it can be expressed as a Tabular object. The important thing is that it is not very ragged.

Other than that, it may be more efficient to save the Dataset object in some compressed format, like MX or WDX.

Hi José Martin,

I now have version 14.2.

Tabular works very well.
Tabular is an excellent addition to Wolfram!!

You could try exporting to Parquet and/or SQLite. For SQLite, you may need to load DatabaseLink using Get or Needs. I’ve recently used both formats successfully when working with large datasets. SQLite has the advantage of being open source with decent free viewer. And, if you don’t need all the data, you can get a selected subset. Similar operations may also be possible with Parquet, but I’ve not tried.

Hi Eric. I just want to save them to import them back in Mathematica. I alreaqdy tried the *.MX format. It worked exporting (took a long time, though), but it didn't work when I tried to importe the data back to Mathematica. One Dataset has about 13 million records.

Thank you Hans. I have used Iconize with smaller sets of data, but for very large ones (more than 13 million records) it doesn't seem to perform very well.