Better options for databases

Issue #122 new
andrew_peterson repo owner created an issue

It would be nice to turn off databases sometimes ... reading/writing to disk, especially the tarfile can be a bottleneck. It would be good to be able to have more ways to deal with this. Simplest is to just shut off non-memory databases. Second might be to re-think the tarfile; this seems like reading may be much slower than from file. Third might be to "pre-load" the whole on-disk database.

Note that the optimal behavior may differ depending on what is being done. Fingerprint derivatives are typically always expensive, so reading from disk is worth it. Neighborlists and fingerprints are often not, so it can be faster to just calculate. We should re-think how this works.

Comments (3)

  1. andrew_peterson reporter

    Another thought is that it would be useful if multiple calculators (an ensemble) could share the same data object. That is, it is loaded into memory once rather than N times for an N-member committee.

  2. Log in to comment