But They Had Become Radical Lesbian Nuns
Sep 5, 2021 · 2 minute readAfter a fairly long wait, there’s a new update to But This Was A Fantasy. Firstly, The Century of The Self is now added to the index, so you can now type “John Denver” or “Freud” and get an appropriate result. That brings us up to 2002 — next up, the mirrors of The Power of Nightmares and The Trap.
However, there’s another new feature in this weekend’s update — the ability to filter results by series. Yes, you can now limit your search of “mid-century furniture” to just episodes from Pandora’s Box if you really want to. My original plan for this was to build separate FAISS indexes for each series (and I may eventually do this still). However, I got bored on Friday afternoon, and given that I am having some thoughts around bloom filters right now (exciting times!!), I threw together a bunch of simple filters, one per series, and now there’s filtering. Each filter is only about 30k - you could easily fit one inside a ZX Spectrum…
Anyhow, one of the consequences of adding filtering is that it’s now possible to get zero results from a series search — this wouldn’t be an issue if I were using separate indexes per series, but what I’m doing right now is pulling out the top 50 results from the main index and doing the filter check afterwards. So if none of those results are from the series you are filtering on…you get nothing. I might add some retry logic for that in the future that’ll pull out more candidates if the initial search comes up empty. Anything to get around building those multiple indexes.
At this point, there’s not much left to do except add the outstanding series; I might move the images to a Google Cloud Storage bucket instead of bundling them all in the container. It was quick and easy to start with, but now I’m uploading about 6Gb of images on every push, which makes changes a lot more heavyweight than they need to be. Otherwise, just keep checking back for more content…