10/24
Greg, vic, rose, octavio
-
Octavio got a new piano on his terrace!
-
Greg got a run thru of Liz’s farm’s data - great case study (or hilarious comic re: common issues)
- E.g. Square has a different list of items/inventory than Quickbooks than online market (have to map to a single actual crop list…….that needs a common farm convention)
- Hypothesis: If you’re dealing with a significant amount of marginally complex (farm) data, with 3 or more sources, it pays & will be worth it to build a common translation layer. Even if it’s just for yourself.
-
Schema news! We are doing our first fork of the Common Farm Convention
- For Pasa - have both surveys & historical data that needed to be ingested, and wanted to use conventions to bring both in in a convergent way that aligns with their specific needs
- Not technically a fork – it’s a child convention that is still completely validatable against the common convention (may have differences in grazing, for example - some stuff that the common convention requires may not be specified in the Pasa-child-version)
- For Greg, this is a requirement – that it is fully validatable, & worth doing the extra three things to make it a child rather than a fork.
-
Rose working on ‘flippers & levers’ for conventions lite, testing if AI can make json standards
- Started with the NAPDC group, needs & gap assessment, proposed some solutions, and the one with the highest score from the group to pursue was AI translation possibilities
- Then got two classes of tasks to work on
- One is a TAP who interviewed a farmer/spoke over the phone, & now needs to fill out what happened in a data ingestion program - verbal notes into data structure
- 2 is pdf/visual data (photos etc) into data structure
- Took fairly real use case, built a fairly real schema, and now rose is making a series of ai ‘recipes’ that we’ll run against test data
- Experimental-trial Data-translation - a Hugging Face Space by our-sci
- Discovering how ‘typed’ your speaking pattern has to be to produce usefully schematized data for this purpose —
One of the planned sessions at GOAT with be ‘Failsharing’ - we have a lot of knowledge of each other and our work, operating with a similar/overlapping set of values, and want the space to be real about what hasn’t worked and why.
We’ve all picked very hard problems to work on! & out of all the times to try and predict the future, this is a very hard one! “Someday it will all work!”