-
Notifications
You must be signed in to change notification settings - Fork 13
UI_Design
A nifty way to show off the power of Mimir would be to be able to just drop a CSV file onto Mimir and be able to start asking questions about it. As it stands, it seems like there are a few challenges in doing so...
- Is there a header row? Should row 1 define the attribute names?
- If there is no header row, how are the attribute names assigned?
- What are the types of each attribute?
- What happens to attribute values that don't conform to the proposed spec.
Questions 3 and 4 seem like the easiest to answer up front through a new type-inference lens.
Come up with a question, figure out how it would be solved end-to-end in Mimir.
A lens for log parsing ... basically a list of pairs of regular expressions and projection terms that build a relation.
A generalization of the schema matching lens that coerces a table of JSON objects into a fixed schema.
Encode a state machine (somehow?) that operates over sequences of records (e.g., for record linkage). Error states or invalid final states trigger new variables.