Michael Shoemate
Michael Shoemate
There is interest in using the type system to enforce compatible pairings of metrics and domains.
We're typically using a metric, SymmetricDistance, that is well-defined on unordered data. On the other hand, we have at least two transformations that are order-dependent: - bounded sum - resize...
There should be an additional loosening term in aggregator relations to account for the potential worst-case change in output of the function when adding or removing input observations changes floating...
Occasionally when we write relations, we only write the relation itself-- no constant, no forward map, no backward map. This can cause chaining to fail, because a hinter does not...
We need a representation for 2-dimensional homogeneously-typed arrays. Assumptions: - row-axis represents records, column-axis features - the number of columns is not known at compile-time Use cases: - used as...
The input domain for `make_split_dataframe` (as well as some other dataframe constructors) is too vague-- `AllDomain`. SymmetricDistance isn't well-defined between strings, and we can't define it generally on strings because...
It should be possible to check if some data is a member of a transformation or measurement's input domain. My initial approach here would be to make the Transformation and...
In order to pass data between rust and bindings languages, there needs to be a data loader and unloader. There is no such data loader/unloader for vectors of nullable types....
I noticed that the HammingDistance implementation was really a restricted sensitivity proof for known N, and came up with a tighter sensitivity. There needs to be another set of constructors,...
I just had a meeting where I got stuck trying to find the SQL notebooks! I mistook `data` for a folder that contained datasets. I think `sql` is a more...