Open jjbrosnan opened 1 year ago
https://deephaven.io/core/docs/how-to-guides/use-deephaven-learn/ suggests that
For real-time tables, only the rows of the table that changed during the most recent update cycle are separated into batches and processed.
In the python library though, we create:
.update(formulas=[f"{future_offset} = __computer.compute(k)", ])
but this does not work with general updating tables.
I am experiencing this issue too. It's quite problematic, as I am not able to use learn.learn on live aggregated data
I believe the issue here is not simply "k
isn't usable because my table isn't append-only", but "dh.learn cannot operate on tables which are not append-only" - can you modify your table setup so that only new rows are created?
This problem has been assessed. It has been determined that the best course of action is to wait for Deephaven's multidimensional array support to be delivered. Until then, it is recommended that users formulate their AI input tables in an append-only way.
It is possible to set io.deephaven.engine.table.impl.select.AbstractFormulaColumn.allowUnsafeRefreshingFormulas=true
. This would likely allow a query to run, but it may produce inaccurate results, so it is discouraged.
@chipkent thank you very much for looking into this! When are the multidimensional arrays expected to be introduced in Deephaven Core?
The design work has been done for the feature, but no developer resources have been allocated yet.
Description
Try using deephaven.learn on data and it will return an error:
This easily fixable if tables are append-only, otherwise it's no good.
Steps to reproduce
Use deephaven.learn. Ping me if you want a script.
Expected results
A new table with additional columns containing ML results.
Actual results
The error above.
Additional details and attachments
If applicable, add any additional screenshots, logs, or other attachments to help explain your problem.
Versions