feat: Pivot table

[chipkent]

Add support for computing pivot tables.

Example:

from deephaven import empty_table

y = empty_table(1000).select(["Row=ii%10", "Col=(int)((ii/10) % 30)", "Sentinel=random()"]).where(
    "Sentinel > 0.3")
ys = y.sum_by(["Row", "Col"])

pvt = pivot(ys, ["Row"], "Col", "Sentinel")
pvt2 = pivot(ys, "Row", "Col", "Sentinel", lambda x: f"Col_{x}")

[jmao-denver]

@chipkent Have you looked at

deephaven-core/engine/table/src/main/java/io/deephaven/engine/table/impl/util/ColumnsToRowsTransform.java

Line 76 in 45ebf06

public class ColumnsToRowsTransform {

?

[chipkent]

I was working from the pivot table query that has been floating around. Someone would need to comment that the java way is preferred before making any changes.

[mofojed]

Side note that in your example in the PR description, it's missing the import for pivot:

from deephaven.experimental.pivot import pivot

And include an example with multiple Rows selected:

from deephaven import empty_table
from deephaven.experimental.pivot import pivot

y = empty_table(1000).select(["Row=ii%10", "Row2=ii%7", "Col=(int)((ii/10) % 30)", "Sentinel=random()"]).where(
    "Sentinel > 0.3")
ys = y.sum_by(["Row", "Row2", "Col"])

pvt = pivot(ys, ["Row", "Row2"], "Col", "Sentinel")
pvt2 = pivot(ys, ["Row", "Row2"], "Col", "Sentinel", lambda x: f"Col_{x}")

Is there a way to convert to a tree table when we have multiple rows? Such that you can expand/collapse the sections.

[mofojed]

I think confusingly worded... it's called value_to_col_name, and says it converts a "value to a column name", which in your example I would expect to be the sum_by value (e.g. values in the Sentinel column); but it's actually receiving the value from within the column_col.

[mofojed]

I would assume for true pivot tables, customers will ask to be able to pass in an array for columns and values as well, similar to Excel: https://support.microsoft.com/en-us/office/pivotby-function-de86516a-90ad-4ced-8522-3a25fac389cf