Skip to main content
Version: Python

rolling_sum_time

rolling_sum_time creates a time-based windowed sum operator to be used in an update_by table operation. Data is windowed by reverse and forward time intervals relative to the current row, and the sum of the window is calculated.

Syntax​

rolling_sum_time(
    ts_col: str,
    cols: list[str],
    rev_time: Union[int, str],
    fwd_time: Union[int, str],
) -> UpdateByOperation

Parameters​

ParameterTypeDescription
ts_colstr

The name of the column containing timestamps.

colslist[str]

The column(s) to be operated on. These can include expressions to rename the output (e.g., NewCol = Col). If None the rolling sum is calculated for all applicable columns.

rev_timeUnion[int,str]

The look-behind window size. This can be expressed as an integer in nanoseconds or a string duration, e.g., "PT00:00:00.001" or "PTnHnMnS", where H is hour, M is minute, and S is second.

fwd_timeUnion[int,str]

The look-forward window size. This can be expressed as an integer in nanoseconds or a string duration, e.g., "PT00:00:00.001" or "PTnHnMnS", where H is hour, M is minute, and S is second.

Returns​

An UpdateByOperation to be used in an update_by table operation.

Examples​

The following example performs an update_by on the source table using three rolling_sum_time operations. Each operation gives varying rev_time and fwd_time values to show how they affect the output. The windows for each operation are as follows:

  • op_before: The window starts five seconds before the current row, and ends one second before the current row.
  • op_after: The window starts one second after the current row, and ends five seconds after of the current row.
  • op_middle: The window starts three seconds before the current row, and ends three seconds after the current row.
from deephaven.updateby import rolling_sum_time
from deephaven.time import to_j_instant
from deephaven import empty_table

base_time = to_j_instant("2023-01-01T00:00:00 ET")

source = empty_table(10).update(
    ["Timestamp = base_time + i * SECOND", "Letter = (i % 2 == 0) ? `A` : `B`", "X = i"]
)

op_before = rolling_sum_time(
    ts_col="Timestamp",
    cols=["WindowBeforeX = X"],
    rev_time=int(5e9),
    fwd_time=int(-1e9),
)
op_after = rolling_sum_time(
    ts_col="Timestamp", cols=["WindowAfterX = X"], rev_time="PT1S", fwd_time="PT5S"
)
op_middle = rolling_sum_time(
    ts_col="Timestamp", cols=["WindowMiddleX = X"], rev_time="PT3S", fwd_time="PT3S"
)

result = source.update_by(ops=[op_before, op_after, op_middle], by=["Letter"])