explode_time_series

ExplodeTimeSeries(features)

Bases: Transform

Transform that explodes nested time series data into individual rows.

Each time series is represented as a list of structs, where each struct contains a timestamp and a value. The value is itself a struct holding multiple feature values at that timestamp. This transform expands the list into separate rows, then unnests the nested structs into columns.

Example

Input DataFrame: ┌─────┬────────────────────────────────────────┐ │ id │ series │ │ --- │ --- │ │ i64 │ list[struct[timestamp: str, value]] │ ├─────┼────────────────────────────────────────┤ │ 1 │ [{t1, {a=1, b=10}}, {t2, {a=2, b=20}}] │ │ 2 │ [{t3, {a=3, b=30}}, {t4, {a=4, b=40}}] │ └─────┴────────────────────────────────────────┘

After applying ExplodeTimeSeries("series"): ┌─────┬────────────┬─────┬─────┐ │ id │ timestamp │ a │ b │ │ --- │ --- │ --- │ --- │ │ i64 │ str │ i64 │ i64 │ ├─────┼────────────┼─────┼─────┤ │ 1 │ t1 │ 1 │ 10 │ │ 1 │ t2 │ 2 │ 20 │ │ 2 │ t3 │ 3 │ 30 │ │ 2 │ t4 │ 4 │ 40 │ └─────┴────────────┴─────┴─────┘

Source code in src/flowcean/polars/transforms/explode_time_series.py

def __init__(self, features: ColumnNameOrSelector) -> None:
    super().__init__()
    self.features = features