Expanding window instead of fixed one to compute indicators

[fingoldo]

Hi, is it possible/easy to implement expanding window?
Currently, indicators output null until "timeperiod" observations are reached:

import polars as pl
import polars_talib as plta

# Sample multi-ticker data
df = pl.DataFrame({
    "ticker": ["AAPL", "AAPL", "AAPL", "AAPL", "AAPL", "AAPL",
               "GOOG", "GOOG", "GOOG", "GOOG", "GOOG", "GOOG"],
    "close": [150.0, 152.0, 151.0, 153.0, 155.0, 154.0,
              2800.0, 2820.0, 2810.0, 2830.0, 2850.0, 2840.0],
})

# The actual answer - use over() for per-group computation
result = df.with_columns(
    plta.sma(pl.col("close"), timeperiod=5).over("ticker").alias("sma_5")
)

print(result)

But i'd rather prefer them to use less observations and output something meaningful instead of null.
I.e., run computations over expanding window up to "timeperiod" observations, rather than waiting for strictly "timeperiod" observations.

[Yvictor]

Thanks for the suggestion. I would classify this as a feature request rather than a bug in the current fixed-window indicators.

The current functions mostly mirror TA-Lib semantics. For most indicators, TA-Lib has a fixed lookback and returns no meaningful value before enough observations are available. If the existing sma(..., timeperiod=5) behavior were changed to expanding-window semantics, the results would no longer match TA-Lib / ta-lib-python and it could break existing users.

If this is added, I think it should be exposed as an explicit option or a separate helper, for example a min_periods / expanding mode, rather than changing the default behavior. For now, Polars rolling/expanding expressions with min_periods are probably the best workaround for this kind of calculation.