quantlab.factor.kunquant_ts.WindowedZScore#

class quantlab.factor.kunquant_ts.WindowedZScore(v, window, v2=None)[source]#

Bases: WindowedCompositiveOp

Rolling z-score along time, (x - rolling_mean) / rolling_std.

Each symbol is standardized against its own trailing window bars; nothing is computed across symbols. NaN inputs propagate, and the first window - 1 bars are NaN until the window is full, as with every KunQuant rolling operator. No fill is applied, so fill missing values in the caller if you need them.

The operator is a composite op: KunQuant replaces it with simpler built-in operators (see decompose) while compiling.

Parameters:

vOpBase: The input series, usually a factor expression.
windowint: Number of trailing bars the mean and standard deviation use.

Examples

>>> Output(WindowedZScore(alpha(all_data), 20), "alpha001")

decompose(options)[source]#

Expand into WindowedAvg, WindowedStddev, Sub and Div.

KunQuant calls this while compiling the graph.

Parameters:

optionsdict: Decomposition options passed by KunQuant. Unused, but required by KunQuant’s composite-operator interface.

Returns:

list[OpBase]: The replacement operators, in dependency order.

Examples

It can also be called directly on an op built inside a Builder:

>>> with Builder():
...     z = WindowedZScore(Input("close"), 20)
>>> [type(op).__name__ for op in z.decompose({})]
['WindowedAvg', 'WindowedStddev', 'Sub', 'Div']