microsoft/qlib · error · ValueError

The rolling window size of Skewness operation should >= 3

Error message

The rolling window size of Skewness operation should >= 3

What it means

Raised by the Skew rolling operator (qlib/data/ops.py) at expression construction when its window size N is nonzero and smaller than 3. Statistical skewness is undefined for fewer than 3 samples, so qlib rejects such windows immediately. N=0 in qlib convention means an expanding window over all history and is allowed.

Source

Thrown at qlib/data/ops.py:925

class Skew(Rolling):
    """Rolling Skewness

    Parameters
    ----------
    feature : Expression
        feature instance
    N : int
        rolling window size

    Returns
    ----------
    Expression
        a feature instance with rolling skewness
    """

    def __init__(self, feature, N):
        if N != 0 and N < 3:
            raise ValueError("The rolling window size of Skewness operation should >= 3")
        super(Skew, self).__init__(feature, N, "skew")


class Kurt(Rolling):
    """Rolling Kurtosis

    Parameters
    ----------
    feature : Expression
        feature instance
    N : int
        rolling window size

    Returns
    ----------
    Expression
        a feature instance with rolling kurtosis
    """

View on GitHub (pinned to 79633dd950)

Solutions

Set N >= 3, e.g. Skew($close, 5).
If you want an expanding (all-history) window, use N=0, which is explicitly allowed.
Filter window-size candidate lists to exclude N in {1,2} before generating expressions.

Example fix

# before
expr = "Skew($close, 2)"

# after
expr = "Skew($close, 5)"  # any N >= 3, or N=0 for expanding window

Defensive patterns

Strategy: validation

Validate before calling

if N != 0 and N < 3:
    raise ValueError("Skew window must be 0 (expanding) or >= 3")

Type guard

def is_valid_skew_window(n: int) -> bool:
    return n == 0 or n >= 3

Prevention

Filter hyperparameter grids: windows = [w for w in windows if w == 0 or w >= 3].
Generate feature expressions from validated parameter lists, not free-form strings.

When it happens

Trigger: Building a feature expression like Skew($close, 2) or Skew($change, 1) — any N in {1, 2}. Evaluated when the expression string is parsed/instantiated, before any data is loaded.

Common situations: Parameter sweeps over window sizes (e.g. trying N=1..10) that include invalid small values; hand-written feature lists with short windows; porting formulas from libraries with different minimums.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/48a0e14642b0ba4c. Report an issue: GitHub.