What causes "invalid argument type for `alpha`" in microsoft/qlib?

Set alpha to 'lw' or 'oas' to use a built-in estimator

microsoft/qlib · error · TypeError

invalid argument type for `alpha`

Error message

invalid argument type for `alpha`

What it means

ShrinkRiskModel.__init__ (qlib/model/riskmodel/shrink.py:69) validates the `alpha` argument: it must be either the string 'lw' or 'oas' (Ledoit-Wolf / Oracle Approximating Shrinkage estimators) or a float in [0, 1]. Any other type (int out of range, None, list, dict) raises TypeError('invalid argument type for `alpha`'). Note plain Python int is also rejected since only float/np.floating are accepted.

Source

Thrown at qlib/model/riskmodel/shrink.py:69

    TGT_CONST_CORR = "const_corr"
    TGT_SINGLE_FACTOR = "single_factor"

    def __init__(self, alpha: Union[str, float] = 0.0, target: Union[str, np.ndarray] = "const_var", **kwargs):
        """
        Args:
            alpha (str or float): shrinking parameter or estimator (`lw`/`oas`)
            target (str or np.ndarray): shrinking target (`const_var`/`const_corr`/`single_factor`)
            kwargs: see `RiskModel` for more information
        """
        super().__init__(**kwargs)

        # alpha
        if isinstance(alpha, str):
            assert alpha in [self.SHR_LW, self.SHR_OAS], f"shrinking method `{alpha}` is not supported"
        elif isinstance(alpha, (float, np.floating)):
            assert 0 <= alpha <= 1, "alpha should be between [0, 1]"
        else:
            raise TypeError("invalid argument type for `alpha`")
        self.alpha = alpha

        # target
        if isinstance(target, str):
            assert target in [
                self.TGT_CONST_VAR,
                self.TGT_CONST_CORR,
                self.TGT_SINGLE_FACTOR,
            ], f"shrinking target `{target} is not supported"
        elif isinstance(target, np.ndarray):
            pass
        else:
            raise TypeError("invalid argument type for `target`")
        if alpha == self.SHR_OAS and target != self.TGT_CONST_VAR:
            raise NotImplementedError("currently `oas` can only support `const_var` as target")
        self.target = target

    def _predict(self, X: np.ndarray) -> np.ndarray:

View on GitHub (pinned to 79633dd950)

Solutions

Set alpha to 'lw' or 'oas' to use a built-in estimator
Set alpha to a float between 0 and 1 inclusive, e.g. alpha=0.1 (write 0.1, not an int)
Check your YAML/JSON config for null/integer alpha values and coerce them to float or estimator strings

Example fix

# before
model = ShrinkRiskModel(alpha=None)  # TypeError

# after
model = ShrinkRiskModel(alpha='lw')
# or fixed float
model = ShrinkRiskModel(alpha=0.1)

Defensive patterns

Strategy: validation

Validate before calling

import numbers
assert alpha in ('lw', 'oas') or (isinstance(alpha, (float, np.floating)) and 0 <= alpha <= 1), f'invalid alpha: {alpha!r}'

Type guard

def is_valid_alpha(alpha) -> bool:
    return alpha in ('lw', 'oas') or (isinstance(alpha, (float, np.floating)) and 0.0 <= alpha <= 1.0)

Try / catch

try:
    model = ShrinkRiskModel(alpha=alpha)
except TypeError as e:
    raise ValueError(f"alpha must be 'lw'/'oas' or float in [0,1], got {alpha!r}") from e

Prevention

In YAML configs write alpha as 0.1 (float) or an estimator string, never null/int
Validate risk-model hyperparameters at config load time

When it happens

Trigger: Constructing ShrinkRiskModel(alpha=None), alpha=1 (int), alpha=[0.1, 0.2], or a misspelled estimator string; passing alpha as an unsupported estimator name like 'james_stein' (that hits the assert instead).

Common situations: Config files with alpha: null or integer alpha (e.g. alpha: 1); users assuming alpha accepts any sklearn covariance estimator name; copy-pasting configs between risk model classes with different alpha semantics.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/31094f44ffc7418a. Report an issue: GitHub.