microsoft/qlib · error · ValueError

{str(e)}. \n\t{warning_info}

Error message

{str(e)}. \n\t{warning_info}

What it means

Raised by NpPairOperator._load_internal (qlib/data/ops.py) when numpy rejects the element-wise operation between the left and right feature series (e.g. np.divmod, np.arctan2 on incompatible data). The original numpy ValueError is caught, a detailed warning_info string (operator, both feature expressions, instrument, and length mismatch note) is logged at debug level, and a new ValueError chaining the original is re-raised. It almost always means the two operand features produced series that cannot be paired — most commonly different lengths.

Source

Thrown at qlib/data/ops.py:331

        check_length = isinstance(series_left, (np.ndarray, pd.Series)) and isinstance(
            series_right, (np.ndarray, pd.Series)
        )
        if check_length:
            warning_info = (
                f"Loading {instrument}: {str(self)}; np.{self.func}(series_left, series_right), "
                f"The length of series_left and series_right is different: ({len(series_left)}, {len(series_right)}), "
                f"series_left is {str(self.feature_left)}, series_right is {str(self.feature_right)}. Please check the data"
            )
        else:
            warning_info = (
                f"Loading {instrument}: {str(self)}; np.{self.func}(series_left, series_right), "
                f"series_left is {str(self.feature_left)}, series_right is {str(self.feature_right)}. Please check the data"
            )
        try:
            res = getattr(np, self.func)(series_left, series_right)
        except ValueError as e:
            get_module_logger("ops").debug(warning_info)
            raise ValueError(f"{str(e)}. \n\t{warning_info}") from e
        else:
            if check_length and len(series_left) != len(series_right):
                get_module_logger("ops").debug(warning_info)
        return res


class Power(NpPairOperator):
    """Power Operator

    Parameters
    ----------
    feature_left : Expression
        feature instance
    feature_right : Expression
        feature instance

    Returns
    ----------

View on GitHub (pinned to 79633dd950)

Solutions

1. Re-dump or repair the qlib bin data so both operand features have data for the requested instrument and range (qlib's dump_bin.py).
2. Simplify the expression: load each operand separately for the failing instrument (enable the ops debug logger: logging.getLogger('ops').setLevel(logging.DEBUG)) to see which side is empty.
3. Wrap operands or the whole pair operator in operators that tolerate NaN (e.g. If/IsNa) or use qlib's Fillna processor afterwards instead of relying on broken inputs.
4. If series lengths differ due to a custom data handler, fix the handler so both features are aligned on the same datetime index.

Example fix

# before (expression that can fail when one side has no data)
fields = ["Div($close, $volume)"]

# after (guard against missing operand data; fill after computing)
fields = ["If(IsNa($volume), NaN, Div($close, $volume))"]
# and/or in the processor list:
# {"class": "Fillna", "kwargs": {"fields_group": "feature"}}

import logging
logging.getLogger("ops").setLevel(logging.DEBUG)
try:
    df = D.features(insts, [expr], start, end)
except ValueError as e:
    if "Please check the data" in str(e):
        logger.error("feature data incomplete for %s; re-dump bins", expr)
        raise

When it happens

Trigger: Using binary element-wise expression operators such as Div(feature_left, feature_right), Sub(...), Gt(...) where one operand is NaN-only, empty, or the two series have different lengths (check_length=True path warns; the np call itself can still fail on shape mismatch). Typical with mismatched data availability between two features for one instrument.

Common situations: Expressions mixing features with different calendars or missing data coverage (e.g. a feature only defined for part of the instruments); data holes in a local bin store for one operand; stale or partially dumped qlib data where one field exists and another does not.

Related errors

• {lack_stock} doesn't have close price in qlib in the latest
• Most of samples are dropped. Please check this task: {task}
• Empty data from dataset, please check your dataset config.
• Empty data from dataset, please check your dataset config.
• The rolling window size of Skewness operation should >= 3

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/3ff4a6c3d23ea74e. Report an issue: GitHub.