microsoft/qlib · error · ValueError

Unsupported reweighter type.

Error message

Unsupported reweighter type.

What it means

DNNModelPytorch.fit() builds sample weights per segment: reweighter=None gives uniform ones; a qlib.dataset.common.Reweighter instance has its reweight(df) called; anything else raises ValueError('Unsupported reweighter type.') while preparing train/valid data, before training starts.

Source

Thrown at qlib/contrib/model/pytorch_nn.py:216

        has_valid = "valid" in dataset.segments
        segments = ["train", "valid"]
        vars = ["x", "y", "w"]
        all_df = defaultdict(dict)  # x_train, x_valid y_train, y_valid w_train, w_valid
        all_t = defaultdict(dict)  # tensors
        for seg in segments:
            if seg in dataset.segments:
                # df_train df_valid
                df = dataset.prepare(
                    seg, col_set=["feature", "label"], data_key=self.valid_key if seg == "valid" else DataHandlerLP.DK_L
                )
                all_df["x"][seg] = df["feature"]
                all_df["y"][seg] = df["label"].copy()  # We have to use copy to remove the reference to release mem
                if reweighter is None:
                    all_df["w"][seg] = pd.DataFrame(np.ones_like(all_df["y"][seg].values), index=df.index)
                elif isinstance(reweighter, Reweighter):
                    all_df["w"][seg] = pd.DataFrame(reweighter.reweight(df))
                else:
                    raise ValueError("Unsupported reweighter type.")

                # get tensors
                for v in vars:
                    all_t[v][seg] = torch.from_numpy(all_df[v][seg].values).float()
                    # if seg == "valid": # accelerate the eval of validation
                    all_t[v][seg] = all_t[v][seg].to(self.device)  # This will consume a lot of memory !!!!

                evals_result[seg] = []
                # free memory
                del df
                del all_df["x"]
                gc.collect()

        save_path = get_or_create_path(save_path)
        stop_steps = 0
        train_loss = 0
        best_loss = np.inf
        # train

View on GitHub (pinned to 79633dd950)

Solutions

1. Subclass qlib.data.dataset.Reweighter and implement reweight(self, data_frame) returning per-sample weights; pass that instance.
2. Pass reweighter=None (or omit) when you don't need weighting.
3. For builtin weighting schemes, check qlib's existing Reweighter implementations and reuse them.

Example fix

# before
model.fit(dataset, reweighter=lambda df: df["label"] ** 2)  # ValueError: Unsupported reweighter type.

# after
from qlib.data.dataset import Reweighter

class AbsLabelReweighter(Reweighter):
    def reweight(self, data_frame):
        return data_frame["label"].abs().values

model.fit(dataset, reweighter=AbsLabelReweighter())

from qlib.data.dataset import Reweighter

if reweighter is not None and not isinstance(reweighter, Reweighter):
    raise TypeError("reweighter must be None or a qlib Reweighter instance")
model.fit(dataset, reweighter=reweighter)

from qlib.data.dataset import Reweighter

def is_valid_reweighter(rw) -> bool:
    return rw is None or isinstance(rw, Reweighter)

try:
    model.fit(dataset, reweighter=rw)
except ValueError as e:
    if "Unsupported reweighter" in str(e):
        raise TypeError("Wrap weighting logic in a qlib Reweighter subclass") from e
    raise

When it happens

Trigger: model.fit(dataset, reweighter=X) with X not None and not an instance of qlib.dataset.common.Reweighter — e.g. a weight array, pandas Series, lambda, or duck-typed custom class lacking the subclass relationship.

Common situations: Hand-rolling sample weights as arrays/callables; using a reweighter class copied from another project that doesn't subclass qlib's Reweighter; misunderstanding that the API is type-based, not duck-typed.

Related errors

• Unsupported reweighter type.
• Unsupported reweighter type.
• optimizer {} is not supported!
• loss {} is not supported!
• optimizer {} is not supported!

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/43cb2d9ce3fa0bff. Report an issue: GitHub.