microsoft/qlib · error · ValueError

Empty data from dataset, please check your dataset config.

Error message

Empty data from dataset, please check your dataset config.

What it means

Thrown by HISTModel.fit when the train or valid segment prepared from the DatasetH is empty. dataset.prepare(['train','valid','test'], col_set=['feature','label'], data_key=DK_L) returned a DataFrame with zero rows for at least one of train/valid, so there is nothing to train on. The library treats this as a dataset configuration problem, not a model problem.

Source

Thrown at qlib/contrib/model/pytorch_hist.py:256

                score = self.metric_fn(pred, label)
                scores.append(score.item())

        return np.mean(losses), np.mean(scores)

    def fit(
        self,
        dataset: DatasetH,
        evals_result=dict(),
        save_path=None,
    ):
        df_train, df_valid, df_test = dataset.prepare(
            ["train", "valid", "test"],
            col_set=["feature", "label"],
            data_key=DataHandlerLP.DK_L,
        )
        if df_train.empty or df_valid.empty:
            raise ValueError("Empty data from dataset, please check your dataset config.")

        if not os.path.exists(self.stock2concept):
            url = "https://github.com/SunsetWolf/qlib_dataset/releases/download/v0/qlib_csi300_stock2concept.npy"
            urllib.request.urlretrieve(url, self.stock2concept)

        stock_index = np.load(self.stock_index, allow_pickle=True).item()
        df_train["stock_index"] = 733
        df_train["stock_index"] = df_train.index.get_level_values("instrument").map(stock_index)
        df_valid["stock_index"] = 733
        df_valid["stock_index"] = df_valid.index.get_level_values("instrument").map(stock_index)

        x_train, y_train, stock_index_train = df_train["feature"], df_train["label"], df_train["stock_index"]
        x_valid, y_valid, stock_index_valid = df_valid["feature"], df_valid["label"], df_valid["stock_index"]

        save_path = get_or_create_path(save_path)

        stop_steps = 0
        best_score = -np.inf

View on GitHub (pinned to 79633dd950)

Solutions

Inspect dataset.prepare('train', col_set=['feature','label']) and dataset.prepare('valid', ...) interactively and confirm both are non-empty
Fix the segment date ranges in your DatasetH config so they overlap the data calendar (check with D.calendar freq='day')
Verify the underlying data exists: qlib.init with the correct provider_uri and confirm instruments resolve to rows (D.list_instruments / D.features on one instrument)
Check that your label expression (e.g. Ref($close,-2)/Ref($close,-1)-1) yields non-NaN values inside the segments

Example fix

# before
dataset = DatasetH(handler, segments={"train": (20200101, 20201231)})  # data ends 2019
model.fit(dataset)

# after
dataset = DatasetH(handler, segments={"train": (20170101, 20181231), "valid": (20190101, 20191231)})
assert not dataset.prepare('train', col_set='feature').empty
model.fit(dataset)

Defensive patterns

Strategy: validation

Validate before calling

for seg in ("train", "valid"):
    df = dataset.prepare(seg, col_set=["feature", "label"], data_key=DataHandlerLP.DK_L)
    if df.empty:
        raise RuntimeError(f"segment '{seg}' is empty; check segments/instruments/data dump")

Try / catch

try:
    model.fit(dataset)
except ValueError as e:
    if "Empty data from dataset" in str(e):
        raise RuntimeError("Dataset segments empty; fix DatasetH config") from e
    raise

Prevention

Always assert train/valid prepared frames are non-empty before fit
Cross-check segment boundaries against D.calendar before building DatasetH
Run a smoke test: D.features(instruments, ['$close'], start, end) returns rows

When it happens

Trigger: Calling model.fit(dataset) where the dataset's train or valid segment has no data after handler processing: date range of segments outside the loaded data calendar, instruments with no records, or a DataHandler label processor (DK_L) that drops/filters every row.

Common situations: Wrong segment dates in DatasetH (e.g. train start after data end), missing bin/expression data for all instruments, an Alpha158 handler whose label columns are all NaN and get dropped, or forgetting to run dump_bin/dump_pickle so the underlying qlib data is empty.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/927e76d34c4d52c7. Report an issue: GitHub.