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

Raised by TabNet model fit() in qlib/contrib/model/pytorch_tabnet.py:172 when the train or valid DataFrame from dataset.prepare(['train','valid']) is empty. Standard qlib data guard: it fires before NaN-filling and training, indicating the handler/dataset config produced zero rows for a required segment.

Source

Thrown at qlib/contrib/model/pytorch_tabnet.py:172

        evals_result=dict(),
        save_path=None,
    ):
        if self.pretrain:
            # there is a  pretrained model, load the model
            self.logger.info("Pretrain...")
            self.pretrain_fn(dataset, self.pretrain_file)
            self.logger.info("Load Pretrain model")
            self.tabnet_model.load_state_dict(torch.load(self.pretrain_file, map_location=self.device))

        # adding one more linear layer to fit the final output dimension
        self.tabnet_model = FinetuneModel(self.out_dim, self.final_out_dim, self.tabnet_model).to(self.device)
        df_train, df_valid = dataset.prepare(
            ["train", "valid"],
            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.")
        df_train.fillna(df_train.mean(), inplace=True)
        x_train, y_train = df_train["feature"], df_train["label"]
        x_valid, y_valid = df_valid["feature"], df_valid["label"]
        save_path = get_or_create_path(save_path)

        stop_steps = 0
        train_loss = 0
        best_score = -np.inf
        best_epoch = 0
        evals_result["train"] = []
        evals_result["valid"] = []

        self.logger.info("training...")
        self.fitted = True

        for epoch_idx in range(self.n_epochs):
            self.logger.info("epoch: %s" % (epoch_idx))
            self.logger.info("training...")

View on GitHub (pinned to 79633dd950)

Solutions

Check which segment is empty: dataset.prepare('train', col_set=['feature','label']).empty and same for 'valid'.
Align segment date ranges with the data calendar and instrument universe.
Verify qlib.init(provider_uri=...) and that the dump contains your instruments.
If using named segments, ensure DatasetH.segments defines 'train' and 'valid'.

Example fix

# before
train: [2008-01-01, 2024-12-31]   # dump ends 2020

# after
train: [2008-01-01, 2014-12-31]
valid: [2015-01-01, 2020-12-31]

Defensive patterns

Strategy: validation

Validate before calling

for seg in ("train", "valid"):
    df = dataset.prepare(seg, col_set=["feature", "label"], data_key="learn")
    if df.empty:
        raise ValueError(f"TabNet fit(): segment '{seg}' empty — check segment config and data dump")

Try / catch

try:
    model.fit(dataset)
except ValueError as e:
    if "Empty data" in str(e):
        print({s: dataset.prepare(s, col_set=["feature", "label"]).shape for s in ("train", "valid")})
        raise
    raise

Prevention

Verify train/valid segments are non-empty before fit(), especially when pretrain segments are also configured.
Keep segment dates inside the dumped calendar and instruments present in the dump.
Smoke-test the handler expression on a small date window first.

When it happens

Trigger: Segment dates outside the dumped data calendar; handler expressions filtering out all rows; wrong provider_uri / no qlib.init; also fires after the optional pretrain branch, so pretrain may succeed while the finetune 'train'/'valid' segments are empty.

Common situations: Using pretrain segments that exist but train/valid boundaries misconfigured; examples run against a market different from the downloaded dump; custom instruments missing from the dump.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/ee762091a18c982d. Report an issue: GitHub.