microsoft/qlib · error · ValueError

unknown loss `%s`

Error message

unknown loss `%s`

What it means

Raised by SFM model loss_fn in qlib/contrib/model/pytorch_sfm.py:426 when self.loss is not 'mse'. The SFM model supports only masked MSE (NaN labels are masked out); any other loss string is a ValueError thrown on the first loss evaluation in fit(). The metric_fn ('', 'loss') also delegates to loss_fn, so a bad loss breaks both.

Source

Thrown at qlib/contrib/model/pytorch_sfm.py:426

                    break

        self.logger.info("best score: %.6lf @ %d" % (best_score, best_epoch))
        self.sfm_model.load_state_dict(best_param)
        torch.save(best_param, save_path)
        if self.device != "cpu":
            torch.cuda.empty_cache()

    def mse(self, pred, label):
        loss = (pred - label) ** 2
        return torch.mean(loss)

    def loss_fn(self, pred, label):
        mask = ~torch.isnan(label)

        if self.loss == "mse":
            return self.mse(pred[mask], label[mask])

        raise ValueError("unknown loss `%s`" % self.loss)

    def metric_fn(self, pred, label):
        mask = torch.isfinite(label)

        if self.metric in ("", "loss"):
            return -self.loss_fn(pred[mask], label[mask])

        raise ValueError("unknown metric `%s`" % self.metric)

    def predict(self, dataset: DatasetH, segment: Union[Text, slice] = "test"):
        if not self.fitted:
            raise ValueError("model is not fitted yet!")

        x_test = dataset.prepare(segment, col_set="feature", data_key=DataHandlerLP.DK_I)
        index = x_test.index
        self.sfm_model.eval()
        x_values = x_test.values
        sample_num = x_values.shape[0]

View on GitHub (pinned to 79633dd950)

Solutions

  1. Set loss: 'mse' (only supported value).
  2. Subclass the SFM model and override loss_fn(), preserving the NaN mask, for custom losses.

Example fix

# before
kwargs:
  loss: huber

# after
kwargs:
  loss: mse
Defensive patterns

Strategy: validation

Validate before calling

assert config.get("loss", "mse") == "mse", "SFM model only supports loss='mse'"

Type guard

def is_supported_sfm_loss(loss: str) -> bool:
    return loss == "mse"

Try / catch

try:
    model.fit(dataset)
except ValueError as e:
    if "unknown loss" in str(e):
        raise ValueError("SFM supports only loss='mse'") from e
    raise

Prevention

When it happens

Trigger: Setting loss to anything except 'mse' in SFM model kwargs and calling fit(); the first train_epoch/test_epoch evaluation triggers the raise.

Common situations: Copying loss names from other frameworks or other qlib models; hand-editing a workflow YAML and introducing a typo like 'msae' or 'MSE'.

Related errors


AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/cc40b8883e2c0b0f. Report an issue: GitHub.