microsoft/qlib · error · ValueError

unknown metric `%s`

Error message

unknown metric `%s`

What it means

Thrown by TransformerTSModel.metric_fn, the per-epoch scoring function used for early stopping and best-model selection. Only '' and 'loss' (negated training loss) are implemented; any other `metric` string raises during the first validation pass.

Source

Thrown at qlib/contrib/model/pytorch_transformer_ts.py:100

    def mse(self, pred, label):
        loss = (pred.float() - label.float()) ** 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 train_epoch(self, data_loader):
        self.model.train()

        for data in data_loader:
            feature = data[:, :, 0:-1].to(self.device)
            label = data[:, -1, -1].to(self.device)

            pred = self.model(feature.float())  # .float()
            loss = self.loss_fn(pred, label)

            self.train_optimizer.zero_grad()
            loss.backward()
            torch.nn.utils.clip_grad_value_(self.model.parameters(), 3.0)
            self.train_optimizer.step()

    def test_epoch(self, data_loader):
        self.model.eval()

View on GitHub (pinned to 79633dd950)

Solutions

  1. Set metric='loss' or '' exactly (no extra whitespace) in the TransformerTSModel constructor.
  2. Use per-model config dicts so metric names valid in other models don't leak into this one.
  3. Subclass and override metric_fn for a custom early-stopping metric like IC.

Example fix

# before
model = TransformerTSModel(..., metric="ic")

# after
model = TransformerTSModel(..., metric="loss")
Defensive patterns

Strategy: validation

Validate before calling

metric = model_kwargs.get("metric", "")
assert metric in ("", "loss"), f"TransformerTSModel metric must be '' or 'loss', got {metric!r}"

Try / catch

try:
    model.fit(ds, valid)
except ValueError as e:
    if "unknown metric" in str(e):
        model_kwargs["metric"] = "loss"
        model = TransformerTSModel(**model_kwargs)
        model.fit(ds, valid)
    else:
        raise

Prevention

When it happens

Trigger: TransformerTSModel(..., metric='ic'|'loss '|any non-empty value other than 'loss') then fit(); metric_fn hits the raise.

Common situations: Configs copied from IC-based workflows; trailing whitespace in the metric string (e.g. 'loss ') which fails the equality; shared hyperparameter dicts across heterogeneous models.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/a124182c512145e5. Report an issue: GitHub.