microsoft/qlib · error · ValueError

unknown metric `%s`

Error message

unknown metric `%s`

What it means

Raised by SANDWICH model metric_fn in qlib/contrib/model/pytorch_sandwich.py:249 when self.metric is not '' or 'loss'. Only two tokens are accepted: empty string and 'loss', both meaning 'use negative training loss as the early-stopping metric'. Any other metric name (e.g. 'ic') raises ValueError because no other metric is implemented.

Source

Thrown at qlib/contrib/model/pytorch_sandwich.py:249

    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 train_epoch(self, x_train, y_train):
        x_train_values = x_train.values
        y_train_values = np.squeeze(y_train.values)
        self.sandwich_model.train()

        indices = np.arange(len(x_train_values))
        np.random.shuffle(indices)

        for i in range(len(indices))[:: self.batch_size]:
            if len(indices) - i < self.batch_size:
                break

            feature = torch.from_numpy(x_train_values[indices[i : i + self.batch_size]]).float().to(self.device)
            label = torch.from_numpy(y_train_values[indices[i : i + self.batch_size]]).float().to(self.device)

            pred = self.sandwich_model(feature)
            loss = self.loss_fn(pred, label)

View on GitHub (pinned to 79633dd950)

Solutions

  1. Set metric: '' (or omit it) or metric: 'loss' in the model kwargs.
  2. If you need IC-based early stopping, subclass and override metric_fn() with an IC computation on masked labels.

Example fix

# before
kwargs:
  metric: ic

# after
kwargs:
  metric: loss   # or '' (default: negative training loss)
Defensive patterns

Strategy: validation

Validate before calling

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

Type guard

def is_supported_sandwich_metric(metric: str) -> bool:
    return metric in ("", "loss")

Try / catch

try:
    model.fit(dataset)
except ValueError as e:
    if "unknown metric" in str(e):
        raise ValueError("Set metric='' or 'loss'; override metric_fn() for custom metrics") from e
    raise

Prevention

When it happens

Trigger: Passing metric='ic', metric='auc', or any non-empty string other than 'loss' to the sandwich model constructor and calling fit(); the raise fires on the first validation evaluation inside the training loop.

Common situations: Copying metric: 'ic' from a DNNModelPytorch/qlib workflow config into the sandwich model; assuming custom metric names from qlib's signal analysis are accepted here.

Related errors


AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/0f0df4a84999ec6d. Report an issue: GitHub.