microsoft/qlib · error · ValueError

model is not fitted yet!

Error message

model is not fitted yet!

What it means

Thrown by TransformerModel.predict when the `fitted` flag is False. fitted is set only after a successful fit loop (best weights reloaded, checkpoint saved), so prediction on an untrained or failed-to-train Transformer model is blocked.

Source

Thrown at qlib/contrib/model/pytorch_transformer.py:217

                stop_steps = 0
                best_epoch = step
                best_param = copy.deepcopy(self.model.state_dict())
            else:
                stop_steps += 1
                if stop_steps >= self.early_stop:
                    self.logger.info("early stop")
                    break

        self.logger.info("best score: %.6lf @ %d" % (best_score, best_epoch))
        self.model.load_state_dict(best_param)
        torch.save(best_param, save_path)

        if self.use_gpu:
            torch.cuda.empty_cache()

    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.model.eval()
        x_values = x_test.values
        sample_num = x_values.shape[0]
        preds = []

        for begin in range(sample_num)[:: self.batch_size]:
            if sample_num - begin < self.batch_size:
                end = sample_num
            else:
                end = begin + self.batch_size

            x_batch = torch.from_numpy(x_values[begin:end]).float().to(self.device)

            with torch.no_grad():
                pred = self.model(x_batch).detach().cpu().numpy()

View on GitHub (pinned to 79633dd950)

Solutions

  1. Call model.fit(dataset, evals_result) successfully first, then predict(dataset, segment='test').
  2. Guard evaluation stages on fit success (try/except around fit, skip predict on failure) and fix the root fit error.
  3. To restore a trained model: torch.load(save_path) → model.load_state_dict(...) → set model.fitted = True before predict.

Example fix

# before
model = TransformerModel(d_feat=6)
model.predict(dataset)  # ValueError

# after
model = TransformerModel(d_feat=6)
model.fit(dataset, evals_result)
model.predict(dataset)
Defensive patterns

Strategy: validation

Validate before calling

if not getattr(model, "fitted", False):
    raise RuntimeError("TransformerModel not fitted — call fit(dataset, evals_result) first")

Type guard

def transformer_ready(model) -> bool:
    return getattr(model, "fitted", False) and getattr(model, "model", None) is not None

Try / catch

try:
    preds = model.predict(dataset)
except ValueError as e:
    if "not fitted" in str(e):
        model.fit(dataset, evals_result)
        preds = model.predict(dataset)
    else:
        raise

Prevention

When it happens

Trigger: TransformerModel.predict(dataset) without a prior successful fit(dataset, evals_result); or after a fit that errored mid-training (OOM, NaN loss, early crash) leaving fitted=False.

Common situations: Rolling-refit scripts where one fold's fit fails but predict still runs; notebook re-execution of only the predict cell; expecting a saved .bin checkpoint alone to make a fresh model instance predict-ready.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/39a9d33bf6df58ef. Report an issue: GitHub.