What causes "model is not fitted yet!" in microsoft/qlib?

Call fit(dataset) successfully before predict(dataset).

microsoft/qlib · error · ValueError

model is not fitted yet!

Error message

model is not fitted yet!

What it means

ALSTMTSModel.predict() first checks the self.fitted flag, which is set to True only after fit() completes its training loop. Calling predict() on a model whose fit() never ran (or raised before finishing) raises ValueError immediately, before any data is prepared. This guard prevents running an untrained, randomly-initialized network on test data.

Source

Thrown at qlib/contrib/model/pytorch_alstm_ts.py:289

                stop_steps = 0
                best_epoch = step
                best_param = copy.deepcopy(self.ALSTM_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.ALSTM_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!")

        dl_test = dataset.prepare(segment, col_set=["feature", "label"], data_key=DataHandlerLP.DK_I)
        dl_test.config(fillna_type="ffill+bfill")
        test_loader = DataLoader(dl_test, batch_size=self.batch_size, num_workers=self.n_jobs)
        self.ALSTM_model.eval()
        preds = []

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

            with torch.no_grad():
                pred = self.ALSTM_model(feature.float()).detach().cpu().numpy()

            preds.append(pred)

        return pd.Series(np.concatenate(preds), index=dl_test.get_index())

View on GitHub (pinned to 79633dd950)

Solutions

Call fit(dataset) successfully before predict(dataset).
If fit() raised earlier, fix that error and rerun fit; fitted only becomes True at the end of a clean fit().
If you meant to restore a trained model, reload its parameters and set model.fitted = True manually before predict().

Example fix

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

# after
model = ALSTMTSModel(d_feat=6)
model.fit(dataset)
model.predict(dataset)

Defensive patterns

Strategy: validation

Validate before calling

if not getattr(model, 'fitted', False):
    raise RuntimeError('ALSTMTSModel is not fitted; call fit() before predict()')

Type guard

def is_fitted(m) -> bool:
    return bool(getattr(m, 'fitted', False))

Try / catch

try:
    pred = model.predict(dataset)
except ValueError as e:
    if 'not fitted' in str(e):
        model.fit(dataset)
        pred = model.predict(dataset)
    else:
        raise

Prevention

Treat a failed fit() as 'model unusable': check model.fitted before predict in pipeline code.
Do not swallow exceptions from fit(); a half-trained model never sets fitted=True.
When restoring saved models in a new process, reload weights and set fitted=True explicitly.

When it happens

Trigger: Instantiating ALSTMTSModel and calling predict(dataset) directly; calling predict() after a fit() that raised (empty data, bad metric, CUDA OOM) so fitted was never set; re-creating the model object in a new process without restoring a fitted state.

Common situations: Notebook workflows where the fit cell errored but later cells keep running; checkpoint-reload scripts that build a fresh model and forget to load weights or refit; an exception during fit() being swallowed by a bare try/except so the caller believes training succeeded.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/71ff803d59c1dad0. Report an issue: GitHub.