What causes "optimizer {} is not supported!" in microsoft/qlib?

microsoft/qlib · error · NotImplementedError

optimizer {} is not supported!

Error message

optimizer {} is not supported!

What it means

DNNModelPt.fit() selects the training optimizer from the optimizer hyperparameter with two branches: 'adam' -> torch.optim.Adam and 'gd' -> torch.optim.SGD, both including weight_decay. Any other string raises NotImplementedError before the ReduceLROnPlateau scheduler is attached. This generic feed-forward model otherwise follows the same optimizer switch pattern as the GATs/ALSTM contrib models.

Source

Thrown at qlib/contrib/model/pytorch_general_nn.py:138

                seed,
                pt_model_uri,
                pt_model_kwargs,
            )
        )

        if self.seed is not None:
            np.random.seed(self.seed)
            torch.manual_seed(self.seed)

        self.logger.info("model:\n{:}".format(self.dnn_model))
        self.logger.info("model size: {:.4f} MB".format(count_parameters(self.dnn_model)))

        if optimizer.lower() == "adam":
            self.train_optimizer = optim.Adam(self.dnn_model.parameters(), lr=self.lr, weight_decay=weight_decay)
        elif optimizer.lower() == "gd":
            self.train_optimizer = optim.SGD(self.dnn_model.parameters(), lr=self.lr, weight_decay=weight_decay)
        else:
            raise NotImplementedError("optimizer {} is not supported!".format(optimizer))

        # === ReduceLROnPlateau learning rate scheduler ===
        self.lr_scheduler = ReduceLROnPlateau(
            self.train_optimizer, mode="min", factor=0.5, patience=5, min_lr=1e-6, threshold=1e-5
        )
        self.fitted = False
        self.dnn_model.to(self.device)

    @property
    def use_gpu(self):
        return self.device != torch.device("cpu")

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

    def loss_fn(self, pred, label, weight=None):
        mask = ~torch.isnan(label)

View on GitHub (pinned to 79633dd950)

Solutions

Use 'adam' or 'gd'.
Change 'sgd' to 'gd'.
Subclass DNNModelPt and add a branch constructing the desired torch.optim class with weight_decay.

Example fix

# before
model = DNNModelPt(optimizer='sgd')

# after
model = DNNModelPt(optimizer='gd')

Defensive patterns

Strategy: validation

Validate before calling

assert optimizer.lower() in ('adam', 'gd'), f"optimizer must be 'adam' or 'gd', got {optimizer!r}"

Type guard

def is_supported_optimizer(name: str) -> bool:
    return name.lower() in ('adam', 'gd')

Try / catch

try:
    model.fit(dataset)
except NotImplementedError as e:
    if 'optimizer' in str(e):
        model = DNNModelPt(optimizer='adam')
        model.fit(dataset)
    else:
        raise

Prevention

Use 'gd' rather than 'sgd' for the plain-SGD option.
Validate optimizer names against the allowlist before fit in tuning loops.
Subclass DNNModelPt early if you need AdamW or other optimizers with weight_decay.

When it happens

Trigger: DNNModelPt(...).fit(dataset) with optimizer='sgd', 'adamw', 'rmsprop', 'adagrad', or any string besides 'adam'/'gd'.

Common situations: The classic 'sgd' vs 'gd' mismatch; hyperparameter tuning scripts that enumerate torch optimizer names; wanting AdamW for proper weight decay handling in the general NN workflow.

Related errors

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/e523b71cbea3a662. Report an issue: GitHub.