microsoft/qlib · error · ValueError

not implemented yet

Error message

not implemented yet

What it means

Thrown by DEnsembleModel.get_loss when self.loss is not "mse". Double Ensemble's sample-reweighting and feature-selection modules need per-sample training loss; only mean squared error is implemented, so any other loss string (e.g. "mae", a LightGBM objective name) is rejected during fit.

Source

Thrown at qlib/contrib/model/double_ensemble.py:225

        g["g_value"].replace(np.nan, 0, inplace=True)

        # divide features into bins_fs bins
        g["bins"] = pd.cut(g["g_value"], self.bins_fs)

        # randomly sample features from bins to construct the new features
        res_feat = []
        sorted_bins = sorted(g["bins"].unique(), reverse=True)
        for i_b, b in enumerate(sorted_bins):
            b_feat = features[g["bins"] == b]
            num_feat = int(np.ceil(self.sample_ratios[i_b] * len(b_feat)))
            res_feat = res_feat + np.random.choice(b_feat, size=num_feat, replace=False).tolist()
        return pd.Index(set(res_feat))

    def get_loss(self, label, pred):
        if self.loss == "mse":
            return (label - pred) ** 2
        else:
            raise ValueError("not implemented yet")

    def retrieve_loss_curve(self, model, df_train, features):
        if self.base_model == "gbm":
            num_trees = model.num_trees()
            x_train, y_train = df_train["feature"].loc[:, features], df_train["label"]
            # Lightgbm need 1D array as its label
            if y_train.values.ndim == 2 and y_train.values.shape[1] == 1:
                y_train = np.squeeze(y_train.values)
            else:
                raise ValueError("LightGBM doesn't support multi-label training")

            N = x_train.shape[0]
            loss_curve = pd.DataFrame(np.zeros((N, num_trees)))
            pred_tree = np.zeros(N, dtype=float)
            for i_tree in range(num_trees):
                pred_tree += model.predict(x_train.values, start_iteration=i_tree, num_iteration=1)
                loss_curve.iloc[:, i_tree] = self.get_loss(y_train, pred_tree)
        else:

View on GitHub (pinned to 79633dd950)

Solutions

1. Use loss="mse" (the supported value)
2. For classification-style tasks, encode the target appropriately and keep mse, or pick a different model class

Example fix

# before
model = DEnsembleModel(loss="mae")

# after
model = DEnsembleModel(loss="mse")

assert loss == "mse", "DEnsembleModel supports only loss='mse' (needed for sample reweighting and feature selection)"

When it happens

Trigger: Constructing DEnsembleModel(loss="mae") or loss="binary" etc.; loss is forwarded into params as the objective, but the reweighting math only knows "mse".

Common situations: Treating loss as a free LightGBM objective parameter; porting a classification objective into Double Ensemble.

Related errors

• nfs-common is not found, please install it by execute: sudo
• Mount failed: requires sudo or permission denied
• mount {provider_uri} on {mount_path} error! Command error
• Mount failed: {e.stderr}
• We can't find the project path

AI-assisted analysis of microsoft/qlib@79633dd950 (2026-08-15). Data as JSON: /api/errors/1f0368bed047b793. Report an issue: GitHub.