FoundationAgents/MetaGPT · error · ValueError

Number of classes {num_classes} not supported

Error message

Number of classes {num_classes} not supported

What it means

Raised by ExpDataset.get_metric when the number of unique values in the target column (NumberOfClasses) does not map to a supported metric: 2 -> 'f1 binary', 2<n<=200 -> 'f1 weighted', >200 or 0 -> 'rmse'. The only remaining case, num_classes == 1, raises — a single-class target is neither a classification nor a regression target.

Source

Thrown at metagpt/ext/sela/data/dataset.py:278

            "metadata": metadata,
            "df_head": df_head_text,
        }
        return dataset_info

    def get_df_head(self, raw_df):
        return raw_df.head().to_string(index=False)

    def get_metric(self):
        dataset_info = self.get_dataset_info()
        num_classes = dataset_info["metadata"]["NumberOfClasses"]
        if num_classes == 2:
            metric = "f1 binary"
        elif 2 < num_classes <= 200:
            metric = "f1 weighted"
        elif num_classes > 200 or num_classes == 0:
            metric = "rmse"
        else:
            raise ValueError(f"Number of classes {num_classes} not supported")
        return metric

    def create_base_requirement(self):
        metric = self.get_metric()
        req = BASE_USER_REQUIREMENT.format(datasetname=self.name, target_col=self.target_col, metric=metric)
        return req

    def save_dataset(self, target_col):
        df, test_df = self.get_raw_dataset()
        if not self.check_dataset_exists() or self.force_update:
            print(f"Saving Dataset {self.name} in {self.dataset_dir}")
            self.split_and_save(df, target_col, test_df=test_df)
        else:
            print(f"Dataset {self.name} already exists")
        if not self.check_datasetinfo_exists() or self.force_update:
            print(f"Saving Dataset info for {self.name}")
            dataset_info = self.get_dataset_info()
            self.save_datasetinfo(dataset_info)

View on GitHub (pinned to 11cdf466d0)

Solutions

1. Check nunique() of the configured target column in raw/train.csv and correct target_col if it is wrong
2. If the dataset truly has one class, it is unusable for this pipeline — pick another target or dataset
3. For regression targets with few distinct values, consider that <=200 unique values will be treated as classification

Example fix

df = pd.read_csv(raw_path)
assert df[target_col].nunique() != 1, "target column is constant"

n = df[target_col].nunique()
assert n != 1, "constant target column"

def is_usable_target(df, target_col) -> bool:
    return target_col in df.columns and df[target_col].nunique() != 1

When it happens

Trigger: The target column of raw/train.csv has exactly one unique value, e.g. a constant label, wrong column selected as target_col, or an id/constant column mistaken for the target.

Common situations: Incorrect target_col configured in datasets.yaml; degenerate dataset; target column that is constant due to an upstream preprocessing bug.

Related errors

• Dataset {dataset_name} not found in config file. Available d
• Dataset {task_name} not found in config file. Available data
• Target column not provided
• Unsupported dataset: {dataset}
• Raw dataset `train.csv` not found in {raw_dir}

AI-assisted analysis of FoundationAgents/MetaGPT@11cdf466d0 (2026-08-14). Data as JSON: /api/errors/2abcaa34f0c0b841. Report an issue: GitHub.