UnboundLocalError: local variable "conditional_cols" referenced before assignment when model trained with argument missing_imputation=False and there are seasonalities

[frlm]

neural_prophet/neuralprophet/data/process.py

Line 435 in 18e3e4d

def _handle_missing_data(

We observed a bug when we try to predict a future dataframe using a trained model with missing_imputation = False and seasonalities set. The variable conditional_cols is used inside an if statement when dropped_trailing_y and predicting are true, but this variable is defined only if config_missing.impute_missing is True

    if config_missing.impute_missing:
        # impute missing values
        data_columns = []
        if n_lags > 0:
            data_columns.append("y")
        if config_lagged_regressors is not None:
            data_columns.extend(config_lagged_regressors.keys())
        if config_regressors is not None and config_regressors.regressors is not None:
            data_columns.extend(config_regressors.regressors.keys())
        if config_events is not None:
            data_columns.extend(config_events.keys())
        conditional_cols = []
        if config_seasonality is not None:
            conditional_cols = list(
                set(
                    [
                        value.condition_name
                        for key, value in config_seasonality.periods.items()
                        if value.condition_name is not None
                    ]
                )
            )
            data_columns.extend(conditional_cols)
        for column in data_columns:
            sum_na = df[column].isna().sum()
            if sum_na > 0:
                log.warning(f"{sum_na} missing values in column {column} were detected in total. ")
                # use 0 substitution for holidays and events missing values
                if config_events is not None and column in config_events.keys():
                    df[column].fillna(0, inplace=True)
                    remaining_na = 0
                else:
                    df.loc[:, column], remaining_na = df_utils.fill_linear_then_rolling_avg(
                        df[column],
                        limit_linear=config_missing.impute_linear,
                        rolling=config_missing.impute_rolling,
                    )
                log.info(f"{sum_na - remaining_na} NaN values in column {column} were auto-imputed.")
                if remaining_na > 0:
                    log.warning(
                        f"More than {2 * config_missing.impute_linear + config_missing.impute_rolling} consecutive \
                            missing values encountered in column {column}. "
                        f"{remaining_na} NA remain after auto-imputation. "
                    )
    _if dropped_trailing_y and predicting:
        # add trailing y values again if in predict mode
        df = pd.concat([df, df_to_add])
        if config_seasonality is not None and len(conditional_cols) > 0:
            df[conditional_cols] = df[conditional_cols].ffill()  # type: ignore_