https://github.com/nixtla logo
#mlforecast
Title
# mlforecast
j

juice tea

02/03/2023, 10:57 AM
hi all👋, Currently, I am facing issues in creating custom functions for the
lag_transforms
the functions from window-ops for the (seasonal)rolling mean, std etc are working fine however with the custom functions i run into issues. so far i have tried the following but i keep getting the same error
Copy code
python 
@jit
def rolling_numpy(input_array, window_size):
    return np.lib.stride_tricks.sliding_window_view(x=input_array, window_shape=window_size).mean(axis=1)

@jit
def rolling_pandas(input_array, window_size,min_samples=1):
    return pd.Series(input_array).rolling(window=window_size,min_periods=min_samples).mean().fillna(method='backfill').values


File /opt/miniconda3/envs/mdf/lib/python3.9/site-packages/mlforecast/forecast.py:195, in MLForecast.preprocess(self, data, id_col, time_col, target_col, static_features, dropna, keep_last_n, max_horizon, return_X_y)
    155 def preprocess(
    156     self,
    157     data: pd.DataFrame,
   (...)
    165     return_X_y: bool = False,
    166 ) -> Union[pd.DataFrame, Tuple[pd.DataFrame, Union[pd.Series, pd.DataFrame]]]:
    167     """Add the features to `data`.
...
    <source elided>
            lagged = shift_array(data[indptr[i] : indptr[i + 1]], lag)
            out[i] = func(lagged, *args)[-1]
            ^
question what is the proper way of creating these custom functions?