Skip to main content

preprocessing.to_model

format_xarray_for_rnn​

def format_xarray_for_rnn(
    ds: xr.Dataset,
    read_from_variable: str = "position_processed",
    keypoints: list[str] | None = None,
    extra_features: list[str] | None = None
) -> Tuple[np.ndarray, Dict[str, Any]]

Formats the xarray dataset for use VAME's RNN model:

  • The x and y coordinates of the centered_reference_keypoint are excluded.
  • The x coordinate of the orientation_reference_keypoint is excluded.
  • The remaining data is flattened and transposed.
  • Optionally, pre-computed scalar features (speed, head direction, ...) are appended as additional rows after the pose-derived features.

Parameters

  • ds (xr.Dataset): The xarray dataset to format.
  • read_from_variable (str, default="position_processed"): The variable to read from the dataset.
  • keypoints (list[str] | None, optional): A list of keypoints to include in the output. If None, all keypoints are included. If provided, only the specified keypoints will be included in the output.
  • extra_features (list[str] | None, optional): Names of pre-computed scalar features to append after the pose-derived features. Each must exist as a data variable in ds with dims ("time",) or ("time", "individuals"). VAME applies no preprocessing to these — the user is responsible for alignment, scaling, and NaN handling. See vame.validate_extra_features.

Returns

  • Tuple[np.ndarray, Dict[str, Any]]: A tuple containing:
  • The formatted array in the shape (n_features, n_samples)
  • A dictionary with feature provenance and processing information Where n_features = 2 * n_keypoints * n_spaces - 3 + len(extra_features or []).