Skip to main content

model.create_training

logger_config​

logger​

traindata_aligned​

def traindata_aligned(config: dict,
                      sessions: List[str] | None = None,
                      test_fraction: float = 0.1,
                      read_from_variable: str = "position_processed",
                      split_mode: Literal["mode_1", "mode_2"] = "mode_2",
                      keypoints_to_include: List[str] | None = None,
                      keypoints_to_exclude: List[str] | None = None) -> None

Create training dataset for aligned data. Save numpy arrays with the test/train info to the project folder.

Parameters

  • config (dict): Configuration parameters dictionary.
  • sessions (List[str], optional): List of session names. If None, all sessions will be used. Defaults to None.
  • test_fraction (float, optional): Fraction of data to use as test data. Defaults to 0.1.
  • read_from_variable (str, optional): Variable name to read from the processed data. Defaults to "position_processed".
  • split_mode (Literal["mode_1", "mode_2"], optional): Mode for splitting data into train/test sets:
  • mode_1: Original mode that takes the initial test_fraction portion of the combined data for testing and the rest for training.
  • mode_2: Takes random continuous chunks from each session proportional to test_fraction for testing and uses the remaining parts for training. Defaults to "mode_2".

Returns

  • None

create_trainset​

@save_state(model=CreateTrainsetFunctionSchema)
def create_trainset(config: dict,
                    test_fraction: float = 0.1,
                    read_from_variable: str = "position_processed",
                    split_mode: Literal["mode_1", "mode_2"] = "mode_2",
                    keypoints_to_include: List[str] | None = None,
                    keypoints_to_exclude: List[str] | None = None,
                    save_logs: bool = True) -> None

Creates training and test datasets for the VAME model. Fills in the values in the "create_trainset" key of the states.json file. Creates the training dataset for VAME at:

  • project_name/
    • data/
      • train/
        • test_seq.npy
        • train_seq.npy
        • metadata.json

The produced test_seq.npy contains the combined data in the shape of (num_features, num_video_frames * test_fraction). The produced train_seq.npy contains the combined data in the shape of (num_features, num_video_frames * (1 - test_fraction)). The metadata.json file contains feature provenance information for tracking which keypoints and coordinates correspond to each feature in the numpy arrays, along with detailed split information for full reproducibility.

Parameters

  • config (dict): Configuration parameters dictionary.
  • test_fraction (float, optional): Fraction of data to use as test data. Defaults to 0.1.
  • read_from_variable (str, optional): Variable name to read from the processed data. Defaults to "position_processed".
  • split_mode (Literal["mode_1", "mode_2"], optional): Mode for splitting data into train/test sets:
  • mode_1: Original mode that takes the initial test_fraction portion of the combined data for testing and the rest for training.
  • mode_2: Takes random continuous chunks from each session proportional to test_fraction for testing and uses the remaining parts for training. Defaults to "mode_2".
  • save_logs (bool, optional): Whether to save logs. Defaults to True.

Returns

  • None