Skip to content

Parameters

AutoML Mode: training¤

The training mode fits a single or multiple models with fixed or self-adjusting hyper parameters.

The training process takes as input images with annotated classification (ground truth) and outputs the fitted model(s).

Parameter overview for the training process.

Category Argument Type Default Description
I/O --path_imagedir str training Path to the directory containing the images.
I/O --path_modeldir str model Path to the output directory in which fitted models and metadata are stored.
I/O --path_gt str None Path to the index/class annotation file if required. (only for 'csv' interface).
I/O --ohe bool False Boolean option whether annotation data is sparse categorical or one-hot encoded.
Configuration --analysis str standard Analysis mode for the AutoML training. Options: ["minimal", "standard", "advanced"].
Configuration --three_dim bool False Boolean, whether data is 2D or 3D.
Configuration --shape_3D str 128x128x128 Desired input shape of 3D volume for architecture (will be cropped into, format: 1x2x3).
Configuration --epochs int 500 Number of epochs. A single epoch is defined as one iteration through the complete data set.
Configuration --batch_size int 24 Number of samples inside a single batch.
Configuration --workers int 1 Number of workers/threads which preprocess batches during runtime.
Configuration --metalearner str mean Key for Metalearner or Aggregate function.
Configuration --architecture str DenseNet121 Key of single or multiple Architectures (only supported for 'analysis=advanced', format: 'KEY' or 'KEY,KEY,KEY).
Other --help bool False show this help message and exit.
List of Architectures

AUCMEDI provides a large library of state-of-the-art and ready-to-use architectures.

List of Metalearner

AutoML Mode: prediction¤

The prediction mode utilizes the fitted model(s) to infer the classification of unknown images.

The prediction process takes as input unknown images and outputs a CSV file with prediction probabilities.

Parameter overview for the prediction process.

Category Argument Type Default Description
I/O --path_imagedir str test Path to the directory containing the images.
I/O --path_modeldir str model Path to the output directory in which fitted models and metadata are stored.
I/O --path_pred str preds.csv Path to the output file in which predicted csv file should be stored.
Configuration --xai_method str None Key for XAI method.
Configuration --xai_directory str xai Path to the output directory in which predicted image xai heatmaps should be stored.
Configuration --batch_size int 24 Number of samples inside a single batch.
Configuration --workers int 1 Number of workers/threads which preprocess batches during runtime.
Other --help bool False show this help message and exit.
List of XAI Methods

AUCMEDI provides a large library of state-of-the-art and ready-to-use XAI methods: aucmedi.xai.methods

AutoML Mode: evaluation¤

The evaluation mode compares ground truth annotations with predicted classifications to estimate model performance.

The evaluation process takes as input images with annotated classification (ground truth) as well as predicted classifications, and outputs various performance evaluation figures and metrics.

Parameter overview for the evaluation process.

Category Argument Type Default Description
I/O --path_imagedir str training Path to the directory containing the ground truth images.
I/O --path_gt str None Path to the index/class annotation CSV file (only required for defining the ground truth via 'csv' instead of 'directory' interface).
I/O --ohe bool False Boolean option whether annotation data is sparse categorical or one-hot encoded.
I/O --path_pred str preds.csv Path to the input file in which predicted csv file is stored.
I/O --path_evaldir str evaluation Path to the directory in which evaluation figures and tables should be stored.
Other --help bool False show this help message and exit.