AutoML Implementation

[muellerdo]

• Implementation of Pipeline Block: Training
• Implementation of Pipeline Block: Inference
• Implementation of Pipeline Block: Evaluation
• Implementation of Pipeline Block: XAI
• Try it out on a dataset
• Implementation of CLI base
• Implementation of Pipeline Block Integration in CLI
• Unittesting for everyone
• Dockerization
• Packaging as binary for PyPI
• Make an example
• Documentation

[muellerdo]

Idea for default automl setup:

• 5-fold cross-validation on dataset
• each fold with different architecture
• some folds with different input size?
• augmenting on / off?

[muellerdo]

Different pre-implemented settings:

• Minimalistic (mobilenetv2 with train/val split)
• Standard (just a single architecture with train/val percentage split and augmenting)
• Stacking
• Bagging
• Advanced (see above / composite)

[muellerdo]

Add xai_decoder parameter to config of pred module

• only supported for single model (minimal/standard)

[muellerdo]

Idea prediction mode:

rename path_imagedir to path_images:

• if path is a directory -> normal application
• if path contains a "," -> list of files -> split by ","
• if path is a file -> no input_interface and only passing to DataGenerator as [file] and image_format == None