Trait-Driven Model Imports

fusilli offers various models for data fusion available in fusilli.fusionmodels. Each model is adaptable to different degrees, and you can modify them as described in Modify the Fusion Models.

Choosing a Model

You have several ways to select a model:

1. Direct Import: Import a specific model from fusilli.fusionmodels at the beginning of your script.

2. Use Model Chooser Function: Utilize fusilli.utils.model_chooser.import_chosen_fusion_models() to filter models based on specific criteria.

Model Chooser Function Usage

The fusilli.utils.model_chooser.import_chosen_fusion_models() function takes a dictionary of criteria. The keys represent criteria names, and the values are the desired criteria.

For instance, to fetch models capable of fusing tabular and image data, use:

from fusilli.utils.model_chooser import import_chosen_fusion_models

criteria = {
    "modality_type": ["tab_img"],
}

models = import_chosen_fusion_models(criteria)

This function returns a list of models that fulfill the specified criteria. Access these models by indexing the list (e.g., models[0], models[1], etc.). It will also display a list of all models meeting that description.

Examples of Criteria

• Attention-Based Fusion for Tabular and Image Data:

criteria = {
    "modality_type": ["tab_img"],
    "fusion_type": ["attention"],
}

• Operation- or Attention-Based Fusion for Tabular and Image Data; also Uni-Modal Benchmark Models:

criteria = {
    "modality_type": ["tab_img", "tabular1", "img"],
    "fusion_type": ["operation", "attention"],
}

• Subspace-Based Fusion for Any Modality Type:

criteria = {
    "fusion_type": ["subspace"],
    "modality_type": "all",
}

• Specific Models by Name: Tabular1Unimodal, Tabular2Unimodal, and ConcatTabularData:

criteria = {
    "class_name": ["Tabular1Unimodal", "Tabular2Unimodal", "ConcatTabularData"],
}