Quick-Start Script

This script provides a simple setup to use fusilli on a single dataset with default parameters.

Note

For a more detailed guide on using Fusilli, refer to the Training and Testing.

Before you run this, you need to:

1. Install fusilli (see How to Install).

2. Prepare your data and specify the paths to your data (see Loading your Data).

3. Specify output file paths (see Folder Configuration).

Running One Model (Regression)

from fusilli.data import prepare_fusion_data
from fusilli.train import train_and_save_models
from fusilli.eval import RealsVsPreds
import matplotlib.pyplot as plt

# Import the example fusion model
from fusilli.fusionmodels.tabularfusion.concat_data import ConcatTabularData

data_paths = {
    "tabular1": "path/to/tabular_1.csv",  # Path to tabular dataset 1
    "tabular2": "path/to/tabular_2.csv",  # Path to tabular dataset 2
    "image": "path/to/image_file.pt",  # Path to image dataset
}

output_paths = {
    "checkpoints": "path/to/checkpoints/dir",  # Unique dir for each experiment
    "losses": "path/to/losses/dir",  # Unique dir for each experiment
    "figures": "path/to/figures/dir",  # Unique dir for each experiment
}

# Get the data module (PyTorch Lightning-compatible data structure)
data_module = prepare_fusion_data(prediction_task="regression",
                                  fusion_model=ConcatTabularData,
                                  data_paths=data_paths,
                                  output_paths=output_paths)

# Train the model and receive a list with the trained model
trained_model = train_and_save_models(data_module=data_module,
                                      fusion_model=ConcatTabularData)

# Evaluate the model by plotting the real values vs. predicted values
RealsVsPreds_figure = RealsVsPreds.from_final_val_data(trained_model)
plt.show()

Comparing Multiple Models (Binary classification)

Here we show how to compare multiple models on a single dataset. Instead of using plt.show(), we can also save the figures to a file using plt.savefig(), so let’s do that.

from fusilli.data import prepare_fusion_data
from fusilli.train import train_and_save_models
from fusilli.eval import ConfusionMatrix, ModelComparison
from fusilli.utils.model_chooser import import_chosen_fusion_models
import matplotlib.pyplot as plt

data_paths = {
    "tabular1": "path/to/tabular_1.csv",  # Path to tabular dataset 1
    "tabular2": "path/to/tabular_2.csv",  # Path to tabular dataset 2
    "image": "path/to/image_file.pt",  # Path to image dataset
}

output_paths = {
    "checkpoints": "path/to/checkpoints/dir",  # Unique dir for each experiment
    "losses": "path/to/losses/dir",  # Unique dir for each experiment
    "figures": "path/to/figures/dir",  # Unique dir for each experiment
}

# import a list of models based on conditions: for example, all tabular models
model_conditions = {
    "modality_type": ["tabular1", "tabular2", "tabular_tabular"],
}

fusion_models = import_chosen_fusion_models(model_conditions)

# Dictionary to store trained models for later comparison
trained_models = {}

for model in fusion_models:

    model_name = model.__name__

    # Get the data module (PyTorch Lightning-compatible data structure)
    data_module = prepare_fusion_data(prediction_task="binary",
                                      fusion_model=model,
                                      data_paths=data_paths,
                                      output_paths=output_paths,
                                      )

        # Train the model and receive a list with the trained model
    trained_model = train_and_save_models(data_module=data_module,
                                          fusion_model=model,)

    # Evaluate the model by plotting the confusion matrix
    ConfusionMatrix_figure = ConfusionMatrix.from_final_val_data(trained_model)
    plt.savefig(f"{output_paths['figures']}/{model_name}_confusion_matrix.png")

    # save trained model
    trained_models[model_name] = trained_model

# compare models
comparison_fig = ModelComparison.from_final_val_data(trained_models)
plt.savefig(f"{output_paths['figures']}/model_comparison.png")