morphoclass.xai.grad_cam_on_models module

Explain model layers using GradCam.

morphoclass.xai.grad_cam_on_models.get_edges_colors_based_on_barcode_colors(tree, colors)

Collect colors for edges based on barcode colors.

Parameters

• tree (tmd.Tree.Tree) – Morphology tree used to create barcode.

• colors (list_like) – List of barcode colors.

Returns

color_edges – List of edge colors.

Return type

list_like

morphoclass.xai.grad_cam_on_models.grad_cam_cnn_model(model, dataset, sample_id)

Explain CNN model.

Plot with feature maps after each feature extractor layer. Starting from original image to the last feature extractor layer. Only one morphology sample is visualized.

Parameters

• model (morphoclass.models.cnnet.CNNet) – Model that will be explained.

• dataset (morphoclass.data.morphology_dataset.MorphologyDataset) – Dataset containing embeddings and morphologies.

• sample_id (int) – The id of embedding in the dataset.

Returns

fig – Figure with explainable plots.

Return type

matplotlib.figure.Figure

morphoclass.xai.grad_cam_on_models.grad_cam_gnn_model(model, dataset, sample_id)

Explain GNN model.

Plot with two rows:

• Original graph and graph with GradCam values within the nodes.

• Heatmap of the original graph (zero-values) and heatmap of the GradCam values on the graph.

Only one morphology sample is visualized.

Parameters

• model (morphoclass.models.man_net.ManNet) – Model that will be explained.

• dataset (morphoclass.data.morphology_dataset.MorphologyDataset) – Dataset containing morphologies.

• sample_id (int) – The id of morphology in the dataset.

Returns

fig – Figure with explainable plots.

Return type

matplotlib.figure.Figure

morphoclass.xai.grad_cam_on_models.grad_cam_perslay_model(model, dataset, sample_id)

Explain PersLay model.

Plot with 3 rows:

• Barcodes: The original barcode and GradCam weighted barcode (colored bar) after each feature extraction layer.

• Persistence diagrams: The original PD and GradCam weighted PD (colored dot) after each feature extraction layer.

• Graph: The original graph and GradCam weighted graph (colored edge) after each feature extraction layer.

Parameters

• model (morphoclass.models.coriander_net.CorianderNet) – Model that will be explained.

• dataset (morphoclass.data.morphology_dataset.MorphologyDataset) – Dataset containing embeddings and morphologies.

• sample_id (int) – The id of embedding in the dataset.

Returns

fig – Figure with explainable plots.

Return type

matplotlib.figure.Figure