Training MOGDx

train.AUROC(logits, targets, meta)[source]

Calculates the Area Under the Receiver Operating Characteristic curve (AUROC).

Parameters:

  • logits (torch.Tensor) – The predicted outcomes from the model.

  • targets (torch.Tensor) – The true outcomes.

  • meta (pd.Series) – Metadata associated with the outcomes.

Returns:

Contains the matplotlib figure object, and the transformed prediction scores.

Return type:

tuple

train.confusion_matrix(logits, targets, display_labels)[source]

Generates a confusion matrix plot from the predicted and true labels.

Parameters:

  • logits (torch.Tensor) – Logits output from the model.

  • targets (torch.Tensor) – True labels for the data.

  • display_labels (list) – List of labels to display on the matrix axes.

Returns:

Seaborn axis grid containing the confusion matrix plot.

Return type:

seaborn.axisgrid

train.evaluate(model, graph, dataloader)[source]

Evaluates the model performance on a validation set.

Parameters:

  • model (torch.nn.Module) – Model to evaluate.

  • graph (dgl.DGLGraph) – Graph from which data is to be loaded.

  • dataloader (torch.utils.data.DataLoader) – DataLoader the supplies the evaluation data.

Returns:

A tuple containing the loss, accuracy, F1 score, precision, recall, logits and labels.

Return type:

tuple

train.layerwise_infer(device, graph, nid, model, batch_size)[source]

Perform inference in a layer-wise manner for subgraph provided.

Parameters:

  • device (str) – The device on which the inference will be computed (e.g., ‘cuda’ or ‘cpu’).

  • graph (dgl.DGLGraph) – The graph containing the nodes.

  • nid (list or tensor) – Node IDs for which inference is to be performed.

  • model (nn.Module) – The trained model.

  • batch_size (int) – The size of batches to process the graph.

Returns:

Accuracy of prediction after inference.

Return type:

float

train.train(g, train_index, device, model, labels, epochs, lr, patience, pretrain=False, pnet=False, batch_size=1024)[source]

Trains a model on the given graph data using specified parameters.

Parameters:

  • g (dgl.DGLGraph) – The graph containing feature and label data.

  • train_index (list) – Indices used for training the model.

  • device (str) – Device type to use (‘cuda’ or ‘cpu’).

  • model (torch.nn.Module) – Model to be trained.

  • labels (torch.Tensor) – Labels for the training data.

  • epochs (int) – Number of epochs to train the model.

  • lr (float) – Learning rate for the optimizer.

  • patience (int) – Patience for early stopping criterion.

  • pretrain (bool) – Whether to preprocess data before training.

  • pnet (bool) – Indicates the use of a Pathway Network model for data preprocessing.

Returns:

Either a graph or a matplotlib figure depending on the ‘pretrain’ parameter.

train.tsne_embedding_plot(emb, meta)[source]

Generates a 2D t-SNE plot of embeddings colored by metadata labels.

Parameters:

  • emb (np.array) – High-dimensional embeddings to be reduced.

  • meta (pd.Series) – Metadata series containing labels to color the embeddings.

Returns:

Displays a t-SNE plot of embeddings.

Return type:

None