def calculate_metrics()

in source/sagemaker/src/package/model/inference.py [0:0]

• 21 lines of code
• 2 McCabe index (conditional complexity)

def calculate_metrics(model, iterator, criterion):
    epoch_loss = 0
    epoch_acc = 0

    model.eval()

    all_predictions = np.array([])
    all_labels = np.array([])
    with torch.no_grad():
        for batch in iterator:
            predictions = model(batch.review).squeeze(1)

            loss = criterion(predictions, batch.sentiment)

            acc = binary_accuracy(predictions, batch.sentiment)

            epoch_loss += loss.item()
            epoch_acc += acc.item()

            all_predictions = np.concatenate(
                (all_predictions, torch.sigmoid(predictions).detach().cpu().numpy()), axis=0)
            all_labels = np.concatenate(
                (all_labels, batch.sentiment.detach().cpu().clone().numpy()), axis=0)

    fpr, tpr, thresholds = roc_curve(all_labels, all_predictions, pos_label=1)
    prec = precision_score(all_labels, np.round(all_predictions, 0))
    rec = recall_score(all_labels, np.round(all_predictions, 0))

    return epoch_loss / iterator.iterations, epoch_acc / iterator.iterations, fpr, tpr, prec, rec