evaluation.py

import torch.nn.functional as F
import torch
from sklearn.metrics import f1_score, roc_auc_score
from utils import fair_metric

def evaluate(x, classifier, hp, encoder, data, args):
    classifier.eval()
    encoder.eval()

    with torch.no_grad():
        h = encoder(data.x, data.edge_index, data.adj_norm_sp)
        output = classifier(h)

    accs, auc_rocs, F1s, paritys, equalitys = {}, {}, {}, {}, {}

    pred_val = (output[data.val_mask].squeeze() > 0).type_as(data.y)
    pred_test = (output[data.test_mask].squeeze() > 0).type_as(data.y)

    accs['val'] = pred_val.eq(
        data.y[data.val_mask]).sum().item() / data.val_mask.sum().item()
    accs['test'] = pred_test.eq(
        data.y[data.test_mask]).sum().item() / data.test_mask.sum().item()

    F1s['val'] = f1_score(data.y[data.val_mask].cpu(
    ).numpy(), pred_val.cpu().numpy())

    F1s['test'] = f1_score(data.y[data.test_mask].cpu(
    ).numpy(), pred_test.cpu().numpy())

    auc_rocs['val'] = roc_auc_score(
        data.y[data.val_mask].cpu().numpy(), output[data.val_mask].detach().cpu().numpy())
    auc_rocs['test'] = roc_auc_score(
        data.y[data.test_mask].cpu().numpy(), output[data.test_mask].detach().cpu().numpy())

    paritys['val'], equalitys['val'] = fair_metric(pred_val.cpu().numpy(), data.y[data.val_mask].cpu(
    ).numpy(), data.sens[data.val_mask].cpu().numpy())

    paritys['test'], equalitys['test'] = fair_metric(pred_test.cpu().numpy(), data.y[data.test_mask].cpu(
    ).numpy(), data.sens[data.test_mask].cpu().numpy())

    return accs, auc_rocs, F1s, paritys, equalitys