def perturb_o_and_get_filtered_rank()

in source/sagemaker/sagemaker_graph_entity_resolution/dgl_entity_resolution/utils.py [0:0]

• 20 lines of code
• 3 McCabe index (conditional complexity)

def perturb_o_and_get_filtered_rank(embedding, w, s, r, o, test_size, triplets_to_filter):
    """ Perturb object in the triplets
    """
    num_entities = embedding.shape[0]
    ranks = []
    for idx in range(test_size):
        if idx % 100 == 0:
            print("test triplet {} / {}".format(idx, test_size))
        target_s = s[idx]
        target_r = r[idx]
        target_o = o[idx]
        filtered_o = filter_o(triplets_to_filter, target_s, target_r, target_o, num_entities)
        target_o_idx = int((filtered_o == target_o).nonzero())
        emb_s = embedding[target_s]
        emb_r = w[target_r]
        emb_o = embedding[filtered_o]
        emb_triplet = emb_s * emb_r * emb_o
        scores = torch.sigmoid(torch.sum(emb_triplet, dim=1))
        _, indices = torch.sort(scores, descending=True)
        rank = int((indices == target_o_idx).nonzero())
        ranks.append(rank)
    return torch.LongTensor(ranks)