def eval_dis()

in src/evaluation/evaluator.py [0:0]

• 26 lines of code
• 5 McCabe index (conditional complexity)

    def eval_dis(self, to_log):
        """
        Evaluate discriminator predictions and accuracy.
        """
        bs = 128
        src_preds = []
        tgt_preds = []

        self.discriminator.eval()

        for i in range(0, self.src_emb.num_embeddings, bs):
            emb = Variable(self.src_emb.weight[i:i + bs].data, volatile=True)
            preds = self.discriminator(self.mapping(emb))
            src_preds.extend(preds.data.cpu().tolist())

        for i in range(0, self.tgt_emb.num_embeddings, bs):
            emb = Variable(self.tgt_emb.weight[i:i + bs].data, volatile=True)
            preds = self.discriminator(emb)
            tgt_preds.extend(preds.data.cpu().tolist())

        src_pred = np.mean(src_preds)
        tgt_pred = np.mean(tgt_preds)
        logger.info("Discriminator source / target predictions: %.5f / %.5f"
                    % (src_pred, tgt_pred))

        src_accu = np.mean([x >= 0.5 for x in src_preds])
        tgt_accu = np.mean([x < 0.5 for x in tgt_preds])
        dis_accu = ((src_accu * self.src_emb.num_embeddings + tgt_accu * self.tgt_emb.num_embeddings) /
                    (self.src_emb.num_embeddings + self.tgt_emb.num_embeddings))
        logger.info("Discriminator source / target / global accuracy: %.5f / %.5f / %.5f"
                    % (src_accu, tgt_accu, dis_accu))

        to_log['dis_accu'] = dis_accu
        to_log['dis_src_pred'] = src_pred
        to_log['dis_tgt_pred'] = tgt_pred