def validate()

in solver/guesser_vilbert.py [0:0]

• 41 lines of code
• 12 McCabe index (conditional complexity)

    def validate(self, specified_set, write_log=False):
        self.model.eval()
        total_loss = 0
        total_hit = 0
        cnt = 0
        if write_log:
            out_file = open('{}_stat_his.txt'.format(self.exp_name), 'w')
        for val_step, data in enumerate(specified_set):
            game, qs, qs_len, answers, end_turn, cats, img_feats, bboxs, bboxs_mask, bboxs_mask_vb, txt_attn_mask, label = self.fetch_data(data)
            with torch.no_grad():
                pred = self.model.forward_session(
                    qs, answers, end_turn, cats, img_feats, bboxs, 
                    bboxs_mask=bboxs_mask, 
                    attention_mask=txt_attn_mask,
                    image_attention_mask=bboxs_mask_vb,
                    return_state_history=self.mode=='test'
                    )
                if self.mode == 'test':
                    pred, stat_his = pred

                loss = self.loss(pred, label).item()
                hit = (pred.argmax(dim=-1) == label).sum().item()
                total_hit += hit
                total_loss += loss
                cnt += len(label)
                if (val_step == 0) or ((val_step+1) % self._progress_step == 0):
                    self.progress("Dev stat. ({}/{}) | Loss - {:.4f} | Acc. - {:.4f}".format(
                        val_step, len(specified_set), total_loss/float(cnt), total_hit/float(cnt)))

                if self.mode == 'test':
                    if write_log:
                        for g, l, p, stat_h in zip(game, label, pred.argmax(dim=-1), stat_his):
                            out_file.write("{}|{}|{}|{}\n".format(g.id, l, p,stat_h.tolist()))


        score = total_hit / float(cnt)
        loss = total_loss / float(len(specified_set))
        if score > self.best_score and self.mode == 'train':
            self.save_checkpoint('best.pth', score)
            self.best_score = score

        self.verbose(["Val stat. @ step {} | Loss - {:.4f} | Acc. - {:.4f}"
                      .format(self.step, loss, score)])

        self.model.train()
        if write_log:
            out_file.close()