def validate()

in solver/self_play.py [0:0]

• 48 lines of code
• 14 McCabe index (conditional complexity)

    def validate(self, specified_set):
        self.model.eval()
        total_hit = 0
        total_cnt = 0
        out_file = open('{}.txt'.format(self.exp_name), 'w')
        if not self.use_gt_question:
            out_file.write('game_id|pred_obj|answer_obj|turn_id|question|answer\n')
        
        for val_step, data in enumerate(specified_set):
            game, img_feat, tgt_cat, tgt_bbox, cats, bboxs, bboxs_mask, label, qs, q_len = self.fetch_data(data)
            with torch.no_grad():
                if self.use_gt_question:
                    pred, dialog = self.model.play_with_gt_question(
                        qs, q_len, tgt_cat, tgt_bbox, cats, bboxs, bboxs_mask, 
                        self.tokenizer.pad_id, self.answer2id, self.answer2token)
                    
                else:
                    pred, dialog, q_log, a_log = self.model.play(
                        img_feat, tgt_cat, tgt_bbox, cats, bboxs, bboxs_mask,
                        self.tokenizer.sos_id, self.tokenizer.pad_id, 
                        self.tokenizer.eoq_id, self.tokenizer.eod_id,
                        self.answer2id, self.answer2token,
                        max_q_len=20, greedy=True, max_turns=5
                    )
                if self.use_gt_question:
                    for b in range(pred.size(0)):
                        out_file.write("{}-{}/{} | {}\n".format(
                            game[b].id, pred[b].argmax(dim=-1).item(), label[b].item(), self.tokenizer.decode(dialog[b].tolist())))
                else:
                    for b in range(pred.size(0)):
                        out_prefix = "{}|{}|{}".format(game[b].id, pred[b].argmax(dim=-1).item(), label[b].item())
                        for t in range(len(q_log[b])):
                            out_str = out_prefix + "|{}|{}|".format(t, self.tokenizer.decode(q_log[b][t].tolist()))
                            if t != len(q_log[b])-1:
                                out_str += "{}".format(self.tokenizer.decode(a_log[b][t].tolist()))
                            out_file.write(out_str+'\n')
            
                total_hit += (pred.argmax(dim=-1) == label).sum().item()
                total_cnt += pred.size(0)
                # if (val_step == 0) or ((val_step+1) % self._progress_step == 0):
                self.progress("Dev stat. |  Acc. - {:.3f}".format(total_hit/float(total_cnt)))
                # Log
                if self.mode == 'train':
                    NOT_IMPLEMENT_YET()

                        
        if self.mode == 'train':

            score = -avg_loss
            if score > self.best_score:
                #self.save_checkpoint('step_{}.pth'.format(self.step), score)
                self.save_checkpoint('best.pth', score)
                self.best_score = score
            self.model.train()

        self.verbose(["Val stat. @ step {} | Acc. - {:.3f}"
                      .format(self.step, total_hit / float(total_cnt))])
        
        out_file.close()