Merge pull request '데이터셋 entity 수정, 데이터셋 저장 수정' (#183) from feat/training_260324 into develop

Reviewed-on: #183

src/main/java/com/kamco/cd/training/common/dto/HyperParam.java

@@ -14,7 +14,7 @@ import lombok.Setter;
 public class HyperParam {
 
   @Schema(description = "모델", example = "G1")
-  private ModelType model; // G1, G2, G3
+  private ModelType model; // G1, G2, G3, G4
 
   // -------------------------
   // Important

src/main/java/com/kamco/cd/training/common/enums/ModelType.java

@@ -12,7 +12,8 @@ import lombok.Getter;
 public enum ModelType implements EnumType {
   G1("G1"),
   G2("G2"),
-  G3("G3");
+  G3("G3"),
+  G4("G4");
 
   private String desc;
 

src/main/java/com/kamco/cd/training/dataset/dto/DatasetDto.java

@@ -254,6 +254,7 @@ public class DatasetDto {
 
     private Long wasteCnt;
     private Long landCoverCnt;
+    private Integer solarPanelCnt;
 
     public SelectDataSet(
         String modelNo,
@@ -308,6 +309,29 @@ public class DatasetDto {
       this.containerCnt = containerCnt;
     }
 
+    public SelectDataSet(
+        String modelNo,
+        Long datasetId,
+        UUID uuid,
+        String dataType,
+        String title,
+        Long roundNo,
+        Integer compareYyyy,
+        Integer targetYyyy,
+        String memo,
+        Integer solarPanelCnt) {
+      this.datasetId = datasetId;
+      this.uuid = uuid;
+      this.dataType = dataType;
+      this.dataTypeName = getDataTypeName(dataType);
+      this.title = title;
+      this.roundNo = roundNo;
+      this.compareYyyy = compareYyyy;
+      this.targetYyyy = targetYyyy;
+      this.memo = memo;
+      this.solarPanelCnt = solarPanelCnt;
+    }
+
     public String getDataTypeName(String groupTitleCd) {
       LearnDataType type = Enums.fromId(LearnDataType.class, groupTitleCd);
       return type == null ? null : type.getText();

src/main/java/com/kamco/cd/training/hyperparam/HyperParamApiController.java

@@ -101,7 +101,7 @@ public class HyperParamApiController {
           LocalDate endDate,
       @Parameter(description = "버전명", example = "G1_000019") @RequestParam(required = false)
           String hyperVer,
-      @Parameter(description = "모델 타입 (G1, G2, G3 중 하나)", example = "G1")
+      @Parameter(description = "모델 타입 (G1, G2, G3, G4 중 하나)", example = "G1")
           @RequestParam(required = false)
           ModelType model,
       @Parameter(

src/main/java/com/kamco/cd/training/model/ModelTrainMngApiController.java

@@ -1,6 +1,7 @@
 package com.kamco.cd.training.model;
 
 import com.kamco.cd.training.common.dto.MonitorDto;
+import com.kamco.cd.training.common.enums.ModelType;
 import com.kamco.cd.training.common.service.SystemMonitorService;
 import com.kamco.cd.training.config.api.ApiResponseDto;
 import com.kamco.cd.training.dataset.dto.DatasetDto;
@@ -68,7 +69,7 @@ public class ModelTrainMngApiController {
       @Parameter(
               description = "모델",
               example = "G1",
-              schema = @Schema(allowableValues = {"G1", "G2", "G3"}))
+              schema = @Schema(allowableValues = {"G1", "G2", "G3", "G4"}))
           @RequestParam(required = false)
           String modelNo,
       @Parameter(description = "페이지 번호") @RequestParam(defaultValue = "0") int page,
@@ -143,9 +144,9 @@ public class ModelTrainMngApiController {
       @Parameter(
               description = "모델 구분",
               example = "",
-              schema = @Schema(allowableValues = {"G1", "G2", "G3"}))
+              schema = @Schema(allowableValues = {"G1", "G2", "G3", "G4"}))
           @RequestParam
-          String modelType,
+          ModelType modelType,
       @Parameter(
               description = "선택 구분",
               example = "",
@@ -153,7 +154,7 @@ public class ModelTrainMngApiController {
           @RequestParam
           String selectType) {
     DatasetReq req = new DatasetReq();
-    req.setModelNo(modelType);
+    req.setModelNo(modelType.getId());
     req.setDataType(selectType);
     return ApiResponseDto.ok(modelTrainMngService.getDatasetSelectList(req));
   }

src/main/java/com/kamco/cd/training/model/dto/ModelTrainMngDto.java

@@ -139,7 +139,7 @@ public class ModelTrainMngDto {
   public static class AddReq {
 
     @NotNull
-    @Schema(description = "모델 종류 G1, G2, G3", example = "G1")
+    @Schema(description = "모델 종류 G1, G2, G3, G4", example = "G1")
     private String modelNo;
 
     @NotNull
@@ -197,10 +197,11 @@ public class ModelTrainMngDto {
     @Schema(description = "폐기물", example = "0")
     private Long wasteCnt;
 
-    @Schema(
+    @Schema(description = "도로, 비닐하우스, 밭, 과수원, 초지, 숲, 물, 모재/자갈, 토분(무덤), 일반토지, 기타", example = "0")
-        description = "도로, 비닐하우스, 밭, 과수원, 초지, 숲, 물, 모재/자갈, 토분(무덤), 일반토지, 태양광, 기타",
-        example = "0")
     private Long LandCoverCnt;
+
+    @Schema(description = "태양광", example = "0")
+    private Long solar;
   }
 
   @Getter

src/main/java/com/kamco/cd/training/model/service/ModelTrainMngService.java

@@ -51,10 +51,10 @@ public class ModelTrainMngService {
 
   @Value("${train.docker.response_dir}")
   private String responseDir;
+
   @Value("${train.docker.symbolic_link_dir}")
   private String symbolicDir;
 
-
   /**
    * 모델학습 조회
    *
@@ -324,6 +324,8 @@ public class ModelTrainMngService {
   public List<SelectDataSet> getDatasetSelectList(DatasetReq req) {
     if (req.getModelNo().equals(ModelType.G1.getId())) {
       return modelTrainMngCoreService.getDatasetSelectG1List(req);
+    } else if (req.getModelNo().equals(ModelType.G4.getId())) {
+      return modelTrainMngCoreService.getDatasetSelectG4List(req);
     } else {
       return modelTrainMngCoreService.getDatasetSelectG2G3List(req);
     }

src/main/java/com/kamco/cd/training/postgres/core/ModelTrainMngCoreService.java

@@ -154,6 +154,8 @@ public class ModelTrainMngCoreService {
       datasetEntity.setWasteCnt(dataset.getSummary().getWasteCnt());
     } else if (addReq.getModelNo().equals(ModelType.G3.getId())) {
       datasetEntity.setLandCoverCnt(dataset.getSummary().getLandCoverCnt());
+    } else if (addReq.getModelNo().equals(ModelType.G4.getId())) {
+      datasetEntity.setSolarCnt(dataset.getSummary().getSolar());
     }
 
     datasetEntity.setCreatedUid(userUtil.getId());
@@ -337,6 +339,16 @@ public class ModelTrainMngCoreService {
     return datasetRepository.getDatasetTransferSelectG2G3List(modelId, modelNo);
   }
 
+  /**
+   * 데이터셋 G4 목록
+   *
+   * @param req
+   * @return
+   */
+  public List<SelectDataSet> getDatasetSelectG4List(DatasetReq req) {
+    return datasetRepository.getDatasetSelectG4List(req);
+  }
+
   // TODO 미사용 끝
 
   /**

src/main/java/com/kamco/cd/training/postgres/entity/ModelDatasetEntity.java

@@ -43,6 +43,9 @@ public class ModelDatasetEntity {
   @Column(name = "land_cover_cnt")
   private Long landCoverCnt;
 
+  @Column(name = "solar_cnt")
+  private Long solarCnt;
+
   @ColumnDefault("now()")
   @Column(name = "created_dttm")
   private ZonedDateTime createdDttm = ZonedDateTime.now();

src/main/java/com/kamco/cd/training/postgres/repository/dataset/DatasetRepositoryCustom.java

@@ -27,6 +27,8 @@ public interface DatasetRepositoryCustom {
 
   List<SelectDataSet> getDatasetSelectG2G3List(DatasetReq req);
 
+  List<SelectDataSet> getDatasetSelectG4List(DatasetReq req);
+
   Long getDatasetMaxStage(int compareYyyy, int targetYyyy);
 
   Long insertDatasetMngData(DatasetMngRegDto mngRegDto);

src/main/java/com/kamco/cd/training/postgres/repository/dataset/DatasetRepositoryImpl.java

@@ -4,6 +4,7 @@ import static com.kamco.cd.training.postgres.entity.QDatasetObjEntity.datasetObj
 import static com.kamco.cd.training.postgres.entity.QModelDatasetMappEntity.modelDatasetMappEntity;
 import static com.kamco.cd.training.postgres.entity.QModelMasterEntity.modelMasterEntity;
 
+import com.kamco.cd.training.common.enums.DetectionClassification;
 import com.kamco.cd.training.common.enums.ModelType;
 import com.kamco.cd.training.dataset.dto.DatasetDto.DatasetMngRegDto;
 import com.kamco.cd.training.dataset.dto.DatasetDto.DatasetReq;
@@ -104,10 +105,6 @@ public class DatasetRepositoryImpl implements DatasetRepositoryCustom {
       builder.and(dataset.dataType.eq(req.getDataType()));
     }
 
-    if (StringUtils.isNotBlank(req.getDataType()) && !"CURRENT".equals(req.getDataType())) {
-      builder.and(dataset.dataType.eq(req.getDataType()));
-    }
-
     if (req.getIds() != null) {
       builder.and(dataset.id.in(req.getIds()));
     }
@@ -126,12 +123,15 @@ public class DatasetRepositoryImpl implements DatasetRepositoryCustom {
                 dataset.targetYyyy,
                 dataset.memo,
                 new CaseBuilder()
-                    .when(datasetObjEntity.targetClassCd.eq("building"))
+                    .when(
+                        datasetObjEntity.targetClassCd.eq(DetectionClassification.BUILDING.getId()))
                     .then(1)
                     .otherwise(0)
                     .sum(),
                 new CaseBuilder()
-                    .when(datasetObjEntity.targetClassCd.eq("container"))
+                    .when(
+                        datasetObjEntity.targetClassCd.eq(
+                            DetectionClassification.CONTAINER.getId()))
                     .then(1)
                     .otherwise(0)
                     .sum()))
@@ -249,19 +249,31 @@ public class DatasetRepositoryImpl implements DatasetRepositoryCustom {
   }
 
   // TODO 미사용 끝
+
   @Override
   public List<SelectDataSet> getDatasetSelectG2G3List(DatasetReq req) {
 
+    String building = DetectionClassification.BUILDING.getId();
+    String container = DetectionClassification.CONTAINER.getId();
+    String waste = DetectionClassification.WASTE.getId();
+    String solar = DetectionClassification.SOLAR.getId();
+
     BooleanBuilder builder = new BooleanBuilder();
     builder.and(dataset.deleted.isFalse());
 
     NumberExpression<Long> selectedCnt = null;
     NumberExpression<Long> wasteCnt =
-        datasetObjEntity.targetClassCd.when("waste").then(1L).otherwise(0L).sum();
+        datasetObjEntity
+            .targetClassCd
+            .when(DetectionClassification.WASTE.getId())
+            .then(1L)
+            .otherwise(0L)
+            .sum();
 
+    // G1, G2, G4 제외
     NumberExpression<Long> elseCnt =
         new CaseBuilder()
-            .when(datasetObjEntity.targetClassCd.notIn("building", "container", "waste"))
+            .when(datasetObjEntity.targetClassCd.notIn(building, container, waste, solar))
             .then(1L)
             .otherwise(0L)
             .sum();
@@ -481,4 +493,51 @@ public class DatasetRepositoryImpl implements DatasetRepositoryCustom {
         .where(dataset.uid.eq(uid), dataset.deleted.isFalse())
         .fetchOne();
   }
+
+  @Override
+  public List<SelectDataSet> getDatasetSelectG4List(DatasetReq req) {
+    BooleanBuilder builder = new BooleanBuilder();
+
+    builder.and(dataset.deleted.isFalse());
+
+    if (StringUtils.isNotBlank(req.getDataType()) && !"CURRENT".equals(req.getDataType())) {
+      builder.and(dataset.dataType.eq(req.getDataType()));
+    }
+
+    if (req.getIds() != null) {
+      builder.and(dataset.id.in(req.getIds()));
+    }
+
+    return queryFactory
+        .select(
+            Projections.constructor(
+                SelectDataSet.class,
+                Expressions.constant(req.getModelNo()),
+                dataset.id,
+                dataset.uuid,
+                dataset.dataType,
+                dataset.title,
+                dataset.roundNo,
+                dataset.compareYyyy,
+                dataset.targetYyyy,
+                dataset.memo,
+                new CaseBuilder()
+                    .when(datasetObjEntity.targetClassCd.eq(DetectionClassification.SOLAR.getId()))
+                    .then(1)
+                    .otherwise(0)
+                    .sum()))
+        .from(dataset)
+        .leftJoin(datasetObjEntity)
+        .on(dataset.id.eq(datasetObjEntity.datasetUid))
+        .where(builder)
+        .groupBy(
+            dataset.id,
+            dataset.uuid,
+            dataset.dataType,
+            dataset.title,
+            dataset.roundNo,
+            dataset.memo)
+        .orderBy(dataset.createdDttm.desc())
+        .fetch();
+  }
 }

src/main/java/com/kamco/cd/training/postgres/repository/model/ModelDatasetMappRepositoryImpl.java

@@ -1,9 +1,11 @@
 package com.kamco.cd.training.postgres.repository.model;
 
 import static com.kamco.cd.training.postgres.entity.QDatasetEntity.datasetEntity;
+import static com.kamco.cd.training.postgres.entity.QDatasetObjEntity.datasetObjEntity;
 import static com.kamco.cd.training.postgres.entity.QModelDatasetMappEntity.modelDatasetMappEntity;
 import static com.kamco.cd.training.postgres.entity.QModelMasterEntity.modelMasterEntity;
 
+import com.kamco.cd.training.common.enums.DetectionClassification;
 import com.kamco.cd.training.common.enums.ModelType;
 import com.kamco.cd.training.postgres.entity.ModelDatasetMappEntity;
 import com.kamco.cd.training.postgres.entity.QDatasetObjEntity;
@@ -11,6 +13,7 @@ import com.kamco.cd.training.postgres.entity.QDatasetTestObjEntity;
 import com.kamco.cd.training.postgres.entity.QDatasetValObjEntity;
 import com.kamco.cd.training.train.dto.ModelTrainLinkDto;
 import com.querydsl.core.types.Projections;
+import com.querydsl.core.types.dsl.BooleanExpression;
 import com.querydsl.jpa.impl.JPAQueryFactory;
 import java.util.List;
 import lombok.RequiredArgsConstructor;
@@ -33,9 +36,44 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
 
   @Override
   public List<ModelTrainLinkDto> findDatasetTrainPath(Long modelId) {
-
     QDatasetObjEntity datasetObjEntity = QDatasetObjEntity.datasetObjEntity;
 
+    // =====================
+    // 조건 분리
+    // =====================
+    BooleanExpression g1 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G1.getId())
+            .and(
+                datasetObjEntity.targetClassCd.in(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId()));
+
+    BooleanExpression g2 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G2.getId())
+            .and(datasetObjEntity.targetClassCd.eq(DetectionClassification.WASTE.getId()));
+
+    BooleanExpression g4 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G4.getId())
+            .and(datasetObjEntity.targetClassCd.eq(DetectionClassification.SOLAR.getId()));
+
+    // G3 = 전체 허용 (fallback)
+    BooleanExpression g3 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G3.getId())
+            .and(
+                datasetObjEntity.targetClassCd.notIn(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId(),
+                    DetectionClassification.WASTE.getId(),
+                    DetectionClassification.SOLAR.getId()));
+
     return queryFactory
         .select(
             Projections.constructor(
@@ -60,17 +98,7 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
             datasetObjEntity
                 .datasetUid
                 .eq(modelDatasetMappEntity.datasetUid)
-                .and(
+                .and(g1.or(g2).or(g4).or(g3)))
-                    modelMasterEntity
-                        .modelNo
-                        .eq(ModelType.G1.getId())
-                        .and(datasetObjEntity.targetClassCd.upper().in("CONTAINER", "BUILDING"))
-                        .or(
-                            modelMasterEntity
-                                .modelNo
-                                .eq(ModelType.G2.getId())
-                                .and(datasetObjEntity.targetClassCd.upper().eq("WASTE")))
-                        .or(modelMasterEntity.modelNo.eq(ModelType.G3.getId()))))
         .where(modelMasterEntity.id.eq(modelId))
         .fetch();
   }
@@ -80,6 +108,42 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
 
     QDatasetValObjEntity datasetValObjEntity = QDatasetValObjEntity.datasetValObjEntity;
 
+    // =====================
+    // 조건 분리
+    // =====================
+    BooleanExpression g1 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G1.getId())
+            .and(
+                datasetValObjEntity.targetClassCd.in(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId()));
+
+    BooleanExpression g2 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G2.getId())
+            .and(datasetValObjEntity.targetClassCd.eq(DetectionClassification.WASTE.getId()));
+
+    BooleanExpression g4 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G4.getId())
+            .and(datasetValObjEntity.targetClassCd.eq(DetectionClassification.SOLAR.getId()));
+
+    // G3 = 전체 허용 (fallback)
+    BooleanExpression g3 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G3.getId())
+            .and(
+                datasetValObjEntity.targetClassCd.notIn(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId(),
+                    DetectionClassification.WASTE.getId(),
+                    DetectionClassification.SOLAR.getId()));
+
     return queryFactory
         .select(
             Projections.constructor(
@@ -104,17 +168,7 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
             datasetValObjEntity
                 .datasetUid
                 .eq(modelDatasetMappEntity.datasetUid)
-                .and(
+                .and(g1.or(g2).or(g4).or(g3)))
-                    modelMasterEntity
-                        .modelNo
-                        .eq(ModelType.G1.getId())
-                        .and(datasetValObjEntity.targetClassCd.upper().in("CONTAINER", "BUILDING"))
-                        .or(
-                            modelMasterEntity
-                                .modelNo
-                                .eq(ModelType.G2.getId())
-                                .and(datasetValObjEntity.targetClassCd.upper().eq("WASTE")))
-                        .or(modelMasterEntity.modelNo.eq(ModelType.G3.getId()))))
         .where(modelMasterEntity.id.eq(modelId))
         .fetch();
   }
@@ -124,6 +178,42 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
 
     QDatasetTestObjEntity datasetTestObjEntity = QDatasetTestObjEntity.datasetTestObjEntity;
 
+    // =====================
+    // 조건 분리
+    // =====================
+    BooleanExpression g1 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G1.getId())
+            .and(
+                datasetTestObjEntity.targetClassCd.in(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId()));
+
+    BooleanExpression g2 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G2.getId())
+            .and(datasetTestObjEntity.targetClassCd.eq(DetectionClassification.WASTE.getId()));
+
+    BooleanExpression g4 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G4.getId())
+            .and(datasetTestObjEntity.targetClassCd.eq(DetectionClassification.SOLAR.getId()));
+
+    // G3 = 전체 허용 (fallback)
+    BooleanExpression g3 =
+        modelMasterEntity
+            .modelNo
+            .eq(ModelType.G3.getId())
+            .and(
+                datasetTestObjEntity.targetClassCd.notIn(
+                    DetectionClassification.CONTAINER.getId(),
+                    DetectionClassification.BUILDING.getId(),
+                    DetectionClassification.WASTE.getId(),
+                    DetectionClassification.SOLAR.getId()));
+
     return queryFactory
         .select(
             Projections.constructor(
@@ -148,17 +238,7 @@ public class ModelDatasetMappRepositoryImpl implements ModelDatasetMappRepositor
             datasetTestObjEntity
                 .datasetUid
                 .eq(modelDatasetMappEntity.datasetUid)
-                .and(
+                .and(g1.or(g2).or(g4).or(g3)))
-                    modelMasterEntity
-                        .modelNo
-                        .eq(ModelType.G1.getId())
-                        .and(datasetTestObjEntity.targetClassCd.upper().in("CONTAINER", "BUILDING"))
-                        .or(
-                            modelMasterEntity
-                                .modelNo
-                                .eq(ModelType.G2.getId())
-                                .and(datasetTestObjEntity.targetClassCd.upper().eq("WASTE")))
-                        .or(modelMasterEntity.modelNo.eq(ModelType.G3.getId()))))
         .where(modelMasterEntity.id.eq(modelId))
         .fetch();
   }

src/main/java/com/kamco/cd/training/train/service/DockerTrainService.java

@@ -262,9 +262,9 @@ public class DockerTrainService {
     c.add("-v");
     c.add(basePath + ":" + basePath); // 심볼릭 링크와 연결되는 실제 파일 경로도 마운트를 해줘야 함
     c.add("-v");
-    c.add(symbolicDir + ":/data");    //요청할경로
+    c.add(symbolicDir + ":/data"); // 요청할경로
     c.add("-v");
-    c.add(responseDir + ":/checkpoints"); //저장될경로
+    c.add(responseDir + ":/checkpoints"); // 저장될경로
 
     // 표준입력 유지 (-it 대신 -i만 사용)
     c.add("-i");

src/main/java/com/kamco/cd/training/train/service/JobRecoveryOnStartupService.java

@@ -384,7 +384,20 @@ public class JobRecoveryOnStartupService {
         return new OutputResult(false, "total-epoch-missing");
       }
 
-      log.info("[RECOVERY] totalEpoch={}. jobId={}", totalEpoch, job.getId());
+      Integer valInterval = extractValInterval(job).orElse(null);
+      if (valInterval == null || valInterval <= 0) {
+        log.warn(
+            "[RECOVERY] valInterval missing or invalid. jobId={}, valInterval={}",
+            job.getId(),
+            valInterval);
+        return new OutputResult(false, "val-interval-missing");
+      }
+
+      log.info(
+          "[RECOVERY] totalEpoch={}. valInterval={}. jobId={}",
+          totalEpoch,
+          valInterval,
+          job.getId());
 
       // 3) val.csv 존재 확인
       Path valCsv = outDir.resolve("val.csv");
@@ -396,14 +409,17 @@ public class JobRecoveryOnStartupService {
       // 4) val.csv 라인 수 확인
       long lines = countNonHeaderLines(valCsv);
 
+      // expected = 실제 val 실행 횟수
+      int expectedLines = totalEpoch / valInterval;
+
       log.info(
           "[RECOVERY] val.csv lines counted. jobId={}, lines={}, expected={}",
           job.getId(),
           lines,
-          totalEpoch);
+          expectedLines);
 
       // 5) 완료 판정
-      if (lines == totalEpoch) {
+      if (lines >= expectedLines) {
         log.info("[RECOVERY] outputs look COMPLETE. jobId={}", job.getId());
         return new OutputResult(true, "ok");
       }
@@ -412,7 +428,7 @@ public class JobRecoveryOnStartupService {
           "[RECOVERY] val.csv line mismatch. jobId={}, lines={}, expected={}",
           job.getId(),
           lines,
-          totalEpoch);
+          expectedLines);
 
       return new OutputResult(
           false, "val.csv-lines-mismatch lines=" + lines + " expected=" + totalEpoch);
@@ -530,4 +546,19 @@ public class JobRecoveryOnStartupService {
       return reason;
     }
   }
+
+  /** paramsJson에서 valInterval 추출 */
+  private Optional<Integer> extractValInterval(ModelTrainJobDto job) {
+    Map<String, Object> params = job.getParamsJson();
+    if (params == null) return Optional.empty();
+
+    Object v = params.get("valInterval");
+    if (v == null) return Optional.empty();
+
+    try {
+      return Optional.of(Integer.parseInt(String.valueOf(v)));
+    } catch (Exception ignore) {
+      return Optional.empty();
+    }
+  }
 }

src/main/java/com/kamco/cd/training/train/service/ModelTrainMetricsJobService.java

@@ -85,34 +85,20 @@ public class ModelTrainMetricsJobService {
 
           int epoch = Integer.parseInt(record.get("Epoch"));
 
-          float aAcc = parseFloatSafe(record.get("aAcc"));
+          Float aAcc = parseFloatSafe(record.get("aAcc"));
-          float mFscore = parseFloatSafe(record.get("mFscore"));
+          Float mFscore = parseFloatSafe(record.get("mFscore"));
-          float mPrecision = parseFloatSafe(record.get("mPrecision"));
+          Float mPrecision = parseFloatSafe(record.get("mPrecision"));
-          float mRecall = parseFloatSafe(record.get("mRecall"));
+          Float mRecall = parseFloatSafe(record.get("mRecall"));
-          float mIoU = parseFloatSafe(record.get("mIoU"));
+          Float mIoU = parseFloatSafe(record.get("mIoU"));
-          float mAcc = parseFloatSafe(record.get("mAcc"));
+          Float mAcc = parseFloatSafe(record.get("mAcc"));
 
-          float changed_fscore = parseFloatSafe(record.get("changed_fscore"));
+          Float changed_fscore = parseFloatSafe(record.get("changed_fscore"));
-          float changed_precision = parseFloatSafe(record.get("changed_precision"));
+          Float changed_precision = parseFloatSafe(record.get("changed_precision"));
-          float changed_recall = parseFloatSafe(record.get("changed_recall"));
+          Float changed_recall = parseFloatSafe(record.get("changed_recall"));
 
-          float unchanged_fscore = parseFloatSafe(record.get("unchanged_fscore"));
+          Float unchanged_fscore = parseFloatSafe(record.get("unchanged_fscore"));
-          float unchanged_precision = parseFloatSafe(record.get("unchanged_precision"));
+          Float unchanged_precision = parseFloatSafe(record.get("unchanged_precision"));
-          float unchanged_recall = parseFloatSafe(record.get("unchanged_recall"));
+          Float unchanged_recall = parseFloatSafe(record.get("unchanged_recall"));
-          //          int epoch = Integer.parseInt(record.get("Epoch"));
-          //          float aAcc = Float.parseFloat(record.get("aAcc"));
-          //          float mFscore = Float.parseFloat(record.get("mFscore"));
-          //          float mPrecision = Float.parseFloat(record.get("mPrecision"));
-          //          float mRecall = Float.parseFloat(record.get("mRecall"));
-          //          float mIoU = Float.parseFloat(record.get("mIoU"));
-          //          float mAcc = Float.parseFloat(record.get("mAcc"));
-          //          float changed_fscore = Float.parseFloat(record.get("changed_fscore"));
-          //          float changed_precision = Float.parseFloat(record.get("changed_precision"));
-          //          float changed_recall = Float.parseFloat(record.get("changed_recall"));
-          //          float unchanged_fscore = Float.parseFloat(record.get("unchanged_fscore"));
-          //          float unchanged_precision =
-          // Float.parseFloat(record.get("unchanged_precision"));
-          //          float unchanged_recall = Float.parseFloat(record.get("unchanged_recall"));
 
           batchArgs.add(
               new Object[] {

src/main/java/com/kamco/cd/training/train/service/TmpDatasetService.java

@@ -19,6 +19,7 @@ public class TmpDatasetService {
 
   @Value("${train.docker.symbolic_link_dir}")
   private String symbolicDir;
+
   /**
    * train, val, test 폴더별로 link
    *

src/main/java/com/kamco/cd/training/train/service/TrainJobWorker.java

@@ -132,7 +132,9 @@ public class TrainJobWorker {
         String failMsg = result.getStatus() + "\n" + result.getLogs();
         log.info("training fail exitCode={} Msg ={}", result.getExitCode(), failMsg);
 
-        if (result.getExitCode() == -1 || result.getExitCode() == 143) {
+        if (result.getExitCode() == -1
+            || result.getExitCode() == 143
+            || result.getExitCode() == 137) {
           // 실패 처리
           modelTrainJobCoreService.markPaused(
               jobId, result.getExitCode(), result.getStatus() + "\n" + result.getLogs());