Merge pull request '파라미터 변경' (#77) from feat/training_260202 into develop

Reviewed-on: #77

src/main/java/com/kamco/cd/training/train/dto/EvalRunRequest.java

@@ -13,4 +13,10 @@ public class EvalRunRequest {
   private String uuid;
   private int epoch; // best_changed_fscore_epoch_1.pth
   private Integer timeoutSeconds;
+  private String datasetFolder;
+  private String outputFolder;
+
+  public String getOutputFolder() {
+    return this.outputFolder.toString();
+  }
 }

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

@@ -311,6 +311,7 @@ public class DockerTrainService {
     addArg(c, "--hue-delta", req.getHueDelta());
 
     addArg(c, "--resume-from", req.getResumeFrom());
+    addArg(c, "--save-interval", 1);
     return c;
   }
 
@@ -414,30 +415,28 @@ public class DockerTrainService {
 
     c.add("docker");
     c.add("run");
-    c.add("--name");
-    c.add(containerName);
     c.add("--rm");
-
     c.add("--gpus");
     c.add("all");
-    if (ipcHost) c.add("--ipc=host");
+    c.add("--ipc=host");
     c.add("--shm-size=" + shmSize);
 
     c.add("-v");
     c.add("/home/kcomu/data" + "/tmp:/data");
+
     c.add("-v");
     c.add(responseDir + ":/checkpoints");
 
-    c.add(image);
+    c.add("kamco-cd-train:latest");
 
     c.add("python");
     c.add("/workspace/change-detection-code/run_evaluation_pipeline.py");
 
-    c.add("--dataset_dir");
+    addArg(c, "--dataset-folder", req.getDatasetFolder());
-    c.add("/data/" + uuid);
+    addArg(c, "--output-folder", req.getOutputFolder());
 
-    c.add("--model");
+    c.add("--epoch");
-    c.add("/checkpoints/" + uuid + "/" + modelFile);
+    c.add(modelFile);
 
     return c;
   }

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

@@ -5,6 +5,7 @@ import com.kamco.cd.training.model.dto.ModelTrainMngDto;
 import com.kamco.cd.training.postgres.core.ModelTrainJobCoreService;
 import com.kamco.cd.training.postgres.core.ModelTrainMngCoreService;
 import com.kamco.cd.training.train.dto.ModelTrainJobQueuedEvent;
+import com.kamco.cd.training.train.dto.TrainRunRequest;
 import java.time.ZonedDateTime;
 import java.util.Map;
 import java.util.UUID;
@@ -32,10 +33,15 @@ public class TestJobService {
     // best epoch 업데이트
     modelTrainMngCoreService.updateModelMasterBestEpoch(modelId, epoch);
 
+    // 파라미터 조회
+    TrainRunRequest trainRunRequest = modelTrainMngCoreService.findTrainRunRequest(modelId);
+
     Map<String, Object> params = new java.util.LinkedHashMap<>();
     params.put("jobType", "EVAL");
     params.put("uuid", String.valueOf(uuid));
     params.put("epoch", epoch);
+    params.put("datasetFolder", trainRunRequest.getDatasetFolder());
+    params.put("outputFolder", trainRunRequest.getOutputFolder());
 
     int nextAttemptNo = modelTrainJobCoreService.findMaxAttemptNo(modelId) + 1;
 

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

@@ -68,8 +68,16 @@ public class TrainJobWorker {
         modelTrainMngCoreService.markStep2InProgress(modelId, jobId);
         String uuid = String.valueOf(params.get("uuid"));
         int epoch = (int) params.get("epoch");
+        String datasetFolder = String.valueOf(params.get("datasetFolder"));
+        String outputFolder = String.valueOf(params.get("outputFolder"));
+
+        EvalRunRequest evalReq = new EvalRunRequest();
+        evalReq.setUuid(uuid);
+        evalReq.setEpoch(epoch);
+        evalReq.setTimeoutSeconds(null);
+        evalReq.setDatasetFolder(datasetFolder);
+        evalReq.setOutputFolder(outputFolder);
 
-        EvalRunRequest evalReq = new EvalRunRequest(uuid, epoch, null);
         result = dockerTrainService.runEvalSync(evalReq, containerName);
 
       } else {