improve config

benchmarks/cloudmask/target/greene_v0.5/cloudmask_v0.5.in.slurm

@@ -84,7 +84,7 @@ echo "USER_SCRATCH: $USER_SCRATCH"
 echo "PROJECT_DIR:  $PROJECT_DIR"
 echo "PYTHON_DIR:   $PYTHON_DIR"
 echo "PROJECT_DATA: $PROJECT_DATA"
-echo "CONTAINERDIR: $CONTAINERDIR"
+echo "TARGET: $TARGET"
 
 
 # mkdir -p $OUTPUTS_DIR
@@ -144,7 +144,7 @@ echo "# check filesystem"
 echo "# ==================================="
 pwd
 ls
-singularity exec --nv $CONTAINERDIR/cloudmask.sif bash -c "python -c \"import os; os.system('ls')\""
+singularity exec --nv ${TARGET}/cloudmask.sif bash -c "python -c \"import os; os.system('ls')\""
 
 PROGRESS "running" 10
 
@@ -159,15 +159,15 @@ echo "# ==================================="
 echo "# start gpu log"
 echo "# ==================================="
 
-cms gpu watch --gpu=0 --delay=0.5 --dense > ${TARGET}/project/{ee.identifier}/gpu0.log &
+cms gpu watch --gpu=0 --delay=0.5 --dense > ${OUTPUTS_DIR}/gpu0.log &
 
 PROGRESS "running" 21
 
 echo "# ==================================="
 echo "# start cloudmask"
 echo "# ==================================="
 
-singularity exec --nv ${CONTAINERDIR}/cloudmask.sif bash -c "cd ${TARGET}/project/{ee.identifier} ; python cloudmask_v0.5.py --config=config.yaml"
+singularity exec --nv ${TARGET}/cloudmask.sif bash -c "cd ${OUTPUTS_DIR} ; python cloudmask_v0.5.py --config=config.yaml"
 
 PROGRESS "running" 99
 

benchmarks/cloudmask/target/greene_v0.5/cloudmask_v0.5.py

@@ -110,7 +110,7 @@ def cloud_inference(config) -> None:
     N_CHANNELS = config['image.N_CHANNELS']
 
     # Load model
-    # modelPath = os.path.expanduser(config['model_file'])
+    # modelPath = os.path.expanduser(config['output.model_file'])
     model = tf.keras.models.load_model(modelPath)
 
     # Read inference files
@@ -150,7 +150,7 @@ def cloud_inference(config) -> None:
         mask = reconstruct_from_patches(config, mask_patches, nx, ny, patch_size=PATCH_SIZE - CROP_SIZE)
 
         # Save reconstructed image (mask)
-        output_dir = os.path.expanduser(config['output_dir'])
+        output_dir = os.path.expanduser(config['output.directory'])
         mask_name = f"{output_dir}/{file_name.name}.h5"
         with h5py.File(mask_name, 'w') as handle:
             handle.create_dataset('mask', data=mask)
@@ -308,7 +308,7 @@ def string_to_boolean(input_string):
     # else: # mode: original
     #     modelPath = os.path.expanduser(config['model_file'])
 
-    modelPath = os.path.expanduser(config['model_file'])
+    modelPath = os.path.expanduser(config['output.model_file'])
 
     tf.keras.models.save_model(model, modelPath)
     banner('END slstr_cloud in training mode.')
@@ -364,11 +364,11 @@ def main():
     print(config)
 
     # setup
-    log_file = os.path.expanduser(config['log_file'])
+    log_file = os.path.expanduser(config['output.log_file'])
     user_name = config["submission.submitter"]
 
     # MLCommons logging
-    mlperf_logfile = os.path.expanduser(config['mlperf_logfile'])
+    mlperf_logfile = os.path.expanduser(config['output.mlperf_logfile'])
     mllog.config(filename=mlperf_logfile)
 
     print("user", user_name)

benchmarks/cloudmask/target/greene_v0.5/config-rivanna.in.yaml

@@ -101,13 +101,22 @@ training:
 # Inference data
 #  inference_dir: "/scratch/{os.USER}/data/cloudmask/data/ssts"
   # Output directory
-output_dir: "{os.TARGET}/outputs"
+output:
+  directory: "./outputs"
+  # Model file
+  model_file: "./outputs/cloudModel.h5"
+  # Log file for recording runtimes
+  log_file: "./outputs/cloudmask_final.log"
+  # Log file for MLCommons logging
+  mlperf_logfile: "./outputs/mlperf_cloudmask_final.log"
+
+output_dir: "./outputs"
 # Model file
-model_file: "{output_dir}/cloudModel-{identifier}.h5"
+model_file: "./outputs/cloudModel.h5"
 # Log file for recording runtimes
-log_file: "{output_dir}/cloudmask_final_{identifier}.log"
+log_file: "./outputs/cloudmask_final.log"
 # Log file for MLCommons logging
-mlperf_logfile: "{output_dir}/mlperf_cloudmask_final_{identifier}.log"
+mlperf_logfile: "./outputs/mlperf_cloudmask_final.log"
 
 
 data:

benchmarks/cloudmask/target/greene_v0.5/config.in.yaml

@@ -60,13 +60,14 @@ experiment:
 #  card_name: a100-dgx
   # card_name: "a100,v100,p100,k80,rtx2080,rtx3090"
   #card_name: "a100"
-  directive: "a100-dgx"
+  # directive: "a100-dgx,a100,v100,p100,k80,rtx2080,rtx3090"
+  directive: "a100-dgx,v100"
   gpu_count: "1"
   cpu_num: 1
   mem: "64GB"
   repeat: "1"
   # epoch: "2,10,30,50,70,100,200"
-  epoch: "1"
+  epoch: "1,2"
   seed: 1234
   learning_rate: 0.001
   batch_size: 32
@@ -78,7 +79,7 @@ experiment:
   early_stoppage_patience: "25"
   early_stoppage: "False"
 
-identifier: "{experiment.card_name}-{experiment.early_stoppage}-{experiment.early_stoppage_patience}-{experiment.epoch}-{experiment.learning_rate}-{experiment.repeat}"
+identifier: "{experiment.directive}_{experiment.early_stoppage}_{experiment.early_stoppage_patience}_{experiment.epoch}_{experiment.learning_rate}_{experiment.repeat}"
 
 system:
   host: "rivanna"
@@ -100,13 +101,22 @@ training:
 # Inference data
 #  inference_dir: "/scratch/{os.USER}/data/cloudmask/data/ssts"
   # Output directory
-output_dir: "{os.TARGET}/outputs"
+output:
+  directory: "./outputs"
+  # Model file
+  model_file: "./outputs/cloudModel.h5"
+  # Log file for recording runtimes
+  log_file: "./outputs/cloudmask_final.log"
+  # Log file for MLCommons logging
+  mlperf_logfile: "./outputs/mlperf_cloudmask_final.log"
+
+output_dir: "./outputs"
 # Model file
-model_file: "{output_dir}/cloudModel-{identifier}.h5"
+model_file: "./outputs/cloudModel.h5"
 # Log file for recording runtimes
-log_file: "{output_dir}/cloudmask_final_{identifier}.log"
+log_file: "./outputs/cloudmask_final.log"
 # Log file for MLCommons logging
-mlperf_logfile: "{output_dir}/mlperf_cloudmask_final_{identifier}.log"
+mlperf_logfile: "./outputs/mlperf_cloudmask_final.log"
 
 
 data:

benchmarks/cloudmask/target/greene_v0.5/deprecated/slstr_cloud.py

@@ -107,7 +107,7 @@ def cloud_inference(config) -> None:
     N_CHANNELS = config['image.N_CHANNELS']
 
     # Load model
-    # modelPath = os.path.expanduser(config['model_file'])
+    # modelPath = os.path.expanduser(config['output.model_file'])
     model = tf.keras.models.load_model(modelPath)
 
     # Read inference files
@@ -147,7 +147,7 @@ def cloud_inference(config) -> None:
         mask = reconstruct_from_patches(config, mask_patches, nx, ny, patch_size=PATCH_SIZE - CROP_SIZE)
 
         # Save reconstructed image (mask)
-        output_dir = os.path.expanduser(config['output_dir'])
+        output_dir = os.path.expanduser(config['output.directory'])
         mask_name = f"{output_dir}/{file_name.name}.h5"
         with h5py.File(mask_name, 'w') as handle:
             handle.create_dataset('mask', data=mask)