!1695 fix some bug in ProtoNet & Pnasnet

Merge pull request !1695 from zhaoting/master

research/cv/ProtoNet/scripts/run_distribution_gpu.sh

@@ -29,7 +29,7 @@ get_real_path(){
 export DATA_PATH=$(get_real_path $1) # dataset path
 export TRAIN_CLASS=$2 # train class, propose 20
 export EPOCHS=$3 # num of epochs
-export DEVICE_NUM=$4 # device_num
+export RANK_SIZE=$4 # device_num
 
 if [ ! -d $DATA_PATH ]
 then
@@ -40,7 +40,7 @@ fi
 rm -rf distribute_output
 mkdir distribute_output
 
-mpirun --allow-run-as-root -n $DEVICE_NUM --output-filename log_output --merge-stderr-to-stdout \
+mpirun --allow-run-as-root -n $RANK_SIZE --output-filename log_output --merge-stderr-to-stdout \
   python ../train.py --dataset_root=$DATA_PATH \
                 --device_target="GPU" \
                 --classes_per_it_tr=$TRAIN_CLASS \

research/cv/ProtoNet/train.py

@@ -24,6 +24,7 @@ from mindspore import dataset as ds
 import mindspore.context as context
 from mindspore.communication.management import init, get_rank
 from mindspore.context import ParallelMode
+from mindspore.common import set_seed
 from src.EvalCallBack import EvalCallBack
 from src.protonet import WithLossCell
 from src.PrototypicalLoss import PrototypicalLoss
@@ -33,7 +34,7 @@ from model_init import init_dataloader
 
 local_data_url = './cache/data'
 local_train_url = './cache/out'
-
+set_seed(1)
 
 def train(opt, tr_dataloader, net, loss_fn, eval_loss_fn, optim, path, rank_id, val_dataloader=None):
     '''
@@ -68,17 +69,15 @@ def main():
     global local_train_url
 
     options = get_parser().parse_args()
-    device_num = int(os.environ.get("DEVICE_NUM", 1))
-
-    if options.device_target == "GPU":
-        rank_id = get_rank()
-
+    device_num = int(os.environ.get("RANK_SIZE", "1"))
+    rank_id = 0
     if options.run_offline:
         if device_num > 1:
             init()
             context.reset_auto_parallel_context()
             context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
                                               gradients_mean=True)
+            rank_id = get_rank()
 
         if options.device_target == "Ascend":
             context.set_context(device_id=options.device_id)
@@ -105,6 +104,7 @@ def main():
             context.reset_auto_parallel_context()
             context.set_auto_parallel_context(device_num=device_num, parallel_mode=ParallelMode.DATA_PARALLEL,
                                               gradients_mean=True)
+            rank_id = get_rank()
             local_data_url = os.path.join(local_data_url, str(device_id))
             local_train_url = os.path.join(local_train_url, str(device_id))
 

research/cv/pnasnet/default_config.yaml

@@ -39,8 +39,6 @@ cutout_length: 56
 # Dataset config
 train_batch_size: 32
 val_batch_size: 125
-# True for GPU, False for Ascend
-drop_remainder: False
 
 #learning rate config
 lr_init: 0.32

research/cv/pnasnet/eval.py

@@ -72,10 +72,9 @@ if __name__ == '__main__':
     else:
         dataset_val_path = os.path.join(dataset_path, 'val')
 
-    drop_remainder = config.drop_remainder
     dataset = create_dataset(dataset_val_path, do_train=False, rank=device_id,
                              group_size=1, batch_size=config.val_batch_size,
-                             drop_remainder=drop_remainder, shuffle=False)
+                             drop_remainder=False, shuffle=False)
 
     loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')
 

research/cv/pnasnet/train.py

@@ -118,12 +118,10 @@ def train():
     else:
         dataset_train_path = os.path.join(dataset_path, 'train')
 
-    drop_remainder = config.drop_remainder
-
     train_dataset = create_dataset(dataset_train_path, True, config.rank, config.group_size,
                                    num_parallel_workers=config.work_nums,
                                    batch_size=config.train_batch_size,
-                                   drop_remainder=drop_remainder, shuffle=True,
+                                   drop_remainder=True, shuffle=True,
                                    cutout=config.cutout, cutout_length=config.cutout_length)
     train_batches_per_epoch = train_dataset.get_dataset_size()