Modularize node, add step training

eval/config_celeba.ini → eval/epoch_configs/config_celeba.ini

@@ -17,7 +17,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.GradientAccumulator
 training_class = GradientAccumulator
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 512
 shuffle = True
 loss_package = torch.nn

eval/config_celeba_100.ini → eval/epoch_configs/config_celeba_100.ini

@@ -17,7 +17,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.Training
 training_class = Training
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 512
 shuffle = True
 loss_package = torch.nn

eval/config_celeba_grow.ini → eval/epoch_configs/config_celeba_grow.ini

@@ -17,7 +17,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.GradientAccumulator
 training_class = GradientAccumulator
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 512
 shuffle = True
 loss_package = torch.nn

eval/config_femnist.ini → eval/epoch_configs/config_femnist.ini

@@ -16,7 +16,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.GradientAccumulator
 training_class = GradientAccumulator
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 1024
 shuffle = True
 loss_package = torch.nn

eval/config_femnist_100.ini → eval/epoch_configs/config_femnist_100.ini

@@ -16,7 +16,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.Training
 training_class = Training
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 1024
 shuffle = True
 loss_package = torch.nn

eval/config_femnist_grow.ini → eval/epoch_configs/config_femnist_grow.ini

@@ -16,7 +16,8 @@ lr = 0.001
 [TRAIN_PARAMS]
 training_package = decentralizepy.training.GradientAccumulator
 training_class = GradientAccumulator
-epochs_per_round = 5
+rounds = 5
+full_epochs = True
 batch_size = 1024
 shuffle = True
 loss_package = torch.nn

eval/ip_addr_7Machines.json

+{
+    "0": "10.90.41.127",
+    "1": "10.90.41.128",
+    "2": "10.90.41.129",
+    "3": "10.90.41.130",
+    "4": "10.90.41.131",
+    "5": "10.90.41.132",
+    "6": "10.90.41.133"
+}
\ No newline at end of file

eval/step_configs/config_celeba.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Celeba
+dataset_class = Celeba
+model_class = CNN
+n_procs = 96
+images_dir = /home/risharma/leaf/data/celeba/data/raw/img_align_celeba
+train_dir = /home/risharma/leaf/data/celeba/per_user_data/train
+test_dir = /home/risharma/leaf/data/celeba/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.GradientAccumulator
+training_class = GradientAccumulator
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_6Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.PartialModel
+sharing_class = PartialModel

eval/step_configs/config_celeba_100.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Celeba
+dataset_class = Celeba
+model_class = CNN
+n_procs = 96
+images_dir = /home/risharma/leaf/data/celeba/data/raw/img_align_celeba
+train_dir = /home/risharma/leaf/data/celeba/per_user_data/train
+test_dir = /home/risharma/leaf/data/celeba/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.Training
+training_class = Training
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_6Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.Sharing
+sharing_class = Sharing

eval/step_configs/config_celeba_grow.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Celeba
+dataset_class = Celeba
+model_class = CNN
+n_procs = 96
+images_dir = /home/risharma/leaf/data/celeba/data/raw/img_align_celeba
+train_dir = /home/risharma/leaf/data/celeba/per_user_data/train
+test_dir = /home/risharma/leaf/data/celeba/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.GradientAccumulator
+training_class = GradientAccumulator
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_6Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.GrowingAlpha
+sharing_class = GrowingAlpha
+init_alpha=0.10
+max_alpha=0.75
+k=6
+metadata_cap=0.65

eval/step_configs/config_femnist.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Femnist
+dataset_class = Femnist
+model_class = CNN
+n_procs = 16
+train_dir = /home/risharma/leaf/data/femnist/per_user_data/train
+test_dir = /home/risharma/leaf/data/femnist/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.GradientAccumulator
+training_class = GradientAccumulator
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_6Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.PartialModel
+sharing_class = PartialModel

eval/step_configs/config_femnist_100.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Femnist
+dataset_class = Femnist
+model_class = CNN
+n_procs = 16
+train_dir = /home/risharma/leaf/data/femnist/per_user_data/train
+test_dir = /home/risharma/leaf/data/femnist/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.Training
+training_class = Training
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_7Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.Sharing
+sharing_class = Sharing

eval/step_configs/config_femnist_grow.ini

+[DATASET]
+dataset_package = decentralizepy.datasets.Femnist
+dataset_class = Femnist
+model_class = CNN
+n_procs = 16
+train_dir = /home/risharma/leaf/data/femnist/per_user_data/train
+test_dir = /home/risharma/leaf/data/femnist/data/test
+; python list of fractions below
+sizes = 
+
+[OPTIMIZER_PARAMS]
+optimizer_package = torch.optim
+optimizer_class = Adam
+lr = 0.001
+
+[TRAIN_PARAMS]
+training_package = decentralizepy.training.GradientAccumulator
+training_class = GradientAccumulator
+rounds = 20
+full_epochs = False
+batch_size = 64
+shuffle = True
+loss_package = torch.nn
+loss_class = CrossEntropyLoss
+
+[COMMUNICATION]
+comm_package = decentralizepy.communication.TCP
+comm_class = TCP
+addresses_filepath = ip_addr_6Machines.json
+
+[SHARING]
+sharing_package = decentralizepy.sharing.GrowingAlpha
+sharing_class = GrowingAlpha
+init_alpha=0.10
+max_alpha=0.75
+k=8
+metadata_cap=0.65
\ No newline at end of file