Skip to content
Snippets Groups Projects

Compare revisions

Changes are shown as if the source revision was being merged into the target revision. Learn more about comparing revisions.

Source

Select target project
No results found

Target

Select target project
  • sacs/decentralizepy
  • mvujas/decentralizepy
  • randl/decentralizepy
3 results
Show changes
Hide whitespace changes
Inline Side-by-side
Showing
with 621 additions and 22 deletions
eval/step_configs/config_celeba_fft.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.FFT
sharing_class = FFT
alpha = 0.1
change_based_selection = True
accumulation = True
eval/step_configs/config_celeba_grow.ini
View file @ d140bf44
......@@ -2,23 +2,23 @@
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
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
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = Adam
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.GradientAccumulator
training_class = GradientAccumulator
rounds = 20
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 64
batch_size = 16
shuffle = True
loss_package = torch.nn
loss_class = CrossEntropyLoss
......
eval/step_configs/config_celeba_manualadapt.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.ManualAdapt
sharing_class = ManualAdapt
change_alpha = [0.1, 0.5]
change_rounds = [10,30]
eval/step_configs/config_celeba.ini → eval/step_configs/config_celeba_partialmodel.ini
View file @ d140bf44
......@@ -2,23 +2,23 @@
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
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
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = Adam
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.GradientAccumulator
training_class = GradientAccumulator
rounds = 20
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 64
batch_size = 16
shuffle = True
loss_package = torch.nn
loss_class = CrossEntropyLoss
......@@ -31,3 +31,4 @@ addresses_filepath = ip_addr_6Machines.json
[SHARING]
sharing_package = decentralizepy.sharing.PartialModel
sharing_class = PartialModel
alpha = 0.1
\ No newline at end of file
eval/step_configs/config_celeba_randomalpha.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.RandomAlpha
sharing_class = RandomAlpha
eval/step_configs/config_celeba_randomalphainc.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.RandomAlphaIncremental
sharing_class = RandomAlphaIncremental
eval/step_configs/config_celeba_roundrobin.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.RoundRobinPartial
sharing_class = RoundRobinPartial
alpha = 0.1
eval/step_configs/config_celeba_100.ini → eval/step_configs/config_celeba_sharing.ini
View file @ d140bf44
......@@ -2,23 +2,23 @@
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
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
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = Adam
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 20
rounds = 4
full_epochs = False
batch_size = 64
batch_size = 16
shuffle = True
loss_package = torch.nn
loss_class = CrossEntropyLoss
......
eval/step_configs/config_celeba_subsampling.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.SubSampling
sharing_class = SubSampling
alpha = 0.1
eval/step_configs/config_celeba_synchronous.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.Synchronous
sharing_class = Synchronous
\ No newline at end of file
eval/step_configs/config_celeba_topkacc.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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
alpha = 0.1
accumulation = True
accumulate_averaging_changes = True
\ No newline at end of file
eval/step_configs/config_celeba_topkacc_norm.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.TopKNormalized
sharing_class = TopKNormalized
alpha = 0.1
accumulation = True
accumulate_averaging_changes = True
\ No newline at end of file
eval/step_configs/config_celeba_topkaccbound.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.LowerBoundTopK
sharing_class = LowerBoundTopK
lower_bound = 0.1
alpha = 0.1
metro_hastings = False
accumulation = True
accumulate_averaging_changes = True
\ No newline at end of file
eval/step_configs/config_celeba_topkparam.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.TopKParams
sharing_class = TopKParams
alpha = 0.1
eval/step_configs/config_celeba_topkrandom.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.TopKPlusRandom
sharing_class = TopKPlusRandom
alpha = 0.1
eval/step_configs/config_celeba_wavelet.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.Celeba
dataset_class = Celeba
model_class = CNN
images_dir = /mnt/nfs/shared/leaf/data/celeba/data/raw/img_align_celeba
train_dir = /mnt/nfs/shared/leaf/data/celeba/per_user_data/train
test_dir = /mnt/nfs/shared/leaf/data/celeba/data/test
; python list of fractions below
sizes =
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 4
full_epochs = False
batch_size = 16
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.Wavelet
sharing_class = Wavelet
change_based_selection = True
alpha = 0.1
wavelet=sym2
level= 4
accumulation = True
accumulate_averaging_changes = True
eval/step_configs/config_cifar_SharingWithRWAsyncDynamic.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.CIFAR10
dataset_class = CIFAR10
model_class = LeNet
train_dir = /mnt/nfs/shared/CIFAR
test_dir = /mnt/nfs/shared/CIFAR
; python list of fractions below
sizes =
random_seed = 99
partition_niid = True
shards = 1
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 65
full_epochs = False
batch_size = 8
shuffle = True
loss_package = torch.nn
loss_class = CrossEntropyLoss
[COMMUNICATION]
comm_package = decentralizepy.communication.TCPRandomWalkRouting
comm_class = TCPRandomWalkRouting
addresses_filepath = ip_addr_6Machines.json
sampler = equi
[SHARING]
sharing_package = decentralizepy.sharing.SharingWithRWAsyncDynamic
sharing_class = SharingWithRWAsyncDynamic
\ No newline at end of file
eval/step_configs/config_cifar_partialmodel.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.CIFAR10
dataset_class = CIFAR10
model_class = LeNet
train_dir = /mnt/nfs/shared/CIFAR
test_dir = /mnt/nfs/shared/CIFAR
; python list of fractions below
sizes =
random_seed = 99
partition_niid = True
shards = 4
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 65
full_epochs = False
batch_size = 8
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
alpha=0.5
eval/step_configs/config_cifar_sharing.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.CIFAR10
dataset_class = CIFAR10
model_class = LeNet
train_dir = /mnt/nfs/shared/CIFAR
test_dir = /mnt/nfs/shared/CIFAR
; python list of fractions below
sizes =
random_seed = 99
partition_niid = True
shards = 4
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 65
full_epochs = False
batch_size = 8
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_cifar_subsampling.ini 0 → 100644
View file @ d140bf44
[DATASET]
dataset_package = decentralizepy.datasets.CIFAR10
dataset_class = CIFAR10
model_class = LeNet
train_dir = /mnt/nfs/shared/CIFAR
test_dir = /mnt/nfs/shared/CIFAR
; python list of fractions below
sizes =
random_seed = 99
partition_niid = True
shards = 4
[OPTIMIZER_PARAMS]
optimizer_package = torch.optim
optimizer_class = SGD
lr = 0.001
[TRAIN_PARAMS]
training_package = decentralizepy.training.Training
training_class = Training
rounds = 65
full_epochs = False
batch_size = 8
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.SubSampling
sharing_class = SubSampling
alpha = 0.5