Skip to content

Commit

Permalink
TCI
Browse files Browse the repository at this point in the history
  • Loading branch information
@TheColdIce
TheColdIce committed Nov 4, 2024
1 parent 1ed6134 commit 7cfd5fe
Show file tree
Hide file tree
Showing 24 changed files with 1,195 additions and 358 deletions.
2 changes: 1 addition & 1 deletion Dockerfile
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,4 +58,4 @@ RUN pip3 install --no-cache-dir -e .
COPY experiments experiments
WORKDIR /flib/experiments

ENTRYPOINT ["python3", "tune.py"]
ENTRYPOINT ["bash"]
287 changes: 240 additions & 47 deletions experiments/best_hyperparams.py
View file
Original file line number Diff line number Diff line change
@@ -1,88 +1,167 @@
LogRegClient_params = {
'centralized': {
'device': 'cuda:0',
'lr_patience': 5,
'es_patience': 15,
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 128,
'optimizer': 'SGD',
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.08507714201331565,
'momentum': 0.8343372940844335,
'weight_decay': 0.0015318920671180095,
'dampening': 0.6260300336301439,
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {},
},
'federated': {
'device': 'cuda:0',
'lr_patience': 5,
'es_patience': 15,
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 128,
'optimizer': 'SGD',
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.029577125072356544,
'momentum': 0.6954937697844237,
'weight_decay': 0.0016911232487892836,
'dampening': 0.007628568457341833,
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {}
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
'clients': {
'c0': {
'device': 'cuda:0',
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
},
'c1': {
'device': 'cuda:0',
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
},
'c2': {
'device': 'cuda:0',
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
}
}
},
'isolated': {
'device': 'cuda:0',
'lr_patience': 100,
'es_patience': 100,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
},
'clients': {
'c0': {
'device': 'cuda:0',
'lr_patience': 5,
'es_patience': 15,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 128,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.03937729543826933,
'weight_decay': 0.0005450625188668391,
'amsgrad': True
'lr': 0.04,
'weight_decay': 0.0,
'amsgrad': False
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {}
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
}
},
'c1': {
'device': 'cuda:0',
'lr_patience': 5,
'es_patience': 15,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 128,
'optimizer': 'SGD',
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.0924277036106537,
'momentum': 0.8913437725949631,
'weight_decay': 0.0006195976833673726,
'dampening': 0.12132015743487427,
'lr': 0.04,
'weight_decay': 0.0,
'amsgrad': False
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
}
},
'c2': {
'device': 'cuda:0',
'lr_patience': 5,
'es_patience': 15,
'n_rounds': 100,
'eval_every': 10,
'batch_size': 128,
'batch_size': 256,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.06800830792699754,
'weight_decay': 0.00012351437711755064,
'lr': 0.04,
'weight_decay': 0.0,
'amsgrad': False
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.6
}
}
}
}
Expand Down Expand Up @@ -240,23 +319,23 @@
'isolated': {
'clients': {
'c0': {
'criterion': 'log_loss',
'splitter': 'best',
'max_depth': 7,
'class_weight': None,
'criterion': 'entropy',
'splitter': 'random',
'max_depth': 12,
'class_weight': 'balanced',
'random_state': 42,
},
'c1': {
'criterion': 'gini',
'criterion': 'log_loss',
'splitter': 'random',
'max_depth': 8,
'max_depth': 10,
'class_weight': 'balanced',
'random_state': 42,
},
'c2': {
'criterion': 'entropy',
'splitter': 'random',
'max_depth': 8,
'splitter': 'best',
'max_depth': 9,
'class_weight': 'balanced',
'random_state': 42,
}
Expand Down Expand Up @@ -394,4 +473,118 @@
}
}
}
}

GraphSAGEClient_params = {
'centralized': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'hidden_dim': 128,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.001994010234629029,
'weight_decay': 0.018678778429870604,
'amsgrad': True,
},
'criterion': 'ClassBalancedLoss',
'criterion_params': {
'gamma': 0.9936137007812603
}
},
'federated': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'device': 'cuda:0',
'hidden_dim': 256,
'optimizer': 'SGD',
'optimizer_params': {
'lr': 0.048630848021750224,
'momentum': 0.3482146849308458,
'weight_decay': 0.25496952117003857,
'dampening': 0.2581674844412982
},
'criterion': 'NLLLoss',
'criterion_params': {
'weight': 0.8529780681438437
}
},
'isolated': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'batch_size': 128,
'hidden_dim': 64,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.01,
'weight_decay': 0.0,
'amsgrad': False,
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {},
'clients': {
'c0': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'hidden_dim': 512,
'optimizer': 'Adam',
'optimizer_params': {
'lr': 0.0010042395371481477,
'weight_decay': 0.14417159251403777,
'amsgrad': False,
},
'criterion': 'CrossEntropyLoss',
'criterion_params': {}
},
'c1': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'hidden_dim': 512,
'optimizer': 'SGD',
'optimizer_params': {
'lr': 0.07331504183228132,
'momentum': 0.23873608202007257,
'weight_decay': 0.14324551236448438,
'dampening': 0.662142529500039,
},
'criterion': 'NLLLoss',
'criterion_params': {
'weight': 0.7676710189846737
}
},
'c2': {
'device': 'cuda:0',
'n_rounds': 100,
'eval_every': 10,
'lr_patience': 100,
'es_patience': 100,
'hidden_dim': 32, #32,
'optimizer': 'SGD',
'optimizer_params': {
'lr': 0.009297617869520835,
'momentum': 0.10013231108969067,
'weight_decay': 0.12498499034883515,
'dampening': 0.4599952706941705,
},
'criterion': 'NLLLoss',
'criterion_params': {
'weight': 0.6950743876647961,
}
}
}
}
}
18 changes: 10 additions & 8 deletions experiments/generate.py
View file
Original file line number Diff line number Diff line change
@@ -1,16 +1,18 @@
import argparse
import os
from flib.sim import DataGenerator

def main():
# Parse arguments
parser = argparse.ArgumentParser()
parser.add_argument('--conf_file', type=str, help='Path to the config file', default='/home/edvin/Desktop/flib/experiments/param_files/3_banks_homo_mid/conf.json')
args = parser.parse_args()

generator = DataGenerator(args.conf_file)
def main(conf_file):
generator = DataGenerator(conf_file)
tx_log_file = generator()
print(f'\nSynthetic AML data generated\n Raw transaction log file: {tx_log_file}')
pass

if __name__ == "__main__":
main()
DATASET = '3_banks_homo_mid' #'30K_accts', '3_banks_homo_mid'
parser = argparse.ArgumentParser()
parser.add_argument('--conf_file', type=str, help='Path to the config file', default=f'/home/edvin/Desktop/flib/experiments/param_files/{DATASET}/conf.json')
args = parser.parse_args()
if not os.path.isabs(args.conf_file):
args.conf_file = os.path.abspath(args.conf_file)
main(args.conf_file)
Loading

0 comments on commit 7cfd5fe

Please sign in to comment.