import torch
from src.abstract_translator import AbstractTranslator
from src.member.abducibles import abducibles, exclusive
from src.member.evaluate import Evaluator
from src.member.local_params import number_of_arguments
from src.member.manager import MemberManager
from src.networks.mnist_nets import COMP_NET
from src.params import useGPU
from src.run import scenario_test
network = COMP_NET()
if useGPU:
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
network.to(device)
translator = AbstractTranslator(abducibles, exclusive)
dataManager = MemberManager()
outputClasses = [10] * number_of_arguments
evaluator = Evaluator()
scenario = 'member/' + str(number_of_arguments)
# model_name = 'model_samples_3000_iter_400_epoch_1.mdl'
model_name = 'model_samples_3000_iter_9000_epoch_3.mdl'
if __name__ == '__main__':
scenario_test(network, outputClasses, translator, dataManager, scenario, model_name, evaluator)
re.search(pattern=r"iter_(\d*)",
string=iter_model.name).groups()[0])
iter_models[n_iter] = iter_model
# evaluate with each of the two test datasets E1 and E2
for test_dataset in test_datasets:
accuracies = list()
# evaluate model at each iteration (the models are sorted according to their iteration number)
for n_iter in sorted(iter_models.keys()):
model_name = iter_models[n_iter].name
accuracy = scenario_test(network,
outputClasses,
translator,
dataManager,
scenario_name,
model_name,
evaluator,
test_dataset=f"{test_dataset}.csv",
train_dataset=train_dataset)
accuracies.append([n_iter, accuracy])
results_path = Path(results_root) / scenario_name
results_path.mkdir(exist_ok=True)
# write results to files
with open(
results_path /
f"train_{train_dataset}_eval_{test_dataset}_results.csv",
"w") as f:
pd.DataFrame(data=accuracies,