models.append(
{
"model":GNCLClassifier,
"n_estimators":m,
"mode":"upper",
"l_reg":l_reg,
"combination_type":"average",
"base_estimator": partial(simpleresnet, size=s, model_type=t),
"optimizer":optimizer,
"scheduler":scheduler,
"loader":loader,
"eval_every":5,
"store_every":0,
"loss_function":nn.CrossEntropyLoss(reduction="none"),
"use_amp":True,
"device":"cuda",
"train_data": torchvision.datasets.FashionMNIST(".", train=True, transform = train_transformation()),
"test_data": torchvision.datasets.FashionMNIST(".", train=False, transform = test_transformation()),
"verbose":True
}
)
try:
base = models[0]["base_estimator"]().cuda()
rnd_input = torch.rand((1, 1, 28, 28)).cuda()
print(summary(base, rnd_input))
except:
pass
run_experiments(basecfg, models)
"sigma":s,
"scale":1,
"epsilon":e
})
)
runs.append(
( {
"method": "SieveStreaming++",
"K":K,
"sigma":s,
"scale":1,
"epsilon":e
})
)
for T in Ts:
runs.append(
( {
"method": "ThreeSieves",
"K":K,
"sigma":s,
"scale":1,
"epsilon":e,
"T":T
})
)
# random.shuffle(runs)
run_experiments(basecfg, runs)
