def train_inception_opp():
print("Running Opportunity")
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = InceptionModel(num_blocks=1,
in_channels=113,
out_channels=32,
bottleneck_channels=2,
kernel_sizes=20,
use_residuals=True,
num_pred_classes=17)
model.to(device)
trainer = OPPTrainer(model=model)
#trainer.fit()
wandb.agent(sweep_id, function=trainer.fit, count=15)
def train_inception_har():
print("Running HAR")
data_folder = Path('../data/UCI_HAR_Dataset')
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
model = InceptionModel(num_blocks=1,
in_channels=9,
out_channels=32,
bottleneck_channels=2,
kernel_sizes=20,
use_residuals=True,
num_pred_classes=6)
model.to(device)
trainer = HARTrainer(model=model, data_folder=data_folder)
#trainer.fit()
wandb.agent(sweep_id, function=trainer.fit, count=648)
def train_inception_wisdm():
print("Running WISDM")
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = InceptionModel(num_blocks=1,
in_channels=3,
out_channels=32,
bottleneck_channels=2,
kernel_sizes=20,
use_residuals=True,
num_pred_classes=6)
model.to(device)
trainer = WISDMTrainer(model=model)
trainer.fit()
savepath = trainer.save_model()
#savepath = Path("data/models/InceptionModel/InceptionModel_model_wisdm_20_epochs.pkl")
new_trainer = load_wisdm_trainer(savepath)
new_trainer.evaluate()
def test_fit_works_for_single_and_multiclass(self, tmp_path,
num_pred_classes):
in_channels = 30
model = InceptionModel(2,
in_channels,
out_channels=30,
bottleneck_channels=12,
kernel_sizes=15,
use_residuals=True,
num_pred_classes=num_pred_classes)
trainer = TrainerForTests(model,
tmp_path,
in_channels,
num_preds=num_pred_classes)
# this just ensures everything runs
trainer.fit(batch_size=50, num_epochs=1)
def train_inception_sc():
data_folder = Path('../data')
model = InceptionModel(num_blocks=1,
in_channels=1,
out_channels=2,
bottleneck_channels=2,
kernel_sizes=41,
use_residuals=True,
num_pred_classes=6)
trainer = UCRTrainer(model=model,
experiment='synthetic_control',
data_folder=data_folder)
trainer.fit()
savepath = trainer.save_model()
new_trainer = load_ucr_trainer(savepath)
new_trainer.evaluate()