def transformer_bce_kfold(fv):
folds = list(range(1, 11))
for fold in folds:
model_name = "transformer_%s_bce_fold%d" % (fv, fold)
criterion = torch.nn.BCELoss(reduction='none')
mtrain.train(fv,
model_name,
criterion,
balance=False,
batchsize=batchsize,
fold=fold)
def fix_break_run():
folds = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
fv = 'matlab'
for fold in folds:
model_name = "transformer_%s_bce_fold%d" % (fv, fold)
criterion = torch.nn.BCELoss(reduction='none')
mtrain.train(fv,
model_name,
criterion,
balance=False,
batchsize=32,
fold=fold)
def transformer_bce_gbalance(fv, size=0):
model_name = "transformer_%s_size%d_bce_gbalance" % (fv, size)
criterion = torch.nn.BCELoss(reduction='none')
mtrain.train(fv, model_name, criterion, balance=True, batchsize=batchsize)
def transformer_fec1(fv, size=0):
model_name = "transformer_%s_size%d_fec1" % (fv, size)
criterion = torch_util.FECLoss(alpha=batchsize * 1, reduction='none')
mtrain.train(fv, model_name, criterion)