def GetEDLLoss(pred, true):
if type(pred) is not torch.Tensor:
pred = torch.Tensor(pred).float().cuda()
elif pred.is_cuda is False:
pred = pred.cuda()
if type(true) is not torch.Tensor:
true = torch.Tensor(true).float().cuda()
elif true.is_cuda is False:
true = true.cuda()
loss = EDL_Loss()(true, pred).item()
return loss
from torch import optim
import joblib
import os
import argparse
import numpy as np
from model import CNNText
from util import PyTorchParameterList2NPArrayList
from loss import EDL_Loss
# config #
inner_rate = 0.01
outer_rate = 0.1
n_subsets = 5
ntrain = 10
loss_fn = EDL_Loss()
print('[Start maml ...]')
parser = argparse.ArgumentParser()
parser.add_argument("--file_path",
help="saving root path of raw data",
default='./test')
parser.add_argument("--seed",
help="reproducible experiment with seeds",
type=int)
parser.add_argument("--out_dim", help="output dimension", type=int, default=6)
parser.add_argument('--niterations',
help='number of iterations',
default=1000,
type=int)