def __init__(self, arch, model, lr=1e-5, batch_size=64):
self.arch = arch
if not os.path.exists(f'{arch}/ckpt/'):
os.makedirs(f'{arch}/ckpt/')
self.device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
self.batch_size = batch_size
self.model = model
self.model.to(self.device)
self.criterion = nn.MSELoss()
self.optimizer = torch.optim.Adam(self.model.parameters(), lr=lr, weight_decay=1e-4)
self.scheduler = StepLR(self.optimizer, step_size=100, gamma=0.5)
self.history = {'loss':[]}
same_seeds(0)
def __init__(self, arch, model, lr, batch_size, wd):
self.arch = arch
if not os.path.exists(f'{arch}/ckpt/'):
os.makedirs(f'{arch}/ckpt/')
self.device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
self.batch_size = batch_size
self.model = model
self.model.to(self.device)
self.criterion = nn.MSELoss()
self.opt = torch.optim.Adam(self.model.parameters(), lr=lr)
self.scheduler = StepLR(self.opt, step_size=50, gamma=0.1)
self.history = {'train_loss': [], 'valid_loss': []}
self.min_loss = math.inf
same_seeds(73)
def __init__(self, arch, model, batch_size, lr, accum_steps, device):
self.arch = arch
if not os.path.exists(arch):
os.makedirs(arch)
self.model = model
self.batch_size = batch_size
self.opt = torch.optim.Adam(self.model.parameters(),
lr=lr,
weight_decay=1e-2)
self.scheduler = StepLR(self.opt, step_size=20, gamma=0.5)
self.accum_steps = accum_steps
self.criteria = torch.nn.CrossEntropyLoss()
self.device = device
self.history = {'train': [], 'valid': []}
self.best_score = math.inf
same_seeds(73)
def __init__(self, arch, device):
self.arch = arch
if not os.path.exists(f'{arch}/ckpts'):
os.makedirs(f'{arch}/ckpts')
self.feature_extractor = FeatureExtractor().to(device)
self.label_predictor = LabelPredictor().to(device)
self.optimizer_F = optim.Adam(self.feature_extractor.parameters(),
lr=1e-4)
self.optimizer_C = optim.Adam(self.label_predictor.parameters(),
lr=1e-4)
self.class_criterion = nn.CrossEntropyLoss()
self.device = device
self.history = {'loss': [], 'acc': []}
self.best_score = math.inf
same_seeds(73)
import os
import sys
import numpy as np
import torch
from torch import optim
from torch.utils.data import DataLoader
import torch.nn as nn
from utils import same_seeds
from dataset_baseline import Image_Dataset
from dataset_baseline import preprocess
from model_baseline import AE
same_seeds(0)
input_filename = sys.argv[1] # ~/Downloads/dataset/trainX.npy
#output_modeldir = sys.argv[2] # ./model
output_filename = sys.argv[2] # ./chekpoints/baseline.pth
# dataset
trainX = np.load(sys.argv[1])
print("trainX", trainX.shape)
trainX_preprocessed = preprocess(trainX)
print("trainX_preprocessed", trainX_preprocessed.shape)
img_dataset = Image_Dataset(trainX_preprocessed)
img_dataloader = DataLoader(img_dataset, batch_size=64, shuffle=True)
# model, loss, optimizer
model = AE().cuda()
criterion = nn.MSELoss()
optimizer = torch.optim.Adam(model.parameters(),
lr=1 * 1e-5,
TAR_INPUT_DIMENSION = args.tar_input_dim
N_DIMENSION = args.n_dim
CLASS_NUM = args.class_num
SHOT_NUM_PER_CLASS = args.shot_num_per_class
QUERY_NUM_PER_CLASS = args.query_num_per_class
EPISODE = args.episode
TEST_EPISODE = args.test_episode
LEARNING_RATE = args.learning_rate
GPU = args.gpu
HIDDEN_UNIT = args.hidden_unit
# Hyper Parameters in target domain data set
TEST_CLASS_NUM = args.test_class_num # the number of class
TEST_LSAMPLE_NUM_PER_CLASS = args.test_lsample_num_per_class # the number of labeled samples per class 5 4 3 2 1
utils.same_seeds(0)
def _init_():
if not os.path.exists('checkpoints'):
os.makedirs('checkpoints')
if not os.path.exists('classificationMap'):
os.makedirs('classificationMap')
_init_()
# load source domain data set
with open(os.path.join('datasets', 'Chikusei_imdb_128.pickle'),
'rb') as handle:
source_imdb = pickle.load(handle)
print(source_imdb.keys())
