import argparse
import glob
import json
import logging
import os
import pickle
import sys
import time
import numpy as np
import torch
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.utils
import torchvision.datasets as dset
from torch.autograd import Variable
from nasbench_analysis import eval_darts_one_shot_model_in_nasbench as naseval
from nasbench_analysis.search_spaces.search_space_1 import SearchSpace1
from nasbench_analysis.search_spaces.search_space_2 import SearchSpace2
from nasbench_analysis.search_spaces.search_space_3 import SearchSpace3
from nasbench_analysis.single_architecture_training.model_search import NetworkIndependent as Network
from optimizers.darts import utils
from optimizers.darts.architect import Architect
from optimizers.darts.genotypes import PRIMITIVES
parser = argparse.ArgumentParser("cifar")
parser.add_argument('--data', type=str, default='../data', help='location of the darts corpus')
parser.add_argument('--batch_size', type=int, default=256, help='batch size')
parser.add_argument('--learning_rate', type=float, default=0.2, help='init learning rate') # Increased learning rate
parser.add_argument('--learning_rate_min', type=float, default=0, help='min learning rate')
parser.add_argument('--momentum', type=float, default=0.9, help='momentum')
parser.add_argument('--weight_decay', type=float, default=1e-4, help='weight decay')
parser.add_argument('--report_freq', type=float, default=50, help='report frequency')
parser.add_argument('--gpu', type=int, default=0, help='gpu device id')
parser.add_argument('--epochs', type=int, default=108, help='num of training epochs')
parser.add_argument('--init_channels', type=int, default=16, help='num of init channels')
parser.add_argument('--layers', type=int, default=9, help='total number of layers')
parser.add_argument('--model_path', type=str, default='saved_models', help='path to save the model')
parser.add_argument('--cutout', action='store_true', default=False, help='use cutout')
parser.add_argument('--cutout_length', type=int, default=16, help='cutout length')
parser.add_argument('--cutout_prob', type=float, default=1.0, help='cutout probability')
parser.add_argument('--drop_path_prob', type=float, default=0.3, help='drop path probability')
parser.add_argument('--save', type=str, default='EXP', help='experiment name')
parser.add_argument('--seed', type=int, default=2, help='random_ws seed')
parser.add_argument('--grad_clip', type=float, default=5, help='gradient clipping')
parser.add_argument('--train_portion', type=float, default=0.5, help='portion of training darts')
parser.add_argument('--unrolled', action='store_true', default=False, help='use one-step unrolled validation loss')
parser.add_argument('--arch_learning_rate', type=float, default=3e-4, help='learning rate for arch encoding')
parser.add_argument('--arch_weight_decay', type=float, default=1e-3, help='weight decay for arch encoding')
parser.add_argument('--output_weights', type=bool, default=True, help='Whether to use weights on the output nodes')
parser.add_argument('--search_space', choices=['1', '2', '3'], default='1')
parser.add_argument('--debug', action='store_true', default=False, help='run only for some batches')
parser.add_argument('--warm_start_epochs', type=int, default=0,
help='Warm start one-shot model before starting architecture updates.')
args = parser.parse_args()
args.save = 'experiments_2/independent_training_nasbench/search_space_{}/search-{}-{}-{}-{}'.format(
args.search_space,
args.save,
time.strftime(
"%Y%m%d-%H%M%S"),
args.seed,
args.search_space)
utils.create_exp_dir(args.save, scripts_to_save=glob.glob('*.py'))
# Dump the config of the run
with open(os.path.join(args.save, 'config.json'), 'w') as fp:
json.dump(args.__dict__, fp)
for handler in logging.root.handlers[:]:
logging.root.removeHandler(handler)
log_format = '%(asctime)s %(message)s'
logging.basicConfig(stream=sys.stdout, level=logging.INFO,
format=log_format, datefmt='%m/%d %I:%M:%S %p')
fh = logging.FileHandler(os.path.join(args.save, 'log.txt'))
fh.setFormatter(logging.Formatter(log_format))
logging.getLogger().addHandler(fh)
CIFAR_CLASSES = 10
def main():
# Select the search space to search in
if args.search_space == '1':
search_space = SearchSpace1()
elif args.search_space == '2':
search_space = SearchSpace2()
elif args.search_space == '3':
search_space = SearchSpace3()
else:
raise ValueError('Unknown search space')
if not torch.cuda.is_available():
logging.info('no gpu device available')
sys.exit(1)
np.random.seed(args.seed)
torch.cuda.set_device(args.gpu)
cudnn.benchmark = True
torch.manual_seed(args.seed)
cudnn.enabled = True
torch.cuda.manual_seed(args.seed)
logging.info('gpu device = %d' % args.gpu)
logging.info("args = %s", args)
criterion = nn.CrossEntropyLoss()
criterion = criterion.cuda()
model = Network(args.init_channels, CIFAR_CLASSES, args.layers, criterion, output_weights=args.output_weights,
steps=search_space.num_intermediate_nodes, search_space=search_space)
model = model.cuda()
logging.info("param size = %fMB", utils.count_parameters_in_MB(model))
optimizer = torch.optim.RMSprop(
model.parameters(),
args.learning_rate,
momentum=args.momentum,
weight_decay=args.weight_decay)
train_transform, valid_transform = utils._data_transforms_cifar10(args)
train_data = dset.CIFAR10(root=args.data, train=True, download=True, transform=train_transform)
num_train = len(train_data)
indices = list(range(num_train))
split = int(np.floor(args.train_portion * num_train))
train_queue = torch.utils.data.DataLoader(
train_data, batch_size=args.batch_size, num_workers=3,
sampler=torch.utils.data.sampler.SubsetRandomSampler(indices[:split]),
pin_memory=True)
valid_queue = torch.utils.data.DataLoader(
train_data, batch_size=args.batch_size, num_workers=3,
sampler=torch.utils.data.sampler.SubsetRandomSampler(indices[split:num_train]),
pin_memory=True)
scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer, float(args.epochs), eta_min=args.learning_rate_min)
architect = Architect(model, args)
arch = search_space.sample(with_loose_ends=False, upscale=False)
arch_parameters = get_weights_from_arch(arch, model)
model._arch_parameters = arch_parameters
try:
for epoch in range(args.epochs):
scheduler.step()
lr = scheduler.get_lr()[0]
# increase the cutout probability linearly throughout search
train_transform.transforms[-1].cutout_prob = args.cutout_prob * epoch / (args.epochs - 1)
logging.info('epoch %d lr %e cutout_prob %e', epoch, lr,
train_transform.transforms[-1].cutout_prob)
# Save the one shot model architecture weights for later analysis
arch_filename = os.path.join(args.save, 'one_shot_architecture_{}.obj'.format(epoch))
with open(arch_filename, 'wb') as filehandler:
numpy_tensor_list = []
for tensor in model.arch_parameters():
numpy_tensor_list.append(tensor.detach().cpu().numpy())
pickle.dump(numpy_tensor_list, filehandler)
# Save the entire one-shot-model
# filepath = os.path.join(args.save, 'one_shot_model_{}.obj'.format(epoch))
# torch.save(model.state_dict(), filepath)
logging.info('architecture', numpy_tensor_list)
# training
train_acc, train_obj = train(train_queue, valid_queue, model, architect, criterion, optimizer, lr, epoch)
logging.info('train_acc %f', train_acc)
# validation
valid_acc, valid_obj = infer(valid_queue, model, criterion)
logging.info('valid_acc %f', valid_acc)
utils.save(model, os.path.join(args.save, 'weights.pt'))
logging.info('STARTING EVALUATION')
test, valid, runtime, params = naseval.eval_one_shot_model(config=args.__dict__,
model=arch_filename)
index = np.random.choice(list(range(3)))
logging.info('TEST ERROR: %.3f | VALID ERROR: %.3f | RUNTIME: %f | PARAMS: %d'
% (test[index],
valid[index],
runtime[index],
params[index])
)
except Exception as e:
logging.exception('message')
def get_weights_from_arch(arch, model):
adjacency_matrix, node_list = arch
num_ops = len(PRIMITIVES)
# Assign the sampled ops to the mixed op weights.
# These are not optimized
alphas_mixed_op = Variable(torch.zeros(model._steps, num_ops).cuda(), requires_grad=False)
for idx, op in enumerate(node_list):
alphas_mixed_op[idx][PRIMITIVES.index(op)] = 1
# Set the output weights
alphas_output = Variable(torch.zeros(1, model._steps + 1).cuda(), requires_grad=False)
for idx, label in enumerate(list(adjacency_matrix[:, -1][:-1])):
alphas_output[0][idx] = label
# Initialize the weights for the inputs to each choice block.
if type(model.search_space) == SearchSpace1:
begin = 3
else:
begin = 2
alphas_inputs = [Variable(torch.zeros(1, n_inputs).cuda(), requires_grad=False) for n_inputs in
range(begin, model._steps + 1)]
for alpha_input in alphas_inputs:
connectivity_pattern = list(adjacency_matrix[:alpha_input.shape[1], alpha_input.shape[1]])
for idx, label in enumerate(connectivity_pattern):
alpha_input[0][idx] = label
# Total architecture parameters
arch_parameters = [
alphas_mixed_op,
alphas_output,
*alphas_inputs
]
return arch_parameters
def train(train_queue, valid_queue, model, architect, criterion, optimizer, lr, epoch):
objs = utils.AvgrageMeter()
top1 = utils.AvgrageMeter()
top5 = utils.AvgrageMeter()
for step, (input, target) in enumerate(train_queue):
model.train()
n = input.size(0)
input = input.cuda()
target = target.cuda(non_blocking=True)
# get a minibatch from the search queue with replacement
try:
input_search, target_search = next(valid_queue_iter)
except:
valid_queue_iter = iter(valid_queue)
input_search, target_search = next(valid_queue_iter)
optimizer.zero_grad()
logits = model(input, discrete=True)
loss = criterion(logits, target)
loss.backward()
nn.utils.clip_grad_norm_(model.parameters(), args.grad_clip)
optimizer.step()
prec1, prec5 = utils.accuracy(logits, target, topk=(1, 5))
objs.update(loss.data.item(), n)
top1.update(prec1.data.item(), n)
top5.update(prec5.data.item(), n)
if step % args.report_freq == 0:
logging.info('train %03d %e %f %f', step, objs.avg, top1.avg, top5.avg)
if args.debug:
break
return top1.avg, objs.avg
def infer(valid_queue, model, criterion):
objs = utils.AvgrageMeter()
top1 = utils.AvgrageMeter()
top5 = utils.AvgrageMeter()
model.eval()
for step, (input, target) in enumerate(valid_queue):
input = input.cuda()
target = target.cuda(non_blocking=True)
logits = model(input, discrete=True)
loss = criterion(logits, target)
prec1, prec5 = utils.accuracy(logits, target, topk=(1, 5))
n = input.size(0)
objs.update(loss.data.item(), n)
top1.update(prec1.data.item(), n)
top5.update(prec5.data.item(), n)
if step % args.report_freq == 0:
logging.info('valid %03d %e %f %f', step, objs.avg, top1.avg, top5.avg)
if args.debug:
break
return top1.avg, objs.avg
if __name__ == '__main__':
main()
