iicd/autodl-projects
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update REA, REINFORCE, and RANDOM

Browse Source
This commit is contained in:
D-X-Y
2020-07-13 10:04:52 +00:00
parent 041a9aa4a3
commit 6dc494be08
12 changed files with 277 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

91
exps/algos-v2/random_wo_share.py Normal file
View File

@@ -0,0 +1,91 @@
##################################################
# Copyright (c) Xuanyi Dong [GitHub D-X-Y], 2020 #
##############################################################################
# Random Search for Hyper-Parameter Optimization, JMLR 2012 ##################
##############################################################################
# python ./exps/algos-v2/random_wo_share.py --dataset cifar10 --search_space tss
##############################################################################
import os, sys, time, glob, random, argparse
import numpy as np, collections
from copy import deepcopy
import torch
import torch.nn as nn
from pathlib import Path
lib_dir = (Path(__file__).parent / '..' / '..' / 'lib').resolve()
if str(lib_dir) not in sys.path: sys.path.insert(0, str(lib_dir))
from config_utils import load_config, dict2config, configure2str
from datasets import get_datasets, SearchDataset
from procedures import prepare_seed, prepare_logger, save_checkpoint, copy_checkpoint, get_optim_scheduler
from utils import get_model_infos, obtain_accuracy
from log_utils import AverageMeter, time_string, convert_secs2time
from models import get_search_spaces
from nas_201_api import NASBench201API, NASBench301API
from .regularized_ea import random_topology_func, random_size_func
def main(xargs, api):
torch.set_num_threads(4)
prepare_seed(xargs.rand_seed)
logger = prepare_logger(args)
search_space = get_search_spaces(xargs.search_space, 'nas-bench-301')
if xargs.search_space == 'tss':
random_arch = random_topology_func(search_space)
else:
random_arch = random_size_func(search_space)
x_start_time = time.time()
logger.log('{:} use nas_bench : {:}'.format(time_string(), nas_bench))
best_arch, best_acc, total_time_cost, history = None, -1, [], []
while total_time_cost[-1] < xargs.time_budget:
arch = random_arch()
accuracy, _, _, total_cost = api.simulate_train_eval(arch, xargs.dataset, '12')
total_time_cost.append(total_cost)
history.append(arch)
if best_arch is None or best_acc < accuracy:
best_acc, best_arch = accuracy, arch
logger.log('[{:03d}] : {:} : accuracy = {:.2f}%'.format(len(history), arch, accuracy))
logger.log('{:} best arch is {:}, accuracy = {:.2f}%, visit {:} archs with {:.1f} s (real-cost = {:.3f} s).'.format(time_string(), best_arch, best_acc, len(history), total_time_cost, time.time()-x_start_time))
info = api.query_info_str_by_arch(best_arch, '200' if xargs.search_space == 'tss' else '90')
logger.log('{:}'.format(info))
logger.log('-'*100)
logger.close()
return logger.log_dir, total_time_cost, history
if __name__ == '__main__':
parser = argparse.ArgumentParser("Random NAS")
parser.add_argument('--dataset', type=str, choices=['cifar10', 'cifar100', 'ImageNet16-120'], help='Choose between Cifar10/100 and ImageNet-16.')
parser.add_argument('--search_space', type=str, choices=['tss', 'sss'], help='Choose the search space.')
parser.add_argument('--time_budget', type=int, default=20000, help='The total time cost budge for searching (in seconds).')
parser.add_argument('--loops_if_rand', type=int, default=500, help='The total runs for evaluation.')
# log
parser.add_argument('--save_dir', type=str, help='Folder to save checkpoints and log.')
parser.add_argument('--rand_seed', type=int, default=-1, help='manual seed')
args = parser.parse_args()
if args.search_space == 'tss':
api = NASBench201API(verbose=False)
elif args.search_space == 'sss':
api = NASBench301API(verbose=False)
else:
raise ValueError('Invalid search space : {:}'.format(args.search_space))
args.save_dir = os.path.join('{:}-{:}'.format(args.save_dir, args.search_space), args.dataset, 'RANDOM')
print('save-dir : {:}'.format(args.save_dir))
if args.rand_seed < 0:
save_dir, all_info = None, {}
for i in range(args.loops_if_rand):
print ('{:} : {:03d}/{:03d}'.format(time_string(), i, args.loops_if_rand))
args.rand_seed = random.randint(1, 100000)
save_dir, all_archs, all_total_times = main(args, api)
all_info[i] = {'all_archs': all_archs,
'all_total_times': all_total_times}
save_path = save_dir / 'results.pth'
print('save into {:}'.format(save_path))
torch.save(all_info, save_path)
else:
main(args, api)

29
exps/algos-v2/REA.py → exps/algos-v2/regularized_ea.py
View File

@@ -3,12 +3,12 @@
##################################################################
# Regularized Evolution for Image Classifier Architecture Search #
##################################################################
# python ./exps/algos-v2/REA.py --dataset cifar10 --search_space tss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/REA.py --dataset cifar100 --search_space tss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/REA.py --dataset ImageNet16-120 --search_space tss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/REA.py --dataset cifar10 --search_space sss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/REA.py --dataset cifar100 --search_space sss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/REA.py --dataset ImageNet16-120 --search_space sss --time_budget 12000 --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset cifar10 --search_space tss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset cifar100 --search_space tss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset ImageNet16-120 --search_space tss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset cifar10 --search_space sss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset cifar100 --search_space sss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
# python ./exps/algos-v2/regularized_ea.py --dataset ImageNet16-120 --search_space sss --ea_cycles 200 --ea_population 10 --ea_sample_size 3 --rand_seed 1
##################################################################
import os, sys, time, glob, random, argparse
import numpy as np, collections
@@ -160,7 +160,7 @@ def regularized_evolution(cycles, population_size, sample_size, time_budget, ran
while len(population) < population_size:
model = Model()
model.arch = random_arch()
model.accuracy, time_cost, total_cost = api.simulate_train_eval(model.arch, dataset, '12')
model.accuracy, _, _, total_cost = api.simulate_train_eval(model.arch, dataset, '12')
# Append the info
population.append(model)
history.append(model)
@@ -183,7 +183,7 @@ def regularized_evolution(cycles, population_size, sample_size, time_budget, ran
# Create the child model and store it.
child = Model()
child.arch = mutate_arch(parent.arch)
child.accuracy, time_cost, total_cost = api.simulate_train_eval(model.arch, dataset, '12')
child.accuracy, _, _, total_cost = api.simulate_train_eval(model.arch, dataset, '12')
# Append the info
population.append(child)
history.append(child)
@@ -195,11 +195,7 @@ def regularized_evolution(cycles, population_size, sample_size, time_budget, ran
def main(xargs, api):
assert torch.cuda.is_available(), 'CUDA is not available.'
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
torch.set_num_threads(xargs.workers)
torch.set_num_threads(4)
prepare_seed(xargs.rand_seed)
logger = prepare_logger(args)
@@ -235,12 +231,11 @@ if __name__ == '__main__':
parser.add_argument('--ea_cycles', type=int, help='The number of cycles in EA.')
parser.add_argument('--ea_population', type=int, help='The population size in EA.')
parser.add_argument('--ea_sample_size', type=int, help='The sample size in EA.')
parser.add_argument('--time_budget', type=int, help='The total time cost budge for searching (in seconds).')
parser.add_argument('--loops_if_rand', type=int, default=500, help='The total runs for evaluation.')
parser.add_argument('--time_budget', type=int, default=20000, help='The total time cost budge for searching (in seconds).')
parser.add_argument('--loops_if_rand', type=int, default=500, help='The total runs for evaluation.')
# log
parser.add_argument('--workers', type=int, default=2, help='number of data loading workers (default: 2)')
parser.add_argument('--save_dir', type=str, default='./output/search', help='Folder to save checkpoints and log.')
parser.add_argument('--rand_seed', type=int, default=-1, help='manual seed')
parser.add_argument('--rand_seed', type=int, default=-1, help='manual seed')
args = parser.parse_args()
if args.search_space == 'tss':

33
exps/algos-v2/reinforce.py
View File

@@ -3,12 +3,12 @@
#####################################################################################################
# modified from https://github.com/pytorch/examples/blob/master/reinforcement_learning/reinforce.py #
#####################################################################################################
# python ./exps/algos-v2/reinforce.py --dataset cifar10 --search_space tss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar100 --search_space tss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset ImageNet16-120 --search_space tss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar10 --search_space sss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar100 --search_space sss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset ImageNet16-120 --search_space sss --time_budget 12000 --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar10 --search_space tss --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar100 --search_space tss --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset ImageNet16-120 --search_space tss --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar10 --search_space sss --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset cifar100 --search_space sss --learning_rate 0.001
# python ./exps/algos-v2/reinforce.py --dataset ImageNet16-120 --search_space sss --learning_rate 0.001
#####################################################################################################
import os, sys, time, glob, random, argparse
import numpy as np, collections
@@ -120,15 +120,10 @@ def select_action(policy):
def main(xargs, api):
assert torch.cuda.is_available(), 'CUDA is not available.'
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
torch.set_num_threads(xargs.workers)
torch.set_num_threads(4)
prepare_seed(xargs.rand_seed)
logger = prepare_logger(args)
search_space = get_search_spaces(xargs.search_space, 'nas-bench-301')
if xargs.search_space == 'tss':
policy = PolicyTopology(search_space)
@@ -144,6 +139,7 @@ def main(xargs, api):
# nas dataset load
logger.log('{:} use api : {:}'.format(time_string(), api))
api.reset_time()
# REINFORCE
x_start_time = time.time()
@@ -153,7 +149,7 @@ def main(xargs, api):
start_time = time.time()
log_prob, action = select_action( policy )
arch = policy.generate_arch( action )
reward, _, current_total_cost = api.simulate_train_eval(arch, xargs.dataset, '12')
reward, _, _, current_total_cost = api.simulate_train_eval(arch, xargs.dataset, '12')
trace.append((reward, arch))
total_costs.append(current_total_cost)
@@ -177,7 +173,7 @@ def main(xargs, api):
logger.log('-'*100)
logger.close()
return logger.log_dir, [api.query_index_by_arch(x[0]) for x in trace], total_costs
return logger.log_dir, [api.query_index_by_arch(x[1]) for x in trace], total_costs
if __name__ == '__main__':
@@ -186,15 +182,14 @@ if __name__ == '__main__':
parser.add_argument('--dataset', type=str, choices=['cifar10', 'cifar100', 'ImageNet16-120'], help='Choose between Cifar10/100 and ImageNet-16.')
parser.add_argument('--search_space', type=str, choices=['tss', 'sss'], help='Choose the search space.')
parser.add_argument('--learning_rate', type=float, help='The learning rate for REINFORCE.')
parser.add_argument('--EMA_momentum', type=float, default=0.9, help='The momentum value for EMA.')
parser.add_argument('--time_budget', type=int, help='The total time cost budge for searching (in seconds).')
parser.add_argument('--loops_if_rand', type=int, default=500, help='The total runs for evaluation.')
parser.add_argument('--EMA_momentum', type=float, default=0.9, help='The momentum value for EMA.')
parser.add_argument('--time_budget', type=int, default=20000, help='The total time cost budge for searching (in seconds).')
parser.add_argument('--loops_if_rand', type=int, default=500, help='The total runs for evaluation.')
# log
parser.add_argument('--workers', type=int, default=2, help='number of data loading workers (default: 2)')
parser.add_argument('--save_dir', type=str, default='./output/search', help='Folder to save checkpoints and log.')
parser.add_argument('--arch_nas_dataset', type=str, help='The path to load the architecture dataset (tiny-nas-benchmark).')
parser.add_argument('--print_freq', type=int, help='print frequency (default: 200)')
parser.add_argument('--rand_seed', type=int, default=-1, help='manual seed')
parser.add_argument('--rand_seed', type=int, default=-1, help='manual seed')
args = parser.parse_args()
if args.search_space == 'tss':

17
exps/algos-v2/run-all.sh Normal file
View File

@@ -0,0 +1,17 @@
#!/bin/bash
# bash ./exps/algos-v2/run-all.sh
echo script name: $0
echo $# arguments
datasets="cifar10 cifar100 ImageNet16-120"
search_spaces="tss sss"
for dataset in ${datasets}
do
for search_space in ${search_spaces}
do
python ./exps/algos-v2/reinforce.py --dataset ${dataset} --search_space ${search_space} --learning_rate 0.001
python ./exps/algos-v2/regularized_ea.py --dataset ${dataset} --search_space ${search_space} --ea_cycles 200 --ea_population 10 --ea_sample_size 3
done
done