Skip to content
Branch: master
Find file Copy path
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
624 lines (511 sloc) 24 KB
# coding: utf-8
# ### This code extends the functionality of to support cross-validation training, allowing you compute the out of sample predicted probabilities for the entire imagenet training set: a necessary step for confident learning and the cleanlab package.
# Here is an example of how to use this file:
# ```bash
# # Four fold cross-validation training.
# $ python3 -a resnet18 -b 256 --lr 0.1 --gpu 0 --cvn 4 --cv 0 /IMAGENET_PATH
# $ python3 -a resnet18 -b 256 --lr 0.1 --gpu 1 --cvn 4 --cv 1 /IMAGENET_PATH
# $ python3 -a resnet18 -b 256 --lr 0.1 --gpu 2 --cvn 4 --cv 2 /IMAGENET_PATH
# $ python3 -a resnet18 -b 256 --lr 0.1 --gpu 3 --cvn 4 --cv 3 /IMAGENET_PATH
# # Combine the results
# $ python3 -a resnet18 --cvn 4 --combine-folds /IMAGENET_PATH
# ```
# In[ ]:
# These imports enhance Python2/3 compatibility.
from __future__ import print_function, absolute_import, division, unicode_literals, with_statement
# In[18]:
import argparse
import os
import random
import shutil
import time
import warnings
import sys
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.optim
import torch.multiprocessing as mp
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torchvision.models as models
from sklearn.model_selection import StratifiedKFold
import copy
import numpy as np
# In[ ]:
num_classes = 1000
# In[19]:
model_names = sorted(name for name in models.__dict__
if name.islower() and not name.startswith("__")
and callable(models.__dict__[name]))
# In[20]:
parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('data', metavar='DIR',
help='path to dataset')
parser.add_argument('-a', '--arch', metavar='ARCH', default='resnet18',
help='model architecture: ' +
' | '.join(model_names) +
' (default: resnet18)')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
help='number of data loading workers (default: 4)')
parser.add_argument('--epochs', default=90, type=int, metavar='N',
help='number of total epochs to run')
parser.add_argument('--start-epoch', default=0, type=int, metavar='N',
help='manual epoch number (useful on restarts)')
parser.add_argument('-b', '--batch-size', default=256, type=int,
help='mini-batch size (default: 256), this is the total '
'batch size of all GPUs on the current node when '
'using Data Parallel or Distributed Data Parallel')
parser.add_argument('--lr', '--learning-rate', default=0.1, type=float,
metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
parser.add_argument('--wd', '--weight-decay', default=1e-4, type=float,
metavar='W', help='weight decay (default: 1e-4)',
parser.add_argument('-p', '--print-freq', default=10, type=int,
metavar='N', help='print frequency (default: 10)')
parser.add_argument('--resume', default='', type=str, metavar='PATH',
help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true',
help='evaluate model on validation set')
parser.add_argument('--pretrained', dest='pretrained', action='store_true',
help='use pre-trained model')
parser.add_argument('--world-size', default=-1, type=int,
help='number of nodes for distributed training')
parser.add_argument('--rank', default=-1, type=int,
help='node rank for distributed training')
parser.add_argument('--dist-url', default='tcp://', type=str,
help='url used to set up distributed training')
parser.add_argument('--dist-backend', default='nccl', type=str,
help='distributed backend')
parser.add_argument('--seed', default=None, type=int,
help='seed for initializing training. ')
parser.add_argument('--cv-seed', default=0, type=int,
help='seed for determining the cv folds. ')
parser.add_argument('--gpu', default=None, type=int,
help='GPU id to use.')
parser.add_argument('--multiprocessing-distributed', action='store_true',
help='Use multi-processing distributed training to launch '
'N processes per node, which has N GPUs. This is the '
'fastest way to use PyTorch for either single node or '
'multi node data parallel training')
parser.add_argument('--cv', '--cv-fold', type=int, default = None,
metavar='N', help='The fold to holdout')
parser.add_argument('--cvn', '--cv-n-folds', default = 0, type=int,
metavar='N', help='The number of folds')
parser.add_argument('-m', '--dir-train-mask', default = None, type=str,
metavar='DIR', help='Boolean mask with True for indices to '
'train with and false for indices to skip.')
parser.add_argument('--combine-folds', action='store_true', default = False,
help='Pass this flag and -a arch to combine probs '
'from all folds. You must pass -a and -cvn flags as well!')
best_acc1 = 0
# In[6]:
def main(args = parser.parse_args()):
if args.seed is not None:
cudnn.deterministic = True
warnings.warn('You have chosen to seed training. '
'This will turn on the CUDNN deterministic setting, '
'which can slow down your training considerably! '
'You may see unexpected behavior when restarting '
'from checkpoints.')
if args.gpu is not None:
warnings.warn('You have chosen a specific GPU. This will completely '
'disable data parallelism.')
if args.dist_url == "env://" and args.world_size == -1:
args.world_size = int(os.environ["WORLD_SIZE"])
args.distributed = args.world_size > 1 or args.multiprocessing_distributed
ngpus_per_node = torch.cuda.device_count()
if args.multiprocessing_distributed:
# Since we have ngpus_per_node processes per node, the total world_size
# needs to be adjusted accordingly
args.world_size = ngpus_per_node * args.world_size
# Use torch.multiprocessing.spawn to launch distributed processes: the
# main_worker process function
mp.spawn(main_worker, nprocs=ngpus_per_node, args=(ngpus_per_node, args))
# Simply call main_worker function
main_worker(args.gpu, ngpus_per_node, args)
# In[ ]:
def main_worker(gpu, ngpus_per_node, args):
global best_acc1
args.gpu = gpu
val_size = 50000
use_crossval = args.cvn > 0
use_mask = args.dir_train_mask is not None
cv_fold =
cv_n_folds = args.cvn
class_weights = None
if use_crossval and use_mask:
raise ValueError('Either args.cvn > 0 or dir-train-mask not None, but not both.')
if args.gpu is not None:
print("Use GPU: {} for training".format(args.gpu))
if args.distributed:
if args.dist_url == "env://" and args.rank == -1:
args.rank = int(os.environ["RANK"])
if args.multiprocessing_distributed:
# For multiprocessing distributed training, rank needs to be the
# global rank among all the processes
args.rank = args.rank * ngpus_per_node + gpu
dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
world_size=args.world_size, rank=args.rank)
# create model
if args.pretrained:
print("=> using pre-trained model '{}'".format(args.arch))
model = models.__dict__[args.arch](pretrained=True)
print("=> creating model '{}'".format(args.arch))
model = models.__dict__[args.arch](num_classes=num_classes)
if args.distributed:
# For multiprocessing distributed, DistributedDataParallel constructor
# should always set the single device scope, otherwise,
# DistributedDataParallel will use all available devices.
if args.gpu is not None:
# When using a single GPU per process and per
# DistributedDataParallel, we need to divide the batch size
# ourselves based on the total number of GPUs we have
args.batch_size = int(args.batch_size / ngpus_per_node)
args.workers = int(args.workers / ngpus_per_node)
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
# DistributedDataParallel will divide and allocate batch_size to all
# available GPUs if device_ids are not set
model = torch.nn.parallel.DistributedDataParallel(model)
elif args.gpu is not None:
model = model.cuda(args.gpu)
# DataParallel will divide and allocate batch_size to all available GPUs
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model = torch.nn.DataParallel(model).cuda()
# define optimizer
optimizer = torch.optim.SGD(model.parameters(),,
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
if args.gpu is None:
checkpoint = torch.load(args.resume)
# Map model to be loaded to specified single gpu.
loc = 'cuda:{}'.format(args.gpu)
checkpoint = torch.load(args.resume, map_location=loc)
args.start_epoch = checkpoint['epoch']
best_acc1 = checkpoint['best_acc1']
if args.gpu is not None:
# In case you load checkpoint from different GPU
best_acc1 =
print("=> loaded checkpoint '{}' (epoch {})"
.format(args.resume, checkpoint['epoch']))
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(, 'train')
valdir = os.path.join(, 'val')
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_dataset = datasets.ImageFolder(
# If training only on a cross-validated portion & make val_set = train_holdout.
if use_crossval:
checkpoint_fn = "model_{}__fold_{}__checkpoint.pth.tar".format(args.arch, cv_fold)
print('Computing fold indices. This takes 15 seconds.')
# Prepare labels
labels = [label for img, label in datasets.ImageFolder(traindir).imgs]
# Split train into train and holdout for particular cv_fold.
kf = StratifiedKFold(n_splits = cv_n_folds, shuffle = True, random_state = args.cv_seed)
cv_train_idx, cv_holdout_idx = list(kf.split(range(len(labels)), labels))[cv_fold]
# Seperate datasets
holdout_dataset = copy.deepcopy(train_dataset)
holdout_dataset.imgs = [train_dataset.imgs[i] for i in cv_holdout_idx]
holdout_dataset.samples = holdout_dataset.imgs
# Subset of holdout used to choose the best model.
val_dataset = copy.deepcopy(holdout_dataset)
val_imgs_idx = np.random.choice(range(len(holdout_dataset.imgs)), size=val_size, replace=False,)
val_dataset.imgs = [holdout_dataset.imgs[i] for i in val_imgs_idx]
val_dataset.samples = val_dataset.imgs
train_dataset.imgs = [train_dataset.imgs[i] for i in cv_train_idx]
train_dataset.samples = train_dataset.imgs
print('Train size:', len(cv_train_idx), len(train_dataset.imgs))
print('Holdout size:', len(cv_holdout_idx), len(holdout_dataset.imgs))
print('Val size (subset of holdout):', len(val_imgs_idx), len(val_dataset.imgs))
checkpoint_fn = "model_{}__checkpoint.pth.tar".format(args.arch)
if use_mask:
checkpoint_fn = "model_{}__masked__checkpoint.pth.tar".format(args.arch)
orig_class_counts = np.bincount(
[lab for img, lab in datasets.ImageFolder(traindir).imgs])
train_bool_mask = np.load(args.dir_train_mask)
# Mask labels
train_dataset.imgs = [img for i, img in enumerate(train_dataset.imgs) if train_bool_mask[i]]
train_dataset.samples = train_dataset.imgs
clean_class_counts = np.bincount(
[lab for img, lab in train_dataset.imgs])
print('Train size:', len(train_dataset.imgs))
# Compute class weights to re-weight loss during training
# Should use the confident joint to estimate the noise matrix then
# class_weights = 1 / p(s=k, y=k) for each class k.
# Here we approximate this with a simpler approach
# class_weights = count(y=k) / count(s=k, y=k)
class_weights = torch.Tensor(orig_class_counts / clean_class_counts)
val_dataset = datasets.ImageFolder(
if args.distributed:
train_sampler =
train_sampler = None
train_loader =
train_dataset, batch_size=args.batch_size, shuffle=(train_sampler is None),
num_workers=args.workers, pin_memory=True, sampler=train_sampler,
val_loader =
batch_size=args.batch_size, shuffle=False,
num_workers=args.workers, pin_memory=True,
# define loss function (criterion)
criterion = nn.CrossEntropyLoss(weight=class_weights).cuda(args.gpu)
if args.evaluate:
validate(val_loader, model, criterion, args)
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
adjust_learning_rate(optimizer, epoch, args)
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch, args)
# evaluate on validation set
acc1 = validate(val_loader, model, criterion, args)
# remember best acc@1, model, and save checkpoint
is_best = acc1 > best_acc1
best_acc1 = max(best_acc1, acc1)
if not args.multiprocessing_distributed or (args.multiprocessing_distributed
and args.rank % ngpus_per_node == 0):
'epoch': epoch + 1,
'arch': args.arch,
'state_dict': model.state_dict(),
'best_acc1': best_acc1,
'optimizer' : optimizer.state_dict(),
is_best = is_best,
filename = checkpoint_fn,
cv_fold = cv_fold,
use_mask = use_mask,
if use_crossval:
holdout_loader =
batch_size=args.batch_size, shuffle=False,
num_workers=args.workers, pin_memory=True,
print("=> loading best model_{}__fold_{}_best.pth.tar".format(args.arch, cv_fold))
checkpoint = torch.load("model_{}__fold_{}_best.pth.tar".format(args.arch, cv_fold))
print("Running forward pass on holdout set of size:", len(holdout_dataset.imgs))
probs = get_probs(holdout_loader, model, args)'model_{}__fold_{}__probs.npy'.format(args.arch, cv_fold), probs)
# In[7]:
def train(train_loader, model, criterion, optimizer, epoch, args):
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to train mode
end = time.time()
for i, (input, target) in enumerate(train_loader):
# measure data loading time
data_time.update(time.time() - end)
if args.gpu is not None:
input = input.cuda(args.gpu, non_blocking=True)
target = target.cuda(args.gpu, non_blocking=True)
# compute output
output = model(input)
loss = criterion(output, target)
# measure accuracy and record loss
acc1, acc5 = accuracy(output, target, topk=(1, 5))
losses.update(loss.item(), input.size(0))
top1.update(acc1[0], input.size(0))
top5.update(acc5[0], input.size(0))
# compute gradient and do SGD step
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0:
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Acc@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Acc@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
epoch, i, len(train_loader), batch_time=batch_time,
data_time=data_time, loss=losses, top1=top1, top5=top5))
# In[69]:
def get_probs(loader, model, args):
# switch to evaluate mode
ntotal = len(loader.dataset.imgs) / float(loader.batch_size)
outputs = []
with torch.no_grad():
end = time.time()
for i, (input, target) in enumerate(loader):
print("\rComplete: {:.1%}".format(i / ntotal), end = "")
if args.gpu is not None:
input = input.cuda(args.gpu, non_blocking=True)
target = target.cuda(args.gpu, non_blocking=True)
# compute output
# Prepare outputs as a single matrix
probs = np.concatenate([
torch.nn.functional.softmax(z, dim = 1) if args.gpu is None else
torch.nn.functional.softmax(z, dim = 1).cpu().numpy()
for z in outputs
return probs
# In[8]:
def validate(val_loader, model, criterion, args):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
with torch.no_grad():
end = time.time()
for i, (input, target) in enumerate(val_loader):
if args.gpu is not None:
input = input.cuda(args.gpu, non_blocking=True)
target = target.cuda(args.gpu, non_blocking=True)
# compute output
output = model(input)
loss = criterion(output, target)
# measure accuracy and record loss
acc1, acc5 = accuracy(output, target, topk=(1, 5))
losses.update(loss.item(), input.size(0))
top1.update(acc1[0], input.size(0))
top5.update(acc5[0], input.size(0))
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0:
print('Test: [{0}/{1}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Acc@1 {top1.val:.3f} ({top1.avg:.3f})\t'
'Acc@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
i, len(val_loader), batch_time=batch_time, loss=losses,
top1=top1, top5=top5))
print(' * Acc@1 {top1.avg:.3f} Acc@5 {top5.avg:.3f}'
.format(top1=top1, top5=top5))
return top1.avg
def save_checkpoint(state, is_best, filename='checkpoint.pth.tar', cv_fold = None, use_mask = False):, filename)
if is_best:
sm = "__masked" if use_mask else ""
sf = "__fold_{}".format(cv_fold) if cv_fold is not None else ""
wfn = 'model_{}{}{}_best.pth.tar'.format(state['arch'], sm, sf)
shutil.copyfile(filename, wfn)
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self):
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def adjust_learning_rate(optimizer, epoch, args):
"""Sets the learning rate to the initial LR decayed by 10 every 30 epochs"""
lr = * (0.1 ** (epoch // 30))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def accuracy(output, target, topk=(1,)):
"""Computes the accuracy over the k top predictions for the specified values of k"""
with torch.no_grad():
maxk = max(topk)
batch_size = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
pred = pred.t()
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batch_size))
return res
# In[33]:
def combine_folds(args):
wfn = 'imagenet__train__model_{}__pyx.npy'.format(args.arch)
print('Make sure you specified the model architecture with flag -a.')
print('This method will overwrite file: {}'.format(wfn))
print('Computing fold indices. This takes 15 seconds.')
# Prepare labels
labels = [label for img, label in datasets.ImageFolder(os.path.join(, "train/")).imgs]
num_classes = 1000
# Intialize pyx array (output of trained network)
pyx = np.empty((len(labels), num_classes))
# Split train into train and holdout for each cv_fold.
kf = StratifiedKFold(n_splits = args.cvn, shuffle = True, random_state = args.cv_seed)
for k, (cv_train_idx, cv_holdout_idx) in enumerate(kf.split(range(len(labels)), labels)):
probs = np.load('model_{}__fold_{}__probs.npy'.format(args.arch, k))
pyx[cv_holdout_idx] = probs[:, :num_classes]
print('Writing final predicted probabilities.'), pyx)
# Compute overall accuracy
print('Computing Accuracy.', flush=True)
acc = sum(np.array(labels) == np.argmax(pyx, axis = 1)) / float(len(labels))
print('Accuracy: {:.25}'.format(acc))
# In[ ]:
if __name__ == '__main__':
args = parser.parse_args()
if args.combine_folds:
You can’t perform that action at this time.