train.py

#!/usr/bin/env python

import argparse
import builtins
import math
import os
import random
import shutil
import time
import warnings
import json
import numpy as np

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 torch.utils.data
import torch.utils.data.distributed

from tensorboard_logger import configure, log_value

from utils.cond_encoder import Encoder
from utils.gcn import GraphNet

import model.builder_patch
import model.builder_graph
import model.loader
from monai.transforms import Compose, RandGaussianNoise, Rand3DElastic, RandAdjustContrast

from dataset_train_patch import COPD_dataset as COPD_dataset_graph
from dataset_train_graph import COPD_dataset as COPD_dataset_patch

parser = argparse.ArgumentParser(description='PyTorch ImageNet Training')
parser.add_argument('--arch', metavar='ARCH', default='custom')
parser.add_argument('--workers-patch', default=32, type=int, metavar='N',
                    help='number of data loading workers (default: 32)')
parser.add_argument('--workers-graph', default=32, type=int, metavar='N',
                    help='number of data loading workers (default: 32)')
parser.add_argument('--epochs', default=200, type=int, metavar='N',
                    help='number of total epochs to run')
parser.add_argument('--batch-size-patch', default=4, type=int,
                    metavar='N',
                    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('--batch-size-graph', default=4, type=int,
                    metavar='N',
                    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=0.03, type=float,
                    metavar='LR', help='initial learning rate', dest='lr')
parser.add_argument('--start-epoch', default=0, type=int, metavar='N',
                    help='manual epoch number (useful on restarts)')
parser.add_argument('--schedule', default=[120, 160], nargs='*', type=int,
                    help='learning rate schedule (when to drop lr by 10x)')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
                    help='momentum of SGD solver')
parser.add_argument('--weight-decay', default=1e-4, type=float,
                    metavar='W', help='weight decay (default: 1e-4)',
                    dest='weight_decay')
parser.add_argument('--print-freq', default=100, 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('--resume-graph', default='', type=str, metavar='PATH',
                    help='path to latest checkpoint (default: none)')
parser.add_argument('--world-size', default=1, type=int,
                    help='number of nodes for distributed training')
parser.add_argument('--rank', default=0, type=int,
                    help='node rank for distributed training')
parser.add_argument('--dist-url', default='tcp://localhost:10001', 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=0, type=int,
                    help='seed for initializing training. ')
parser.add_argument('--num-sub-epoch', default=10, type=int)
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')
# COPD data configs:
parser.add_argument('--num-patch', default=581, type=int,
                    help='total number of patches in the atlas image.')
parser.add_argument('--root-dir', default='/pghbio/dbmi/batmanlab/lisun/copd/gnn_shared/data/patch_data_32_6_reg/',
                    help='root directory of COPD data')
parser.add_argument('--label-name', default=["FEV1pp_utah", "FEV1_FVC_utah", "finalGold"], nargs='+',
                    help='phenotype label names')
parser.add_argument('--label-name-set2', default=["Exacerbation_Frequency", "MMRCDyspneaScor"], nargs='+',
                    help='phenotype label names')
parser.add_argument('--visual-score', default=["Emph_Severity", "Emph_Paraseptal"], nargs='+',
                    help='phenotype label names')
parser.add_argument('--P2-Pheno', default=["Exacerbation_Frequency_P2"], nargs='+',
                    help='phenotype label names')
parser.add_argument('--fold', default=0, type=int,
                    help='fold index of cross validation')
parser.add_argument('--nhw-only', action='store_true',
                    help='only include white people')
# MoCo specific configs:
parser.add_argument('--moco-dim', default=128, type=int,
                    help='feature dimension (default: 128)')
parser.add_argument('--moco-k-graph', default=128, type=int,
                    help='queue size; number of negative keys (default: 65536)')
parser.add_argument('--moco-k-patch', default=128, type=int,
                    help='queue size; number of negative keys (default: 65536)')
parser.add_argument('--moco-m', default=0.999, type=float,
                    help='moco momentum of updating key encoder (default: 0.999)')
parser.add_argument('--moco-t', default=0.2, type=float,
                    help='softmax temperature (default: 0.07)')
# options for moco v2
parser.add_argument('--mlp', action='store_true',
                    help='use mlp head')
parser.add_argument('--cos', action='store_true',
                    help='use cosine lr schedule')
# experiment directory
parser.add_argument('--exp-name', default='debug')

def main():
    # read configurations
    args = parser.parse_args()

    # define experiment directory
    exp_dir = os.path.join('./exp', args.exp_name)
    if os.path.isdir(exp_dir) and (not os.path.isfile(args.resume)):
        #shutil.rmtree(exp_dir)
        os.makedirs(exp_dir, exist_ok=True)
    elif (not os.path.isdir(exp_dir)) and (not os.path.isfile(args.resume)):
        os.makedirs(exp_dir, exist_ok=True)
    elif (not os.path.isdir(exp_dir)) and os.path.isfile(args.resume):
        raise NotImplementedError("Experiment directory does not exist.")

    # save configurations to a dictionary
    with open(os.path.join(exp_dir, 'configs.json'), 'w') as f:
        json.dump(vars(args), f, indent=2)

    if args.seed is not None:
        random.seed(args.seed)
        torch.manual_seed(args.seed)
        torch.cuda.manual_seed_all(args.seed)
        torch.backends.cudnn.benchmark = True
        #cudnn.deterministic = True

    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
    print("Distributed:", args.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))
    else:
        # Simply call main_worker function
        main_worker(args.gpu, ngpus_per_node, args)


def main_worker(gpu, ngpus_per_node, args):
    args.gpu = gpu

    # suppress printing if not master
    if args.multiprocessing_distributed and args.gpu != 0:
        def print_pass(*args):
            pass
        builtins.print = print_pass

    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)
    if args.rank == 0:
       configure(os.path.join('./exp', args.exp_name))

    # create model
    model_patch = moco.builder_v3.MoCo(
        Encoder,
        args.num_patch, args.moco_dim, args.moco_k_patch, args.moco_m, args.moco_t, args.mlp)

    model_graph = moco.builder_graph.MoCo(
        GraphNet, args.gpu,
        args.moco_dim, args.moco_k_graph, args.moco_m, args.moco_t, args.mlp)

    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:
            torch.cuda.set_device(args.gpu)
            model_patch.cuda(args.gpu)
            model_graph.cuda(args.gpu)
            # 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_patch = int(args.batch_size_patch / ngpus_per_node)
            args.batch_size_graph = int(args.batch_size_graph / ngpus_per_node)
            args.workers_patch = int((args.workers_patch + ngpus_per_node - 1) / ngpus_per_node)
            args.workers_graph = int((args.workers_graph + ngpus_per_node - 1) / ngpus_per_node)
            model_patch = torch.nn.parallel.DistributedDataParallel(model_patch, device_ids=[args.gpu])
            model_graph = torch.nn.parallel.DistributedDataParallel(model_graph, device_ids=[args.gpu], find_unused_parameters=True)
        else:
            model.cuda()
            # 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:
        torch.cuda.set_device(args.gpu)
        model = model.cuda(args.gpu)
        # comment out the following line for debugging
        raise NotImplementedError("Only DistributedDataParallel is supported.")
    else:
        # AllGather implementation (batch shuffle, queue update, etc.) in
        # this code only supports DistributedDataParallel.
        raise NotImplementedError("Only DistributedDataParallel is supported.")

    # define loss function (criterion) and optimizer
    criterion = nn.CrossEntropyLoss().cuda(args.gpu)

    optimizer_patch = torch.optim.SGD(model_patch.parameters(), args.lr,
                                momentum=args.momentum,
                                weight_decay=args.weight_decay)
    optimizer_graph = torch.optim.SGD(model_graph.parameters(), args.lr,
                                momentum=args.momentum,
                                weight_decay=args.weight_decay)
    # save the initial model
    if not args.resume:
        save_checkpoint({
        'epoch': 0,
        'arch': args.arch,
        'state_dict': model_patch.state_dict(),
        'optimizer': optimizer_patch.state_dict(),
        }, is_best=False,
        filename=os.path.join(os.path.join('./exp', args.exp_name), 'checkpoint_patch_init.pth.tar'))
        save_checkpoint({
        'epoch': 0,
        'arch': args.arch,
        'state_dict': model_graph.state_dict(),
        'optimizer': optimizer_graph.state_dict(),
        }, is_best=False,
        filename=os.path.join(os.path.join('./exp', args.exp_name), 'checkpoint_graph_init.pth.tar'))

    # 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_patch = torch.load(args.resume)
                checkpoint_graph = torch.load(args.resume_graph)
            else:
                # Map model to be loaded to specified single gpu.
                loc = 'cuda:{}'.format(args.gpu)
                checkpoint_patch = torch.load(args.resume, map_location=loc)
                checkpoint_graph = torch.load(args.resume_graph, map_location=loc)
            args.start_epoch = checkpoint_patch['epoch']
            model_patch.load_state_dict(checkpoint_patch['state_dict'])
            model_graph.load_state_dict(checkpoint_graph['state_dict'])
            optimizer_patch.load_state_dict(checkpoint_patch['optimizer'])
            optimizer_graph.load_state_dict(checkpoint_graph['optimizer'])
            print("=> loaded checkpoint '{}' (epoch {})"
                  .format(args.resume, checkpoint_patch['epoch']))
        else:
            print("=> no checkpoint found at '{}'".format(args.resume))
            exit()

    transform_re = Rand3DElastic(mode='bilinear', prob=1.0,
        sigma_range=(8, 12),
        magnitude_range=(0, 1024+240), #[-1024, 240] -> [0, 1024+240]
        spatial_size=(32, 32, 32),
        translate_range=(12, 12, 12),
        rotate_range=(np.pi/18, np.pi/18, np.pi/18),
        scale_range=(0.1, 0.1, 0.1),
        padding_mode='border',
        device=torch.device('cuda:'+str(args.gpu)))
    transform_rgn = RandGaussianNoise(prob=0.25, mean=0.0, std=50)
    transform_rac = RandAdjustContrast(prob=0.25)

    train_tratransforms = Compose([transform_rac, transform_rgn, transform_re])

    train_dataset_patch = COPD_dataset_patch("train", args, moco.loader.TwoCropsTransform(train_tratransforms))
    train_dataset_graph = COPD_dataset_graph("train", args, moco.loader.TwoCropsTransform(train_tratransforms))

    if args.distributed:
        train_sampler_patch = torch.utils.data.distributed.DistributedSampler(train_dataset_patch)
        train_sampler_graph = torch.utils.data.distributed.DistributedSampler(train_dataset_graph)
    else:
        train_sampler = None

    train_loader_patch = torch.utils.data.DataLoader(
        train_dataset_patch, batch_size=args.batch_size_patch, shuffle=(train_sampler_patch is None),
        num_workers=args.workers_patch, pin_memory=True, sampler=train_sampler_patch, drop_last=True)
    train_loader_graph = torch.utils.data.DataLoader(
        train_dataset_graph, batch_size=args.batch_size_graph, shuffle=(train_sampler_graph is None),
        num_workers=args.workers_graph, pin_memory=True, sampler=train_sampler_graph, drop_last=True)

    for epoch in range(args.start_epoch, args.epochs):
        if args.distributed:
            train_sampler_patch.set_epoch(epoch)
        adjust_learning_rate(optimizer_patch, epoch, args)
        # train for one epoch
        train_patch(train_loader_patch, model_patch, criterion, optimizer_patch, epoch, args)
        # save model for every epoch
        if not args.multiprocessing_distributed or (args.multiprocessing_distributed
                and args.rank % ngpus_per_node == 0):
            save_checkpoint({
                'epoch': epoch + 1,
                'arch': args.arch,
                'state_dict': model_patch.state_dict(),
                'optimizer' : optimizer_patch.state_dict(),
            }, is_best=False, filename=os.path.join(os.path.join('./exp', args.exp_name), 'checkpoint_patch_{:04d}.pth.tar'.format(epoch)))

        for sub_epoch in range(args.num_sub_epoch):
            if args.distributed:
                train_sampler_graph.set_epoch(args.num_sub_epoch*epoch+sub_epoch)
            adjust_learning_rate(optimizer_graph, args.num_sub_epoch*epoch+sub_epoch, args)
            train_graph(train_loader_graph, model_graph, model_patch, criterion, optimizer_graph, args.num_sub_epoch*epoch+sub_epoch, args)
            if not args.multiprocessing_distributed or (args.multiprocessing_distributed
                and args.rank % ngpus_per_node == 0):
                save_checkpoint({
                'epoch': args.num_sub_epoch*epoch+sub_epoch + 1,
                'arch': args.arch,
                'state_dict': model_graph.state_dict(),
                'optimizer' : optimizer_graph.state_dict(),
                }, is_best=False, filename=os.path.join(os.path.join('./exp', args.exp_name), 'checkpoint_graph_{:04d}.pth.tar'.format(args.num_sub_epoch*epoch+sub_epoch)))


def train_graph(train_loader, model, model_patch, criterion, optimizer, epoch, args):
    batch_time = AverageMeter('Time', ':6.3f')
    data_time = AverageMeter('Data', ':6.3f')
    losses = AverageMeter('Loss', ':.4e')
    top1 = AverageMeter('Acc@1', ':6.2f')
    top5 = AverageMeter('Acc@5', ':6.2f')

    len_loader = len(train_loader)
    progress = ProgressMeter(
        len_loader,
        [batch_time, data_time, losses, top1, top5],
        prefix="Epoch: [{}]".format(epoch))

    # switch to train mode
    model.train()
    model_patch.eval()
    model_patch.module.encoder_q.flag = 1
    end = time.time()
 
    for i, data in enumerate(train_loader, 0):
        # measure data loading time
        data_time.update(time.time() - end)
        images, patch_loc, adj  = data
        if args.gpu is not None:
            patch_loc = patch_loc.float().cuda(args.gpu, non_blocking=True)
            adj = adj.float().cuda(args.gpu, non_blocking=True)

        # extract feature
        graph_batch_0 = []
        graph_batch_1 = []
        with torch.no_grad():
            for j in range(args.batch_size_graph):
                graph_batch_0.append( model_patch.module.encoder_q(images[0][j].cuda(args.gpu, non_blocking=True),\
                                                        patch_loc[j])[:,:args.moco_dim] ) # img in:[num_patch,1,size,size,size], output:[num_patch,moco_dim]
                graph_batch_1.append( model_patch.module.encoder_q(images[1][j].cuda(args.gpu, non_blocking=True),\
                                                        patch_loc[j])[:,:args.moco_dim] ) # patch_loc in:[num_patch,3]
        graph_batch_0 = torch.stack(graph_batch_0)
        graph_batch_1 = torch.stack(graph_batch_1) # output:[batch,num_patch,moco_dim]
        #print(graph_batch_0.shape)
        #print(adj.shape)
        # compute output
        output, target = model(im_q=[graph_batch_0, adj], im_k=[graph_batch_1, adj]) # adj:[batch,num_patch,num_patch]
        loss = criterion(output, target)

        # acc1/acc5 are (K+1)-way contrast classifier accuracy
        # measure accuracy and record loss
        acc1, acc5 = accuracy(output, target, topk=(1, 5))
        losses.update(loss.item(), images[0].size(0))
        top1.update(acc1[0], images[0].size(0))
        top5.update(acc5[0], images[0].size(0))

        # compute gradient and do SGD step
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # measure elapsed time
        batch_time.update(time.time() - end)
        end = time.time()

        if i % args.print_freq == 0:
            progress.display_graph(i)
            if args.rank==0:
                log_value('graph/epoch', epoch, i+epoch*len_loader)
                log_value('graph/loss', progress.meters[2].avg, i+epoch*len_loader)
                log_value('graph/acc_1', progress.meters[3].avg, i+epoch*len_loader)
                log_value('graph/acc_5', progress.meters[4].avg, i+epoch*len_loader)

def train_patch(train_loader, model, criterion, optimizer, epoch, args):
    batch_time = AverageMeter('Time', ':6.3f')
    data_time = AverageMeter('Data', ':6.3f')
    losses = AverageMeter('Loss', ':.4e')
    top1 = AverageMeter('Acc@1', ':6.2f')
    top5 = AverageMeter('Acc@5', ':6.2f')

    dataset_len = train_loader.dataset.sid_list_len
    progress = ProgressMeter(
        len(train_loader),
        [batch_time, data_time, losses, top1, top5],
        prefix="Epoch: [{}]".format(epoch))

    # switch to train mode
    model.train()
    model.module.encoder_q.flag = 0
    end = time.time()
    num_iter_epoch = len(train_loader)
    num_iter_sub_epoch = num_iter_epoch // args.num_patch
    print("num_iter_sub_epoch:", num_iter_sub_epoch)
        
    patch_idx = -1 # init patch_idx
    for i, data in enumerate(train_loader, 0):
        # measure data loading time
        data_time.update(time.time() - end)
        if i % num_iter_sub_epoch == 0:
            patch_idx+=1
            if patch_idx == args.num_patch: # tail issue
                break
            train_loader.dataset.set_patch_idx(patch_idx)

        images, patch_loc_idx, adj = data

        if args.gpu is not None:
            images[0] = images[0].cuda(args.gpu, non_blocking=True)
            images[1] = images[1].cuda(args.gpu, non_blocking=True)
            patch_loc_idx = patch_loc_idx.float().cuda(args.gpu, non_blocking=True)

        # compute output
        output, target = model(patch_idx=patch_idx, im_q=[images[0], patch_loc_idx], im_k=[images[1], patch_loc_idx])
        loss = criterion(output, target)

        # acc1/acc5 are (K+1)-way contrast classifier accuracy
        # measure accuracy and record loss
        acc1, acc5 = accuracy(output, target, topk=(1, 5))
        losses.update(loss.item(), images[0].size(0))
        top1.update(acc1[0], images[0].size(0))
        top5.update(acc5[0], images[0].size(0))

        # compute gradient and do SGD step
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # measure elapsed time
        batch_time.update(time.time() - end)
        end = time.time()

        if (i % args.print_freq == 0) and (i > 0):
            progress.display_patch(i, patch_idx)
            if args.rank==0:
                step = i + num_iter_epoch * epoch
                log_value('patch/epoch', epoch, step)
                log_value('patch/loss', progress.meters[2].avg, step)
                log_value('patch/acc_1', progress.meters[3].avg, step)
                log_value('patch/acc_5', progress.meters[4].avg, step)

def save_checkpoint(state, is_best, filename='checkpoint.pth.tar'):
    torch.save(state, filename)
    if is_best:
        shutil.copyfile(filename, 'model_best.pth.tar')


class AverageMeter(object):
    """Computes and stores the average and current value"""
    def __init__(self, name, fmt=':f'):
        self.name = name
        self.fmt = fmt
        self.reset()

    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 __str__(self):
        fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
        return fmtstr.format(**self.__dict__)


class ProgressMeter(object):
    def __init__(self, num_batches, meters, prefix="", patch_idx=0):
        self.batch_fmtstr = self._get_batch_fmtstr(num_batches)
        self.meters = meters
        self.prefix = prefix
        self.patch_idx = patch_idx

    def display_graph(self, batch):
        entries = [self.prefix + self.batch_fmtstr.format(batch)]
        entries += [str(meter) for meter in self.meters]
        print('\t'.join(entries))

    def display_patch(self, batch, patch_idx):
        entries = [self.prefix + self.batch_fmtstr.format(batch)]
        entries += ["Patch :[{}]".format(patch_idx)]
        entries += [str(meter) for meter in self.meters]
        print('\t'.join(entries))

    def _get_batch_fmtstr(self, num_batches):
        num_digits = len(str(num_batches // 1))
        fmt = '{:' + str(num_digits) + 'd}'
        return '[' + fmt + '/' + fmt.format(num_batches) + ']'


def adjust_learning_rate(optimizer, epoch, args):
    """Decay the learning rate based on schedule"""
    lr = args.lr
    if args.cos:  # cosine lr schedule
        lr *= 0.5 * (1. + math.cos(math.pi * epoch / args.epochs))
    else:  # stepwise lr schedule
        for milestone in args.schedule:
            lr *= 0.1 if epoch >= milestone else 1.
    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


if __name__ == '__main__':
    main()