# Copyright 2021 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# -*- coding: utf-8 -*-

import argparse
import datetime
import numpy as np
import time
import torch
# import torch_npu
import torch.backends.cudnn as cudnn
import json

from pathlib import Path

# from timm.data import Mixup
from mixup import Mixup
from timm.models import create_model
from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
from timm.scheduler import create_scheduler
# from timm.optim import create_optimizer
from timm.utils import ApexScaler

from datasets import build_dataset
from engine import train_one_epoch, evaluate
from losses import DistillationLoss
from samplers import RASampler
import gvt
import utils
import collections

from torch import optim as optim
from apex.optimizers import NpuFusedAdamW

use_npu = False


def add_weight_decay(model, weight_decay=1e-5, skip_list=()):
    decay = []
    no_decay = []
    for name, param in model.named_parameters():
        if not param.requires_grad:
            continue  # frozen weights
        if len(param.shape) == 1 or name.endswith(".bias") or name in skip_list:
            no_decay.append(param)
        else:
            decay.append(param)
    return [
        {'params': no_decay, 'weight_decay': 0.},
        {'params': decay, 'weight_decay': weight_decay}]


def create_optimizer(args, model, filter_bias_and_bn=True):
    opt_lower = args.opt.lower()
    weight_decay = args.weight_decay
    if weight_decay and filter_bias_and_bn:
        skip = {}
        if hasattr(model, 'no_weight_decay'):
            skip = model.no_weight_decay()
        parameters = add_weight_decay(model, weight_decay, skip)
        weight_decay = 0.
    else:
        parameters = model.parameters()

    opt_args = dict(lr=args.lr, weight_decay=weight_decay)
    if hasattr(args, 'opt_eps') and args.opt_eps is not None:
        opt_args['eps'] = args.opt_eps
    if hasattr(args, 'opt_betas') and args.opt_betas is not None:
        opt_args['betas'] = args.opt_betas

    opt_split = opt_lower.split('_')
    opt_lower = opt_split[-1]
    
    if opt_lower == 'adamw':
        optimizer = optim.AdamW(parameters, **opt_args)
    elif opt_lower == 'npufusedadamw':
        optimizer = NpuFusedAdamW(parameters, **opt_args)
    else:
        assert False and "Invalid optimizer"
        raise ValueError

    return optimizer


def useNPU():
    return use_npu


# to get the 参数解析器
def get_args_parser():
    parser = argparse.ArgumentParser('PVT training and evaluation script', add_help=False)
    parser.add_argument('--batch-size', default=64, type=int)
    parser.add_argument('--epochs', default=300, type=int)

    # Model parameters
    parser.add_argument('--model', default='pcpvt_small', type=str, metavar='MODEL',
                        help='Name of model to train')
    parser.add_argument('--input-size', default=224, type=int, help='images input size')

    # todo change dropout rate to 0.5
    parser.add_argument('--drop', type=float, default=0.0, metavar='PCT',
                        help='Dropout rate (default: 0.)')
    parser.add_argument('--drop-path', type=float, default=0.1, metavar='PCT',
                        help='Drop path rate (default: 0.1)')

    # Optimizer parameters
    # todo maybe use fusedadamw is better
    parser.add_argument('--opt', default='npufusedadamw', type=str, metavar='OPTIMIZER',
                        help='Optimizer (default: "npufusedadam"')
    parser.add_argument('--opt-eps', default=1e-8, type=float, metavar='EPSILON',
                        help='Optimizer Epsilon (default: 1e-8)')
    parser.add_argument('--opt-betas', default=None, type=float, nargs='+', metavar='BETA',
                        help='Optimizer Betas (default: None, use opt default)')
    parser.add_argument('--clip-grad', type=float, default=5, metavar='NORM',
                        help='Clip gradient norm (default: None, no clipping)')
    parser.add_argument('--momentum', type=float, default=0.9, metavar='M',
                        help='SGD momentum (default: 0.9)')
    parser.add_argument('--weight-decay', type=float, default=0.05,
                        help='weight decay (default: 0.05)')
    # Learning rate schedule parameters
    parser.add_argument('--sched', default='cosine', type=str, metavar='SCHEDULER',
                        help='LR scheduler (default: "cosine"')
    # todo change learning rate to 1e-3
    parser.add_argument('--lr', type=float, default=1e-3, metavar='LR',
                        help='learning rate (default: 5e-4)')
    parser.add_argument('--lr-noise', type=float, nargs='+', default=None, metavar='pct, pct',
                        help='learning rate noise on/off epoch percentages')
    parser.add_argument('--lr-noise-pct', type=float, default=0.67, metavar='PERCENT',
                        help='learning rate noise limit percent (default: 0.67)')
    parser.add_argument('--lr-noise-std', type=float, default=1.0, metavar='STDDEV',
                        help='learning rate noise std-dev (default: 1.0)')
    parser.add_argument('--warmup-lr', type=float, default=1e-6, metavar='LR',
                        help='warmup learning rate (default: 1e-6)')
    parser.add_argument('--min-lr', type=float, default=1e-5, metavar='LR',
                        help='lower lr bound for cyclic schedulers that hit 0 (1e-5)')

    parser.add_argument('--decay-epochs', type=float, default=30, metavar='N',
                        help='epoch interval to decay LR')
    parser.add_argument('--warmup-epochs', type=int, default=5, metavar='N',
                        help='epochs to warmup LR, if scheduler supports')
    parser.add_argument('--cooldown-epochs', type=int, default=10, metavar='N',
                        help='epochs to cooldown LR at min_lr, after cyclic schedule ends')
    parser.add_argument('--patience-epochs', type=int, default=10, metavar='N',
                        help='patience epochs for Plateau LR scheduler (default: 10')
    parser.add_argument('--decay-rate', '--dr', type=float, default=0.1, metavar='RATE',
                        help='LR decay rate (default: 0.1)')

    # Augmentation parameters
    parser.add_argument('--color-jitter', type=float, default=0.4, metavar='PCT',
                        help='Color jitter factor (default: 0.4)')
    parser.add_argument('--aa', type=str, default='rand-m9-mstd0.5-inc1', metavar='NAME',
                        help='Use AutoAugment policy. "v0" or "original". " + \
                             "(default: rand-m9-mstd0.5-inc1)'),
    parser.add_argument('--smoothing', type=float, default=0.1, help='Label smoothing (default: 0.1)')
    parser.add_argument('--train-interpolation', type=str, default='bicubic',
                        help='Training interpolation (random, bilinear, bicubic default: "bicubic")')

    parser.add_argument('--repeated-aug', action='store_true')
    parser.add_argument('--no-repeated-aug', action='store_false', dest='repeated_aug')
    parser.set_defaults(repeated_aug=False)

    # * Random Erase params
    parser.add_argument('--reprob', type=float, default=0.25, metavar='PCT',
                        help='Random erase prob (default: 0.25)')
    parser.add_argument('--remode', type=str, default='pixel',
                        help='Random erase mode (default: "pixel")')
    parser.add_argument('--recount', type=int, default=1,
                        help='Random erase count (default: 1)')
    parser.add_argument('--resplit', action='store_true', default=False,
                        help='Do not random erase first (clean) augmentation split')

    # * Mixup params
    parser.add_argument('--mixup', type=float, default=0.8,
                        help='mixup alpha, mixup enabled if > 0. (default: 0.8)')
    parser.add_argument('--cutmix', type=float, default=1.0,
                        help='cutmix alpha, cutmix enabled if > 0. (default: 1.0)')
    parser.add_argument('--cutmix-minmax', type=float, nargs='+', default=None,
                        help='cutmix min/max ratio, overrides alpha and enables cutmix if set (default: None)')
    parser.add_argument('--mixup-prob', type=float, default=1.0,
                        help='Probability of performing mixup or cutmix when either/both is enabled')
    parser.add_argument('--mixup-switch-prob', type=float, default=0.5,
                        help='Probability of switching to cutmix when both mixup and cutmix enabled')
    parser.add_argument('--mixup-mode', type=str, default='batch',
                        help='How to apply mixup/cutmix params. Per "batch", "pair", or "elem"')

    # * Finetuning params
    parser.add_argument('--finetune', default='', help='finetune from checkpoint')
    parser.add_argument('--max_step', default=None)

    # Dataset parameters
    parser.add_argument('--data-path', default='/datasets01/imagenet_full_size/061417/', type=str,
                        help='dataset path')
    parser.add_argument('--data-set', default='IMNET', choices=['CIFAR', 'IMNET', 'INAT', 'INAT19'],
                        type=str, help='Image Net dataset path')
    parser.add_argument('--use-mcloader', action='store_true', default=False, help='Use mcloader')
    parser.add_argument('--inat-category', default='name',
                        choices=['kingdom', 'phylum', 'class', 'order', 'supercategory', 'family', 'genus', 'name'],
                        type=str, help='semantic granularity')

    # todo specify output dir while training
    parser.add_argument('--output_dir', default='',
                        help='path where to save, empty for no saving')
    parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')
    parser.add_argument('--seed', default=0, type=int)
    parser.add_argument('--resume', default='', help='resume from checkpoint')
    parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
                        help='start epoch')
    parser.add_argument('--eval', action='store_true', help='Perform evaluation only')
    # distributed evaluation
    parser.add_argument('--dist-eval', action='store_true', default=False, help='Enabling distributed evaluation')
    parser.add_argument('--num_workers', default=10, type=int)
    parser.add_argument('--pin-mem', action='store_true',
                        help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
    parser.add_argument('--no-pin-mem', action='store_false', dest='pin_mem',
                        help='')
    # default to use pin cpu memory
    parser.set_defaults(pin_mem=True)

    # distributed training parameters
    parser.add_argument('--world_size', default=1, type=int,
                        help='number of distributed processes')
    parser.add_argument('--dist_url', default='env://', help='url used to set up distributed training')

    # test throught
    parser.add_argument('--throughout', action='store_true', help='Perform throughout only')
    return parser


# 在评估时，计算吞吐率
@torch.no_grad()
def throughput(data_loader, model, logger):
    model.eval()

    for idx, (images, _) in enumerate(data_loader):
        if useNPU():
            images = images.npu(non_blocking=True)
        else:
            images = images.cuda(non_blocking=True)
        batch_size = images.shape[0]
        for i in range(50):
            model(images)
        if useNPU():
            torch.npu.synchronize()
        else:
            torch.cuda.synchronize()
        logger.info(f"throughput averaged with 30 times")
        tic1 = time.time()
        for i in range(30):
            model(images)
        if useNPU():
            torch.npu.synchronize()
        else:
            torch.cuda.synchronize()
        tic2 = time.time()
        logger.info(f"batch_size {batch_size} throughput {30 * batch_size / (tic2 - tic1)}")
        return


# 主入口函数
def main(args):
    if 'npu' in args.device:
        devs_per_node = torch.npu.device_count()
    else:
        devs_per_node = torch.cuda.device_count()
    print(f"number of devices available: {devs_per_node}")
    # 针对每个进程开启分布式模式
    use_npu = 'npu' in args.device
    utils.init_distributed_mode(args, use_npu)
    # 打印参数列表
    print(args)

    # torch.device代表将torch.Tensor分配到的设备的对象
    # print(f"device before: {torch.cuda.current_device()}")
    device = torch.device(args.device)
    print(f"using device: {device}")

    if 'npu' in args.device:
        print(f"device after: {torch.npu.current_device()}")
    else:
        print(f"device after: {torch.cuda.current_device()}")

    # fix the seed for reproducibility
    seed = args.seed + utils.get_rank()
    torch.manual_seed(seed)
    np.random.seed(seed)
    cudnn.benchmark = True

    dataset_train, args.nb_classes = build_dataset(is_train=True, args=args)
    dataset_val, _ = build_dataset(is_train=False, args=args)

    if args.distributed:
        num_tasks = utils.get_world_size()
        global_rank = utils.get_rank()
        if args.repeated_aug:
            sampler_train = RASampler(
                dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
            )
        else:
            # 打印分支及所采用的sampler
            print(f"rank {global_rank} is using DistributedSampler with {num_tasks} tasks")
            sampler_train = torch.utils.data.DistributedSampler(
                dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
            )
        if args.dist_eval:
            if len(dataset_val) % num_tasks != 0:
                print('Warning: Enabling distributed evaluation with an eval dataset not divisible by process number. '
                      'This will slightly alter validation results as extra duplicate entries are added to achieve '
                      'equal num of samples per-process.')
            sampler_val = torch.utils.data.DistributedSampler(
                dataset_val, num_replicas=num_tasks, rank=global_rank, shuffle=False)
        else:
            sampler_val = torch.utils.data.SequentialSampler(dataset_val)
    else:
        sampler_train = torch.utils.data.RandomSampler(dataset_train)
        sampler_val = torch.utils.data.SequentialSampler(dataset_val)

    data_loader_train = torch.utils.data.DataLoader(
        dataset_train, sampler=sampler_train,
        batch_size=args.batch_size,
        num_workers=args.num_workers,
        pin_memory=args.pin_mem,
        drop_last=True,
    )

    data_loader_val = torch.utils.data.DataLoader(
        dataset_val, sampler=sampler_val,
        # val batch size = 192 if train batch-size is 128
        batch_size=int(1.5 * args.batch_size),
        num_workers=args.num_workers,
        pin_memory=args.pin_mem,
        drop_last=False
    )

    # mixup: applies different params to each element or whole batch
    mixup_fn = None
    mixup_active = args.mixup > 0 or args.cutmix > 0. or args.cutmix_minmax is not None
    if mixup_active:
        mixup_fn = Mixup(
            mixup_alpha=args.mixup, cutmix_alpha=args.cutmix, cutmix_minmax=args.cutmix_minmax,
            prob=args.mixup_prob, switch_prob=args.mixup_switch_prob, mode=args.mixup_mode,
            label_smoothing=args.smoothing, num_classes=args.nb_classes)

    print(f"Creating model: {args.model}")

    model = create_model(
        args.model,
        pretrained=False,
        num_classes=args.nb_classes,
        drop_rate=args.drop,
        drop_path_rate=args.drop_path,
        drop_block_rate=None,
    )
    model.to(device)
    model_ema = None

    assert torch.backends.cudnn.enabled, "Amp requires cudnn backend to be enabled."

    # 创建优化器等，使用了timm库
    optimizer = create_optimizer(args, model)

    model, optimizer = utils.initialize_amp(model, optimizer)

    if args.distributed:
        # 自动处理并行化的wrapper
        # if utils.hasApex():
        #    print("using apex parallel:")
        #    import apex
        #    model = apex.parallel.DistributedDataParallel(model, delay_allreduce=True)
        # else:
        print("using torch parallel")
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
        model_without_ddp = model.module
    else:
        model_without_ddp = model

    # 打印需要训练的参数数量
    n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
    print('number of params:', n_parameters)

    # 分布式训练需要对learning rate进行缩放
    linear_scaled_lr = args.lr * args.batch_size * utils.get_world_size() / 512.0
    args.lr = linear_scaled_lr
    loss_scaler = ApexScaler()

    lr_scheduler, _ = create_scheduler(args, optimizer)

    criterion = LabelSmoothingCrossEntropy()

    if args.mixup > 0.:
        # smoothing is handled with mixup label transform
        criterion = SoftTargetCrossEntropy()
    elif args.smoothing:
        criterion = LabelSmoothingCrossEntropy(smoothing=args.smoothing)
    else:
        criterion = torch.nn.CrossEntropyLoss()

    criterion = DistillationLoss(
        criterion, None, 'none', 0, 0
    )

    if not args.output_dir:
        args.output_dir = args.model
        if utils.is_main_process():
            import os
            if not os.path.exists(args.model):
                os.mkdir(args.model)

    output_dir = Path(args.output_dir)
    # 是否恢复训练
    if args.resume:
        if args.resume.startswith('https'):
            checkpoint = torch.hub.load_state_dict_from_url(
                args.resume, map_location='cpu', check_hash=True)
        else:
            checkpoint = torch.load(args.resume, map_location='cpu')
        if 'model' in checkpoint:
            model_without_ddp.load_state_dict(checkpoint['model'])
        else:
            model_without_ddp.load_state_dict(checkpoint)

        if not args.eval and 'optimizer' in checkpoint and 'lr_scheduler' in checkpoint and 'epoch' in checkpoint:
            optimizer.load_state_dict(checkpoint['optimizer'])
            lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
            args.start_epoch = checkpoint['epoch'] + 1
            if 'scaler' in checkpoint:
                loss_scaler.load_state_dict(checkpoint['scaler'])
    # 是否只进行评估
    if args.eval:
        test_stats = evaluate(data_loader_val, model, device)
        print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
        return
    # 是否只进行吞吐量测试
    if args.throughout:
        from logger import create_logger
        logger = create_logger(output_dir=output_dir, dist_rank=utils.get_rank(), name=args.model)
        throughput(data_loader_val, model, logger)
        return

    print(f"Start training for {args.epochs} epochs")
    start_time = time.time()
    max_accuracy = 0.0
    # 指定开始epoch和结束epoch
    for epoch in range(args.start_epoch, args.epochs):
        if args.distributed:
            # 对于shuffle有用
            data_loader_train.sampler.set_epoch(epoch)

        # 获取训练信息
        train_stats = train_one_epoch(
            model, criterion, data_loader_train,
            optimizer, device, epoch, loss_scaler,
            args.clip_grad, model_ema, mixup_fn,
            set_training_mode=args.finetune == '',  # keep in eval mode during finetuning
            max_steps=args.max_step
        )

        lr_scheduler.step(epoch)
        if args.output_dir:
            checkpoint_paths = [output_dir / 'checkpoint.pth']
            for checkpoint_path in checkpoint_paths:
                utils.save_on_master({
                    'model': model_without_ddp.state_dict(),
                    'optimizer': optimizer.state_dict(),
                    'lr_scheduler': lr_scheduler.state_dict(),
                    'epoch': epoch,
                    'scaler': loss_scaler.state_dict(),
                    'args': args,
                }, checkpoint_path)

        # 获取验证信息
        test_stats = evaluate(data_loader_val, model, device)
        print(f"Accuracy of the network on the {len(dataset_val)} test images: {test_stats['acc1']:.1f}%")
        if test_stats["acc1"] > max_accuracy:
            if args.output_dir:
                checkpoint_paths = [output_dir / 'checkpoint_best.pth']
                for checkpoint_path in checkpoint_paths:
                    utils.save_on_master({
                        'model': model_without_ddp.state_dict(),
                        'optimizer': optimizer.state_dict(),
                        'lr_scheduler': lr_scheduler.state_dict(),
                        'epoch': epoch,
                        'args': args,
                    }, checkpoint_path)

        max_accuracy = max(max_accuracy, test_stats["acc1"])
        print(f'Max accuracy: {max_accuracy:.2f}%')

        log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
                     **{f'test_{k}': v for k, v in test_stats.items()},
                     'epoch': epoch,
                     'n_parameters': n_parameters}
        # 将训练和测试信息写入json文件
        if args.output_dir and utils.is_main_process():
            with (output_dir / "log.txt").open("a") as f:
                f.write(json.dumps(log_stats) + "\n")

    total_time = time.time() - start_time
    total_time_str = str(datetime.timedelta(seconds=int(total_time)))
    print('Training time {}'.format(total_time_str))


if __name__ == '__main__':
    parser = argparse.ArgumentParser('Twins training and evaluation script', parents=[get_args_parser()])
    args = parser.parse_args()
    if args.output_dir:
        Path(args.output_dir).mkdir(parents=True, exist_ok=True)
    if 'npu' in args.device:
        use_npu = True
    main(args)