（笔记）yolov5-5中train.py一些简单注释

train.py里面加了很多额外的功能，使得整体看起来比较复杂，其实核心部分主要就是

读取数据集，

加载模型，

训练中损失的计算。

这里简单的将train.py按每部分的功能进行了一些注释。

import argparse
import logging
import math
import os
import random
import time
from copy import deepcopy
from pathlib import Path
from threading import Thread

import numpy as np
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.optim.lr_scheduler as lr_scheduler
import torch.utils.data
import yaml
from torch.cuda import amp
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.utils.tensorboard import SummaryWriter
from tqdm import tqdm

import test  # import test.py to get mAP after each epoch
from models.experimental import attempt_load
from models.yolo import Model
from utils.autoanchor import check_anchors
from utils.datasets import create_dataloader
from utils.general import labels_to_class_weights, increment_path, labels_to_image_weights, init_seeds, \
    fitness, strip_optimizer, get_latest_run, check_dataset, check_file, check_git_status, check_img_size, \
    check_requirements, print_mutation, set_logging, one_cycle, colorstr
from utils.google_utils import attempt_download
from utils.loss import ComputeLoss
from utils.plots import plot_images, plot_labels, plot_results, plot_evolution
from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first, is_parallel
from utils.wandb_logging.wandb_utils import WandbLogger, check_wandb_resume

logger = logging.getLogger(__name__)


def train(hyp, opt, device, tb_writer=None):
    logger.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))              #读取hyp超参数文件
    save_dir, epochs, batch_size, total_batch_size, weights, rank = \
        Path(opt.save_dir), opt.epochs, opt.batch_size, opt.total_batch_size, opt.weights, opt.global_rank

    '''
    创建目录，设置模型、txt等保存的路径
    save_dir = Path(opt.save_dir) ，save_dir就是根据opt参数里面所设置的路径而生成的目录
    '''
    # Directories
    wdir = save_dir / 'weights'
    wdir.mkdir(parents=True, exist_ok=True)  # make dir
    last = wdir / 'last.pt'
    best = wdir / 'best.pt'
    results_file = save_dir / 'results.txt'

    '''
    将本次运行的超参数(hyp),和选项操作(opt)给保存成yaml格式,
    保存在了每次训练得到的exp文件中，这两个yaml显示了我们本次训练所选择的超参数和opt参数，opt参数是train代码下面那一堆参数选择
    '''
    # Save run settings
    with open(save_dir / 'hyp.yaml', 'w') as f:
        yaml.dump(hyp, f, sort_keys=False)
    with open(save_dir / 'opt.yaml', 'w') as f:
        yaml.dump(vars(opt), f, sort_keys=False)

    '''
    配置:画图开关,cuda,种子,读取数据集相关的yaml文件
    '''
    # Configure
    plots = not opt.evolve  # create plots
    cuda = device.type != 'cpu'
    init_seeds(2 + rank)
    with open(opt.data,encoding='utf-8') as f:
        data_dict = yaml.load(f, Loader=yaml.SafeLoader)  # data dict
    is_coco = opt.data.endswith('coco.yaml')

    '''
    加载相关日志功能:如logger,wandb
    '''
    # Logging- Doing this before checking the dataset. Might update data_dict
    loggers = {'wandb': None}  # loggers dict
    if rank in [-1, 0]:
        opt.hyp = hyp  # add hyperparameters
        run_id = torch.load(weights).get('wandb_id') if weights.endswith('.pt') and os.path.isfile(weights) else None
        wandb_logger = WandbLogger(opt, Path(opt.save_dir).stem, run_id, data_dict)
        loggers['wandb'] = wandb_logger.wandb
        data_dict = wandb_logger.data_dict
        if wandb_logger.wandb:
            weights, epochs, hyp = opt.weights, opt.epochs, opt.hyp  # WandbLogger might update weights, epochs if resuming

    nc = 1 if opt.single_cls else int(data_dict['nc'])  # number of classes
    names = ['item'] if opt.single_cls and len(data_dict['names']) != 1 else data_dict['names']  # class names
    assert len(names) == nc, '%g names found for nc=%g dataset in %s' % (len(names), nc, opt.data)  # check

    '''
    加载模型
    '''
    # Model
    pretrained = weights.endswith('.pt')
    if pretrained:
        with torch_distributed_zero_first(rank):
            attempt_download(weights)  # download if not found locally
        ckpt = torch.load(weights, map_location=device)  # load checkpoint
        model = Model(opt.cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device)  # create
        exclude = ['anchor'] if (opt.cfg or hyp.get('anchors')) and not opt.resume else []  # exclude keys
        state_dict = ckpt['model'].float().state_dict()  # to FP32
        state_dict = intersect_dicts(state_dict, model.state_dict(), exclude=exclude)  # intersect
        model.load_state_dict(state_dict, strict=False)  # load
        logger.info('Transferred %g/%g items from %s' % (len(state_dict), len(model.state_dict()), weights))  # report
    else:
        model = Model(opt.cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device)  # create
    with torch_distributed_zero_first(rank):
        check_dataset(data_dict)  # check
    train_path = data_dict['train']
    test_path = data_dict['val']

    '''
    冰冻一些层，使得这些层在反向传播的时候不再更新权重,需要冻结的层,可以写在freeze列表中
    '''
    # Freeze
    freeze = []  # parameter names to freeze (full or partial)
    for k, v in model.named_parameters():
        v.requires_grad = True  # train all layers
        if any(x in k for x in freeze):
            print('freezing %s' % k)
            v.requires_grad = False

    '''
    nbs为名义批次,比如实际批次为16,那么64/16=4,每4次迭代，才进行一次反向传播更新权重，可以节约显存.
    '''
    # Optimizer
    nbs = 64  # nominal batch size
    accumulate = max(round(nbs / total_batch_size), 1)  # accumulate loss before optimizing
    hyp['weight_decay'] *= total_batch_size * accumulate / nbs  # scale weight_decay
    logger.info(f"Scaled weight_decay = {hyp['weight_decay']}")

    '''
    设置优化器，权重weight使用了正则化,偏置bias则不使用正则化
    '''
    pg0, pg1, pg2 = [], [], []  # optimizer parameter groups
    for k, v in model.named_modules():
        if hasattr(v, 'bias') and isinstance(v.bias, nn.Parameter):
            pg2.append(v.bias)  # biases
        if isinstance(v, nn.BatchNorm2d):
            pg0.append(v.weight)  # no decay
        elif hasattr(v, 'weight') and isinstance(v.weight, nn.Parameter):
            pg1.append(v.weight)  # apply decay

    if opt.adam:
        optimizer = optim.Adam(pg0, lr=hyp['lr0'], betas=(hyp['momentum'], 0.999))  # adjust beta1 to momentum
    else:
        optimizer = optim.SGD(pg0, lr=hyp['lr0'], momentum=hyp['momentum'], nesterov=True)

    optimizer.add_param_group({'params': pg1, 'weight_decay': hyp['weight_decay']})  # add pg1 with weight_decay
    optimizer.add_param_group({'params': pg2})  # add pg2 (biases)
    logger.info('Optimizer groups: %g .bias, %g conv.weight, %g other' % (len(pg2), len(pg1), len(pg0)))
    del pg0, pg1, pg2

    # Scheduler https://arxiv.org/pdf/1812.01187.pdf
    # https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR

    '''
    设置学习率策略:两者可供选择，线性学习率和余弦退火学习率
    '''
    if opt.linear_lr:
        lf = lambda x: (1 - x / (epochs - 1)) * (1.0 - hyp['lrf']) + hyp['lrf']  # linear
    else:
        lf = one_cycle(1, hyp['lrf'], epochs)  # cosine 1->hyp['lrf'] 余弦退火方式
    scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
    # plot_lr_scheduler(optimizer, scheduler, epochs)

    '''
    设置ema（指数移动平均）:目的是为了收敛的曲线更加平滑
    '''
    # EMA
    ema = ModelEMA(model) if rank in [-1, 0] else None

    '''
    继续接着训练,需要加载优化器,ema模型,训练结果txt,周期
    '''
    # Resume
    start_epoch, best_fitness = 0, 0.0
    if pretrained:
        # Optimizer
        if ckpt['optimizer'] is not None:
            optimizer.load_state_dict(ckpt['optimizer'])
            best_fitness = ckpt['best_fitness']

        # EMA
        if ema and ckpt.get('ema'):
            ema.ema.load_state_dict(ckpt['ema'].float().state_dict())
            ema.updates = ckpt['updates']

        # Results
        if ckpt.get('training_results') is not None:
            results_file.write_text(ckpt['training_results'])  # write results.txt

        # Epochs
        start_epoch = ckpt['epoch'] + 1
        if opt.resume:
            assert start_epoch > 0, '%s training to %g epochs is finished, nothing to resume.' % (weights, epochs)
        if epochs < start_epoch:
            logger.info('%s has been trained for %g epochs. Fine-tuning for %g additional epochs.' %
                        (weights, ckpt['epoch'], epochs))
            epochs += ckpt['epoch']  # finetune additional epochs

        del ckpt, state_dict

    '''
    模型默认的下采样倍率model.stride: [8,16,32]
    gs代表模型下采样的最大步长: 后续为了保证输入模型的图片宽高是最大步长的整数倍
    nl代表模型输出的尺度,默认为3个尺度, 分别下采样8倍，16倍，32倍.   nl=3

    imgsz, imgsz_test代表训练和测试的图片大小，比如opt.img_size=[640,480]，那么训练图片的最大边为640,测试图片最大边为480
    如果opt.img_size=[640],那么自动补成[640,640]

    当然比如这边imgsz是640,那么训练的图片是640*640吗，不一定，具体看你怎么设置，默认是padding成正方形进行训练的.
    '''
    # Image sizes
    gs = max(int(model.stride.max()), 32)  # grid size (max stride)
    nl = model.model[-1].nl  # number of detection layers (used for scaling hyp['obj'])
    imgsz, imgsz_test = [check_img_size(x, gs) for x in opt.img_size]  # verify imgsz are gs-multiples

    '''
    多卡训练
    '''
    # DP mode
    if cuda and rank == -1 and torch.cuda.device_count() > 1:
        model = torch.nn.DataParallel(model)

    # SyncBatchNorm
    if opt.sync_bn and cuda and rank != -1:
        model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
        logger.info('Using SyncBatchNorm()')

    '''
    加载数据集
    '''
    # Trainloader
    dataloader, dataset = create_dataloader(train_path, imgsz, batch_size, gs, opt,
                                            hyp=hyp, augment=True, cache=opt.cache_images, rect=opt.rect, rank=rank,
                                            world_size=opt.world_size, workers=opt.workers,
                                            image_weights=opt.image_weights, quad=opt.quad, prefix=colorstr('train: '))

    '''
    检验加载的数据集是否正确:  利用数据集中的最大类别 best_fitness:
                best_fitness = fi
            wandb_logger.end_epoch(best_result=best_fitness == fi)

            # Save model
            if (not opt.nosave) or (final_epoch and not opt.evolve):  # if save
                ckpt = {'epoch': epoch,
                        'best_fitness': best_fitness,
                        'training_results': results_file.read_text(),
                        'model': deepcopy(model.module if is_parallel(model) else model).half(),
                        'ema': deepcopy(ema.ema).half(),
                        'updates': ema.updates,
                        'optimizer': optimizer.state_dict(),
                        'wandb_id': wandb_logger.wandb_run.id if wandb_logger.wandb else None}

                # Save last, best and delete
                torch.save(ckpt, last)
                if best_fitness == fi:
                    torch.save(ckpt, best)
                if wandb_logger.wandb:
                    if ((epoch + 1) % opt.save_period == 0 and not final_epoch) and opt.save_period != -1:
                        wandb_logger.log_model(
                            last.parent, opt, epoch, fi, best_model=best_fitness == fi)
                del ckpt

        # end epoch ----------------------------------------------------------------------------------------------------
    # end training
    if rank in [-1, 0]:
        # Plots
        if plots:
            plot_results(save_dir=save_dir)  # save as results.png
            if wandb_logger.wandb:
                files = ['results.png', 'confusion_matrix.png', *[f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R')]]
                wandb_logger.log({"Results": [wandb_logger.wandb.Image(str(save_dir / f), caption=f) for f in files
                                              if (save_dir / f).exists()]})
        # Test best.pt
        logger.info('%g epochs completed in %.3f hours.\n' % (epoch - start_epoch + 1, (time.time() - t0) / 3600))
        if opt.data.endswith('coco.yaml') and nc == 80:  # if COCO
            for m in (last, best) if best.exists() else (last):  # speed, mAP tests
                results, _, _ = test.test(opt.data,
                                          batch_size=batch_size * 2,
                                          imgsz=imgsz_test,
                                          conf_thres=0.001,
                                          iou_thres=0.7,
                                          model=attempt_load(m, device).half(),
                                          single_cls=opt.single_cls,
                                          dataloader=testloader,
                                          save_dir=save_dir,
                                          save_json=True,
                                          plots=False,
                                          is_coco=is_coco)

        # Strip optimizers
        final = best if best.exists() else last  # final model
        for f in last, best:
            if f.exists():
                strip_optimizer(f)  # strip optimizers
        if opt.bucket:
            os.system(f'gsutil cp {final} gs://{opt.bucket}/weights')  # upload
        if wandb_logger.wandb and not opt.evolve:  # Log the stripped model
            wandb_logger.wandb.log_artifact(str(final), type='model',
                                            name='run_' + wandb_logger.wandb_run.id + '_model',
                                            aliases=['last', 'best', 'stripped'])
        wandb_logger.finish_run()
    else:
        dist.destroy_process_group()
    torch.cuda.empty_cache()
    return results


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights', type=str, default='yolov5s.pt', help='initial weights path')                        #选择用来训练的网络模型路径，当default为空时，就是没有预训练模型，从头开始训练
    parser.add_argument('--cfg', type=str, default='models/yolov5s.yaml', help='model.yaml path')                        #网络模型的配置参数，地址在models/hub中的yaml文件
    parser.add_argument('--data', type=str, default='data/new.yaml', help='data.yaml path')                              #数据集地址
    parser.add_argument('--hyp', type=str, default='data/hyp.scratch.yaml', help='hyperparameters path')                 #超参数，data里面的hyp.xx.yaml两个二选一
    parser.add_argument('--epochs', type=int, default=300)                                                               #设置训练多少轮
    parser.add_argument('--batch-size', type=int, default=32, help='total batch size for all GPUs')                      #设置batch_size，每次送入网络多少张图片
    parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='[train, test] image sizes')         #设置训练和预测时候的图片尺寸大小（保持一致）
    parser.add_argument('--rect', action='store_true', help='rectangular training')                                      #矩阵训练方式，默认关闭
    parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')            #继续上次中止的训练，在填入default="上次训练的权重位置"即可接着训练
    parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')                              #是否只保存最后一次训练的权重，不用设置
    parser.add_argument('--notest', action='store_true', help='only test final epoch')                                   #是否只在最后一轮进行测试，不用设置
    parser.add_argument('--noautoanchor', action='store_true', help='disable autoanchor check')                          #是否禁止采用锚框，不用设置
    parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters')                                  #超参数进化，超参数的调优，默认关闭
    parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')                                          #不用管
    parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')                  #是否把图片缓存用于更好的训练中，默认关闭
    parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')        #对上一轮测试效果不是很好的图片
                                                                                                                         #在下一轮中对这些图片加一些相关的权重，着重训练，默认关闭
    parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')                            #选择GPU还是CPU不用设置系统会自动选择
    parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')                             #对图片尺寸进行变换，多尺度训练，默认关闭
    parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')              #为的单类别还是多类别，默认多类别
    parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')                          #Adam优化器，默认是不用的，用的是随机梯度下降
    parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')          #不用看
    parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')                       #不用看
    parser.add_argument('--workers', type=int, default=4, help='maximum number of dataloader workers')                   #可以先将workers改为0，训练之后没什么问题再调大
    parser.add_argument('--project', default='runs/train', help='save to project/name')                                  #训练过的权重文件保存路径
    parser.add_argument('--entity', default=None, help='W&B entity')                                                     #不用管
    parser.add_argument('--name', default='exp', help='save to project/name')                                            #存放训练好的权重文件的文件名
    parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')            #如果设置为True，每次训练的结果就不会每次新建一个exp而是在一个文件，没啥用
    parser.add_argument('--quad', action='store_true', help='quad dataloader')                                           #是否选择quad dataloader这种取数据方式，当训练尺寸>640时效果更好
                                                                                                                         #在640尺寸上效果没有默认dataloder好，默认不用
    parser.add_argument('--linear-lr', action='store_true', help='linear LR')                                            #一种学习率优化方式，默认的是余弦退火方式
    parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')                    #标签平滑可以设置为0.01、0.005，防止分类算法中过拟合的情况产生
    parser.add_argument('--upload_dataset', action='store_true', help='Upload dataset as W&B artifact table')
    parser.add_argument('--bbox_interval', type=int, default=-1, help='Set bounding-box image logging interval for W&B')
    parser.add_argument('--save_period', type=int, default=-1, help='Log model after every "save_period" epoch')         #不用管
    # parser.add_argument('--artifact_alias', type=str, default="latest", help='version of dataset artifact to be used')   作者还没实现
    opt = parser.parse_args()

    # Set DDP variables
    opt.world_size = int(os.environ['WORLD_SIZE']) if 'WORLD_SIZE' in os.environ else 1
    opt.global_rank = int(os.environ['RANK']) if 'RANK' in os.environ else -1
    set_logging(opt.global_rank)
    if opt.global_rank in [-1, 0]:
        check_git_status()
        check_requirements()

    # Resume
    wandb_run = check_wandb_resume(opt)
    if opt.resume and not wandb_run:  # resume an interrupted run
        ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run()  # specified or most recent path
        assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
        apriori = opt.global_rank, opt.local_rank
        with open(Path(ckpt).parent.parent / 'opt.yaml') as f:
            opt = argparse.Namespace(**yaml.load(f, Loader=yaml.SafeLoader))  # replace
        opt.cfg, opt.weights, opt.resume, opt.batch_size, opt.global_rank, opt.local_rank = '', ckpt, True, opt.total_batch_size, *apriori  # reinstate
        logger.info('Resuming training from %s' % ckpt)
    else:
        # opt.hyp = opt.hyp or ('hyp.finetune.yaml' if opt.weights else 'hyp.scratch.yaml')
        opt.data, opt.cfg, opt.hyp = check_file(opt.data), check_file(opt.cfg), check_file(opt.hyp)  # check files
        assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
        opt.img_size.extend([opt.img_size[-1]] * (2 - len(opt.img_size)))  # extend to 2 sizes (train, test)
        opt.name = 'evolve' if opt.evolve else opt.name
        opt.save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok | opt.evolve)  # increment run

    # DDP mode
    opt.total_batch_size = opt.batch_size
    device = select_device(opt.device, batch_size=opt.batch_size)
    if opt.local_rank != -1:
        assert torch.cuda.device_count() > opt.local_rank
        torch.cuda.set_device(opt.local_rank)
        device = torch.device('cuda', opt.local_rank)
        dist.init_process_group(backend='nccl', init_method='env://')  # distributed backend
        assert opt.batch_size % opt.world_size == 0, '--batch-size must be multiple of CUDA device count'
        opt.batch_size = opt.total_batch_size // opt.world_size

    # Hyperparameters
    with open(opt.hyp) as f:
        hyp = yaml.load(f, Loader=yaml.SafeLoader)  # load hyps

    # Train 训练模式
    logger.info(opt)
    if not opt.evolve:
        tb_writer = None  # init loggers
        if opt.global_rank in [-1, 0]:
            prefix = colorstr('tensorboard: ')
            logger.info(f"{prefix}Start with 'tensorboard --logdir {opt.project}', view at http://localhost:6006/")
            tb_writer = SummaryWriter(opt.save_dir)  # Tensorboard
        train(hyp, opt, device, tb_writer)

    # Evolve hyperparameters (optional) 进化超参数
    else:
        # Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
        meta = {'lr0': (1, 1e-5, 1e-1),  # initial learning rate (SGD=1E-2, Adam=1E-3)
                'lrf': (1, 0.01, 1.0),  # final OneCycleLR learning rate (lr0 * lrf)
                'momentum': (0.3, 0.6, 0.98),  # SGD momentum/Adam beta1
                'weight_decay': (1, 0.0, 0.001),  # optimizer weight decay
                'warmup_epochs': (1, 0.0, 5.0),  # warmup epochs (fractions ok)
                'warmup_momentum': (1, 0.0, 0.95),  # warmup initial momentum
                'warmup_bias_lr': (1, 0.0, 0.2),  # warmup initial bias lr
                'box': (1, 0.02, 0.2),  # box loss gain
                'cls': (1, 0.2, 4.0),  # cls loss gain
                'cls_pw': (1, 0.5, 2.0),  # cls BCELoss positive_weight
                'obj': (1, 0.2, 4.0),  # obj loss gain (scale with pixels)
                'obj_pw': (1, 0.5, 2.0),  # obj BCELoss positive_weight
                'iou_t': (0, 0.1, 0.7),  # IoU training threshold
                'anchor_t': (1, 2.0, 8.0),  # anchor-multiple threshold
                'anchors': (2, 2.0, 10.0),  # anchors per output grid (0 to ignore)
                'fl_gamma': (0, 0.0, 2.0),  # focal loss gamma (efficientDet default gamma=1.5)
                'hsv_h': (1, 0.0, 0.1),  # image HSV-Hue augmentation (fraction)
                'hsv_s': (1, 0.0, 0.9),  # image HSV-Saturation augmentation (fraction)
                'hsv_v': (1, 0.0, 0.9),  # image HSV-Value augmentation (fraction)
                'degrees': (1, 0.0, 45.0),  # image rotation (+/- deg)
                'translate': (1, 0.0, 0.9),  # image translation (+/- fraction)
                'scale': (1, 0.0, 0.9),  # image scale (+/- gain)
                'shear': (1, 0.0, 10.0),  # image shear (+/- deg)
                'perspective': (0, 0.0, 0.001),  # image perspective (+/- fraction), range 0-0.001
                'flipud': (1, 0.0, 1.0),  # image flip up-down (probability)
                'fliplr': (0, 0.0, 1.0),  # image flip left-right (probability)
                'mosaic': (1, 0.0, 1.0),  # image mixup (probability)
                'mixup': (1, 0.0, 1.0)}  # image mixup (probability)

        assert opt.local_rank == -1, 'DDP mode not implemented for --evolve'
        opt.notest, opt.nosave = True, True  # only test/save final epoch
        # ei = [isinstance(x, (int, float)) for x in hyp.values()]  # evolvable indices
        yaml_file = Path(opt.save_dir) / 'hyp_evolved.yaml'  # save best result here
        if opt.bucket:
            os.system('gsutil cp gs://%s/evolve.txt .' % opt.bucket)  # download evolve.txt if exists

        for _ in range(300):  # generations to evolve
            if Path('evolve.txt').exists():  # if evolve.txt exists: select best hyps and mutate
                # Select parent(s)
                parent = 'single'  # parent selection method: 'single' or 'weighted'
                x = np.loadtxt('evolve.txt', ndmin=2)
                n = min(5, len(x))  # number of previous results to consider
                x = x[np.argsort(-fitness(x))][:n]  # top n mutations
                w = fitness(x) - fitness(x).min()  # weights
                if parent == 'single' or len(x) == 1:
                    # x = x[random.randint(0, n - 1)]  # random selection
                    x = x[random.choices(range(n), weights=w)[0]]  # weighted selection
                elif parent == 'weighted':
                    x = (x * w.reshape(n, 1)).sum(0) / w.sum()  # weighted combination

                # Mutate
                mp, s = 0.8, 0.2  # mutation probability, sigma
                npr = np.random
                npr.seed(int(time.time()))
                g = np.array([x[0] for x in meta.values()])  # gains 0-1
                ng = len(meta)
                v = np.ones(ng)
                while all(v == 1):  # mutate until a change occurs (prevent duplicates)
                    v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
                for i, k in enumerate(hyp.keys()):  # plt.hist(v.ravel(), 300)
                    hyp[k] = float(x[i + 7] * v[i])  # mutate

            # Constrain to limits
            for k, v in meta.items():
                hyp[k] = max(hyp[k], v[1])  # lower limit
                hyp[k] = min(hyp[k], v[2])  # upper limit
                hyp[k] = round(hyp[k], 5)  # significant digits

            # Train mutation
            results = train(hyp.copy(), opt, device)

            # Write mutation results
            print_mutation(hyp.copy(), results, yaml_file, opt.bucket)

        # Plot results
        plot_evolution(yaml_file)
        print(f'Hyperparameter evolution complete. Best results saved as: {yaml_file}\n'
              f'Command to train a new model with these hyperparameters: $ python train.py --hyp {yaml_file}')