【图像分类】学习率更新策略优化器

admin 管理员组

文章数量: 887007

【图像分类】学习率更新策略/优化器

项目地址：

项目说明：开箱即用，涵盖主流模型的图像分类|主干网络学习/对比/魔改项目

在models下的配置文件中，学习率更新方式以及参数均在lr_config中，在上一期讲网络如何被搭建时说了type是我们所需的调用关键字。在core/optimizers/lr_update.py中可以看到目前所支持的学习率更新策略。

常用的为StepLrUpdater与CosineAnnealingLrUpdater，均继承了LrUpdater类，先来看看它

class LrUpdater(object):"""LR Scheduler in MMCV.Args:by_epoch (bool): LR changes epoch by epochwarmup (string): Type of warmup used. It can be None(use no warmup),'constant', 'linear' or 'exp'warmup_iters (int): The number of iterations or epochs that warmuplastswarmup_ratio (float): LR used at the beginning of warmup equals towarmup_ratio * initial_lrwarmup_by_epoch (bool): When warmup_by_epoch == True, warmup_itersmeans the number of epochs that warmup lasts, otherwise means thenumber of iteration that warmup lasts"""def __init__(self,by_epoch=True,warmup=None,warmup_iters=0,warmup_ratio=0.1,warmup_by_epoch=False):# validate the "warmup" argumentif warmup is not None:if warmup not in ['constant', 'linear', 'exp']:raise ValueError(f'"{warmup}" is not a supported type for warming up, valid'' types are "constant" and "linear"')if warmup is not None:assert warmup_iters > 0, \'"warmup_iters" must be a positive integer'assert 0 < warmup_ratio <= 1.0, \'"warmup_ratio" must be in range (0,1]'self.by_epoch = by_epochself.warmup = warmupself.warmup_iters = warmup_itersself.warmup_ratio = warmup_ratioself.warmup_by_epoch = warmup_by_epochif self.warmup_by_epoch:self.warmup_epochs = self.warmup_itersself.warmup_iters = Noneelse:self.warmup_epochs = Noneself.base_lr = []  # initial lr for all param groupsself.regular_lr = []  # expected lr if no warming up is performed# 给optimizer设置新lrdef _set_lr(self, runner, lr_groups):for param_group, lr in zip(runner.get("optimizer").param_groups,lr_groups):param_group['lr'] = lrdef get_lr(self, runner, base_lr):raise NotImplementedError# 获取新lrdef get_regular_lr(self, runner):return [self.get_lr(runner, _base_lr) for _base_lr in self.base_lr]# 获取warmup学习率更新策略的lrdef get_warmup_lr(self, cur_iters):def _get_warmup_lr(cur_iters, regular_lr):if self.warmup == 'constant':warmup_lr = [_lr * self.warmup_ratio for _lr in regular_lr]elif self.warmup == 'linear':k = (1 - cur_iters / self.warmup_iters) * (1 -self.warmup_ratio)warmup_lr = [_lr * (1 - k) for _lr in regular_lr]elif self.warmup == 'exp':k = self.warmup_ratio**(1 - cur_iters / self.warmup_iters)warmup_lr = [_lr * k for _lr in regular_lr]return warmup_lrif isinstance(self.regular_lr, dict):lr_groups = {}for key, regular_lr in self.regular_lr.items():lr_groups[key] = _get_warmup_lr(cur_iters, regular_lr)return lr_groupselse:return _get_warmup_lr(cur_iters, self.regular_lr)# 记录初始lrdef before_run(self, runner):# NOTE: when resuming from a checkpoint, if 'initial_lr' is not saved,# it will be set according to the optimizer paramsfor group in runner.get("optimizer").param_groups:group.setdefault('initial_lr', group['lr'])self.base_lr = [group['initial_lr'] for group in runner.get("optimizer").param_groups]# 在周期更新前获取新lr并在optimizer中更新它def before_train_epoch(self, runner):if self.warmup_iters is None: # 即self.warmup_by_epoch为True，warmup_epochs = warmup_itersepoch_len = len(runner.get("train_loader"))self.warmup_iters = self.warmup_epochs * epoch_len # 按周期更新则warmup iters = warmup_epochs * datasets//batch size# 不按周期更新则没必要在此进行lr更新，在下一步before_train_iter中更新if not self.by_epoch:returnself.regular_lr = self.get_regular_lr(runner)self._set_lr(runner, self.regular_lr)# 首先判断是否按周期更新lr，若按迭代次数更新即by_epoch为False，大于等于warmup_iters使用正常lr更新方式，小于则用warmup方式更新lrdef before_train_iter(self, runner):cur_iter = runner.get("iter")if not self.by_epoch:self.regular_lr = self.get_regular_lr(runner)if self.warmup is None or cur_iter >= self.warmup_iters:self._set_lr(runner, self.regular_lr)else:warmup_lr = self.get_warmup_lr(cur_iter)self._set_lr(runner, warmup_lr)elif self.by_epoch:# 按周期更新lr，若当前迭代步数大于warmup迭代步数阈值，直接返回，使用before_train_epoch中的lrif self.warmup is None or cur_iter > self.warmup_iters:returnelif cur_iter == self.warmup_iters: # 等于则用常规方式self._set_lr(runner, self.regular_lr)else:warmup_lr = self.get_warmup_lr(cur_iter) # 小于则用warmup方式self._set_lr(runner, warmup_lr)

LrUpdater执行的顺序为：

1. __init__，初始化，主要针对warmup，需注意的是若warmup_by_epoch=True则warmup_iters代表的是warmup作用的epoch数，若warmup_by_epoch=False，则warmup_iters代表的是warmup作用的iters数，1 epoch = total images // batch size iters
2. before_run，在这里记录了初始学习率，通过获取配置文件optimizer_cfg中的lr定义初始学习率
3. before_train_epoch，在每个epoch更新前获取新lr并在optimizer中更新学习率。若by_epoch=False，即按iter更新学习率，直接返回。否则大于等于warmup_iters存在周期时使用常规lr更新方式
   1. get_regular_lr，调用配置文件中type指定的学习率更新方式，如StepLrUpdater与CosineAnnealingLrUpdater，通过get_lr获取当前epoch/iter下的学习率
   2. _set_lr给优化器设置最新的学习率
4. before_train_iter，在每个iter更新前获取新lr并在optimizer中更新学习率。首先判断是否按epoch更新lr
   1. 若by_epoch = False，即按iter更新，大于等于warmup_iters使用正常lr更新方式，小于则用warmup方式更新lr
   2. 若by_epoch = True，按epoch更新lr，若当前迭代步数大于warmup迭代步数阈值，直接返回，使用before_train_epoch中的lr，等于则用常规方式，小于则用warmup方式
   3. get_warmup_lr获取warmup学习率更新策略的lr

说完了核心LrUpdater类，再说说StepLrUpdater与CosineAnnealingLrUpdater，我们知道主要功能是根据当前iter或epoch计算最新学习率并返回，先看看StepLrUpdater

class StepLrUpdater(LrUpdater):"""Step LR scheduler with min_lr clipping.Args:step (int | list[int]): Step to decay the LR. If an int value is given,regard it as the decay interval. If a list is given, decay LR atthese steps.gamma (float, optional): Decay LR ratio. Default: 0.1.min_lr (float, optional): Minimum LR value to keep. If LR after decayis lower than `min_lr`, it will be clipped to this value. If Noneis given, we don't perform lr clipping. Default: None."""def __init__(self, step, gamma=0.1, min_lr=None, **kwargs):self.step = stepself.gamma = gammaself.min_lr = min_lrsuper(StepLrUpdater, self).__init__(**kwargs)def get_lr(self, runner, base_lr):progress = runner.get('epoch') if self.by_epoch else runner.get('iter')# calculate exponential termif isinstance(self.step, int):exp = progress // self.stepelse:exp = len(self.step)for i, s in enumerate(self.step):if progress < s:exp = ibreaklr = base_lr * (self.gamma**exp)if self.min_lr is not None:# clip to a minimum valuelr = max(lr, self.min_lr)return lr

step：步骤衰减LR，有两种形式，一是int，另一个是list。举例：

• step=2，即每2个iter/epoch更新一次学习率
• step=[20,60,80]，即0-20，21-60，61-80，80-X间的iter/epoch更新一次学习率

gamma：学习率衰减比率，对应公式lr = base_lr * (self.gamma**exp)

min_lr：最小学习率，若设置了该值，则当学习率小于该值时使用min_lr

再看看CosineAnnealingLrUpdater

class CosineAnnealingLrUpdater(LrUpdater):def __init__(self, min_lr=None, min_lr_ratio=None, **kwargs):assert (min_lr is None) ^ (min_lr_ratio is None)self.min_lr = min_lrself.min_lr_ratio = min_lr_ratiosuper(CosineAnnealingLrUpdater, self).__init__(**kwargs)def get_lr(self, runner, base_lr):if self.by_epoch:progress = runner.get('epoch')max_progress = runner.get('max_epochs')else:progress = runner.get('iter')max_progress = runner.get('max_iters')if self.min_lr_ratio is not None:target_lr = base_lr * self.min_lr_ratioelse:target_lr = self.min_lrreturn annealing_cos(base_lr, target_lr, progress / max_progress)

min_lr：最小学习率

min_lr_ratio：最小学习率比率，min_lr_ratio与min_lr不能同时存在，因为这两个参数仅是用于计算最终学习率，当设置了min_lr，则最终学习率为min_lr；若设置了min_lr_ratio，则最终学习率为base_lr * min_lr_ratio

annealing_cos，学习率更新计算方法，具体如下

def annealing_cos(start, end, factor, weight=1):"""Calculate annealing cos learning rate.Cosine anneal from `weight * start + (1 - weight) * end` to `end` aspercentage goes from 0.0 to 1.0.Args:start (float): The starting learning rate of the cosine annealing.end (float): The ending learing rate of the cosine annealing.factor (float): The coefficient of `pi` when calculating the currentpercentage. Range from 0.0 to 1.0.weight (float, optional): The combination factor of `start` and `end`when calculating the actual starting learning rate. Default to 1."""cos_out = cos(pi * factor) + 1return end + 0.5 * weight * (start - end) * cos_out

至此学习率更新介绍完毕，最后说下优化器是如何调用，对应配置文件中的optimizer_cfg，在上一期讲网络如何被搭建时说了type是我们所需的调用关键字，此处type与torch.optim中的优化器一一对应，也就是不需要额外重写优化器类。

在tools/train.py中可看到import torch.optim as optim，对应构建optimizer的语句为

optimizer = eval('optim.' + optimizer_cfg.pop('type'))(params=model.parameters(),**optimizer_cfg) 

还是借助eval，所以如果有修改优化器的需要，大家可以查阅pytorch手册自行修改。

以上就是本次学习率更新策略/优化器的全部内容，在后续可视化教程中会绘制学习率曲线

本文标签： 图像分类学习率更新策略优化器