fastNLP/fastNLP/core/optimizer.py

"""
optimizer 模块定义了 fastNLP 中所需的各种优化器，一般做为 :class:`~fastNLP.Trainer` 的参数使用。

"""
__all__ = [
    "Optimizer",
    "SGD",
    "Adam",
    "AdamW"
]

import torch
import math
import torch
from torch.optim.optimizer import Optimizer as TorchOptimizer


class Optimizer(object):
    """
    别名：:class:`fastNLP.Optimizer` :class:`fastNLP.core.optimizer.Optimizer`

    :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models.
    :param kwargs: additional parameters.
    """
    
    def __init__(self, model_params, **kwargs):
        if model_params is not None and not hasattr(model_params, "__next__"):
            raise RuntimeError("model parameters should be a generator, rather than {}.".format(type(model_params)))
        self.model_params = model_params
        self.settings = kwargs
    
    def construct_from_pytorch(self, model_params):
        raise NotImplementedError
    
    def _get_require_grads_param(self, params):
        """
        将params中不需要gradient的删除
        :param iterable params: parameters
        :return: list(nn.Parameters)
        """
        return [param for param in params if param.requires_grad]

class NullOptimizer(Optimizer):
    """
    当不希望Trainer更新optimizer时，传入本optimizer，但请确保通过callback的方式对参数进行了更新。

    """
    def __init__(self):
        super().__init__(None)

    def construct_from_pytorch(self, model_params):
        pass

    def __getattr__(self, item):
        def pass_func(*args, **kwargs):
            pass

        return pass_func


class SGD(Optimizer):
    """
    别名：:class:`fastNLP.SGD` :class:`fastNLP.core.optimizer.SGD`

    :param float lr: learning rate. Default: 0.01
    :param float momentum: momentum. Default: 0
    :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models.
    """
    
    def __init__(self, lr=0.001, momentum=0, model_params=None):
        if not isinstance(lr, float):
            raise TypeError("learning rate has to be float.")
        super(SGD, self).__init__(model_params, lr=lr, momentum=momentum)
    
    def construct_from_pytorch(self, model_params):
        if self.model_params is None:
            # careful! generator cannot be assigned.
            return torch.optim.SGD(self._get_require_grads_param(model_params), **self.settings)
        else:
            return torch.optim.SGD(self._get_require_grads_param(self.model_params), **self.settings)


class Adam(Optimizer):
    """
    别名：:class:`fastNLP.Adam` :class:`fastNLP.core.optimizer.Adam`

    :param float lr: learning rate
    :param float weight_decay:
    :param model_params: a generator. E.g. ``model.parameters()`` for PyTorch models.
    """
    
    def __init__(self, lr=0.001, weight_decay=0, betas=(0.9, 0.999), eps=1e-8, amsgrad=False, model_params=None):
        if not isinstance(lr, float):
            raise TypeError("learning rate has to be float.")
        super(Adam, self).__init__(model_params, lr=lr, betas=betas, eps=eps, amsgrad=amsgrad,
                                   weight_decay=weight_decay)
    
    def construct_from_pytorch(self, model_params):
        if self.model_params is None:
            # careful! generator cannot be assigned.
            return torch.optim.Adam(self._get_require_grads_param(model_params), **self.settings)
        else:
            return torch.optim.Adam(self._get_require_grads_param(self.model_params), **self.settings)


class AdamW(TorchOptimizer):
    r"""对AdamW的实现，该实现应该会在pytorch更高版本中出现，https://github.com/pytorch/pytorch/pull/21250。这里提前加入
    
    .. todo::
        翻译成中文
    
    The original Adam algorithm was proposed in `Adam: A Method for Stochastic Optimization`_.
    The AdamW variant was proposed in `Decoupled Weight Decay Regularization`_.
    Arguments:
        params (iterable): iterable of parameters to optimize or dicts defining
            parameter groups
        lr (float, optional): learning rate (default: 1e-3)
        betas (Tuple[float, float], optional): coefficients used for computing
            running averages of gradient and its square (default: (0.9, 0.99))
        eps (float, optional): term added to the denominator to improve
            numerical stability (default: 1e-8)
        weight_decay (float, optional): weight decay coefficient (default: 1e-2)
        amsgrad (boolean, optional): whether to use the AMSGrad variant of this
            algorithm from the paper `On the Convergence of Adam and Beyond`_
            (default: False)
    .. _Adam\: A Method for Stochastic Optimization:
        https://arxiv.org/abs/1412.6980
    .. _Decoupled Weight Decay Regularization:
        https://arxiv.org/abs/1711.05101
    .. _On the Convergence of Adam and Beyond:
        https://openreview.net/forum?id=ryQu7f-RZ
    """

    def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8,
                 weight_decay=1e-2, amsgrad=False):
        if not 0.0 <= lr:
            raise ValueError("Invalid learning rate: {}".format(lr))
        if not 0.0 <= eps:
            raise ValueError("Invalid epsilon value: {}".format(eps))
        if not 0.0 <= betas[0] < 1.0:
            raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
        if not 0.0 <= betas[1] < 1.0:
            raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
        defaults = dict(lr=lr, betas=betas, eps=eps,
                        weight_decay=weight_decay, amsgrad=amsgrad)
        super(AdamW, self).__init__(params, defaults)

    def __setstate__(self, state):
        super(AdamW, self).__setstate__(state)
        for group in self.param_groups:
            group.setdefault('amsgrad', False)

    def step(self, closure=None):
        """Performs a single optimization step.
        Arguments:
            closure (callable, optional): A closure that reevaluates the model
                and returns the loss.
        """
        loss = None
        if closure is not None:
            loss = closure()

        for group in self.param_groups:
            for p in group['params']:
                if p.grad is None:
                    continue

                # Perform stepweight decay
                p.data.mul_(1 - group['lr'] * group['weight_decay'])

                # Perform optimization step
                grad = p.grad.data
                if grad.is_sparse:
                    raise RuntimeError('Adam does not support sparse gradients, please consider SparseAdam instead')
                amsgrad = group['amsgrad']

                state = self.state[p]

                # State initialization
                if len(state) == 0:
                    state['step'] = 0
                    # Exponential moving average of gradient values
                    state['exp_avg'] = torch.zeros_like(p.data)
                    # Exponential moving average of squared gradient values
                    state['exp_avg_sq'] = torch.zeros_like(p.data)
                    if amsgrad:
                        # Maintains max of all exp. moving avg. of sq. grad. values
                        state['max_exp_avg_sq'] = torch.zeros_like(p.data)

                exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
                if amsgrad:
                    max_exp_avg_sq = state['max_exp_avg_sq']
                beta1, beta2 = group['betas']

                state['step'] += 1

                # Decay the first and second moment running average coefficient
                exp_avg.mul_(beta1).add_(1 - beta1, grad)
                exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
                if amsgrad:
                    # Maintains the maximum of all 2nd moment running avg. till now
                    torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
                    # Use the max. for normalizing running avg. of gradient
                    denom = max_exp_avg_sq.sqrt().add_(group['eps'])
                else:
                    denom = exp_avg_sq.sqrt().add_(group['eps'])

                bias_correction1 = 1 - beta1 ** state['step']
                bias_correction2 = 1 - beta2 ** state['step']
                step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1

                p.data.addcdiv_(-step_size, exp_avg, denom)

        return loss