task2/optimizers.py - master - akp1n/ADVML

task2/optimizers.py
160 строк5 KB
Artemiy
fix
24 фев 2025, 20:45
24 фев 2025, 20:4557cd6cd
import math
import torch
from torch.optim import Optimizer


class MyNesterov(Optimizer):

    def __init__(self, params, lr=0.01, momentum=0.9, weight_decay=0.0):
        if lr <= 0.0:
            raise ValueError(f"Invalid learning rate: {lr}")
        if momentum < 0.0:
            raise ValueError(f"Invalid momentum value: {momentum}")

        defaults = dict(lr=lr, momentum=momentum, weight_decay=weight_decay)
        super().__init__(params, defaults)

    def step(self, closure=None):

        loss = None
        if closure is not None:
            loss = closure()

        for group in self.param_groups:
            lr = group['lr']
            momentum = group['momentum']
            weight_decay = group['weight_decay']

            for p in group['params']:
                if p.grad is None:
                    continue
                grad = p.grad.data

                if weight_decay != 0:
                    grad = grad.add(p.data, alpha=weight_decay)

                param_state = self.state[p]
                if 'momentum_buffer' not in param_state:
                    buf = param_state['momentum_buffer'] = torch.clone(grad).detach()
                else:
                    buf = param_state['momentum_buffer']

                    buf.mul_(momentum).add_(grad)

                nesterov_grad = grad.add(buf, alpha=momentum)
                p.data.add_(nesterov_grad, alpha=-lr)

        return loss


class MyAdam(Optimizer):

    def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-8, weight_decay=0.0):
        if lr <= 0.0:
            raise ValueError(f"Invalid learning rate: {lr}")
        if eps <= 0.0:
            raise ValueError(f"Invalid epsilon value: {eps}")

        defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay)
        super().__init__(params, defaults)

    def step(self, closure=None):
        loss = None
        if closure is not None:
            loss = closure()

        for group in self.param_groups:
            lr = group['lr']
            beta1, beta2 = group['betas']
            eps = group['eps']
            weight_decay = group['weight_decay']

            for p in group['params']:
                if p.grad is None:
                    continue
                grad = p.grad.data

                # Weight decay
                if weight_decay != 0:
                    grad = grad.add(p.data, alpha=weight_decay)

                state = self.state[p]

                if 'step' not in state:
                    state['step'] = 0
                if 'exp_avg' not in state:
                    state['exp_avg'] = torch.zeros_like(p.data)
                if 'exp_avg_sq' not in state:
                    state['exp_avg_sq'] = torch.zeros_like(p.data)

                exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']

                state['step'] += 1
                t = state['step']

                exp_avg.mul_(beta1).add_(grad, alpha=(1 - beta1))
                exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=(1 - beta2))

                bias_correction1 = 1 - beta1 ** t
                bias_correction2 = 1 - beta2 ** t

                denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(eps)
                step_size = lr * math.sqrt(bias_correction2) / bias_correction1

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

        return loss


class MyRMSProp(Optimizer):

    def __init__(self, params, lr=1e-2, alpha=0.99, eps=1e-8, weight_decay=0.0, momentum=0.0):
        if lr <= 0.0:
            raise ValueError(f"Invalid learning rate: {lr}")
        if alpha < 0.0 or alpha >= 1.0:
            raise ValueError(f"Invalid alpha: {alpha}")

        defaults = dict(lr=lr, alpha=alpha, eps=eps,
                        weight_decay=weight_decay, momentum=momentum)
        super().__init__(params, defaults)

    def step(self, closure=None):
        loss = None
        if closure is not None:
            loss = closure()

        for group in self.param_groups:
            lr = group['lr']
            alpha = group['alpha']
            eps = group['eps']
            weight_decay = group['weight_decay']
            momentum = group['momentum']

            for p in group['params']:
                if p.grad is None:
                    continue
                grad = p.grad.data

                if weight_decay != 0:
                    grad = grad.add(p.data, alpha=weight_decay)

                state = self.state[p]

                if 'square_avg' not in state:
                    state['square_avg'] = torch.zeros_like(p.data)
                if momentum > 0.0 and 'momentum_buffer' not in state:
                    state['momentum_buffer'] = torch.zeros_like(p.data)

                square_avg = state['square_avg']
                square_avg.mul_(alpha).addcmul_(grad, grad, value=1 - alpha)

                avg = square_avg.sqrt().add_(eps)

                if momentum > 0.0:
                    buf = state['momentum_buffer']
                    buf.mul_(momentum).addcdiv_(grad, avg)
                    p.data.add_(buf, alpha=-lr)
                else:
                    p.data.addcdiv_(grad, avg, value=-lr)

        return loss