shared_optim.py

from __future__ import division
import math
import torch
import torch.optim as optim


class SharedRMSprop(optim.RMSprop):
    """Implements RMSprop algorithm with shared states.
    """

    def __init__(self,
                 params,
                 lr=7e-4,
                 alpha=0.99,
                 eps=0.1,
                 weight_decay=0,
                 momentum=0,
                 centered=False):
        super(SharedRMSprop, self).__init__(params, lr, alpha, eps,
                                            weight_decay, momentum, centered)

        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['step'] = torch.zeros(1)
                state['grad_avg'] = p.data.new().resize_as_(p.data).zero_()
                state['square_avg'] = p.data.new().resize_as_(p.data).zero_()
                state['momentum_buffer'] = p.data.new().resize_as_(
                    p.data).zero_()

    def share_memory(self):
        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['square_avg'].share_memory_()
                state['step'].share_memory_()
                state['grad_avg'].share_memory_()

    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
                grad = p.grad.data
                state = self.state[p]

                square_avg = state['square_avg']
                alpha = group['alpha']

                state['step'] += 1

                if group['weight_decay'] != 0:
                    grad = grad.add(group['weight_decay'], p.data)

                square_avg.mul_(alpha).addcmul_(1 - alpha, grad, grad)

                if group['centered']:
                    grad_avg = state['grad_avg']
                    grad_avg.mul_(alpha).add_(1 - alpha, grad)
                    avg = square_avg.addcmul(
                        -1, grad_avg, grad_avg).sqrt().add_(group['eps'])
                else:
                    avg = square_avg.sqrt().add_(group['eps'])

                if group['momentum'] > 0:
                    buf = state['momentum_buffer']
                    buf.mul_(group['momentum']).addcdiv_(grad, avg)
                    p.data.add_(-group['lr'], buf)
                else:
                    p.data.addcdiv_(-group['lr'], grad, avg)

        return loss


class SharedAdam(optim.Adam):
    """Implements Adam algorithm with shared states.
    """

    def __init__(self,
                 params,
                 lr=1e-3,
                 betas=(0.9, 0.999),
                 eps=1e-3,
                 weight_decay=0):
        super(SharedAdam, self).__init__(params, lr, betas, eps, weight_decay)

        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['step'] = torch.zeros(1)
                state['exp_avg'] = p.data.new().resize_as_(p.data).zero_()
                state['exp_avg_sq'] = p.data.new().resize_as_(p.data).zero_()

    def share_memory(self):
        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['step'].share_memory_()
                state['exp_avg'].share_memory_()
                state['exp_avg_sq'].share_memory_()

    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
                grad = p.grad.data
                state = self.state[p]

                exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
                beta1, beta2 = group['betas']

                state['step'] += 1

                if group['weight_decay'] != 0:
                    grad = grad.add(group['weight_decay'], p.data)

                # 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)

                denom = exp_avg_sq.sqrt().add_(group['eps'])

                bias_correction1 = 1 - beta1**state['step'][0]
                bias_correction2 = 1 - beta2**state['step'][0]
                step_size = group['lr'] * \
                    math.sqrt(bias_correction2) / bias_correction1

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

        return loss


sample_lr = [
    0.0001, 0.00009, 0.00008, 0.00007, 0.00006, 0.00005, 0.00004, 0.00003,
    0.00002, 0.00001, 0.000009, 0.000008, 0.000007, 0.000006, 0.000005,
    0.000004, 0.000003, 0.000002, 0.000001
]


class SharedLrSchedAdam(optim.Adam):
    """Implements Adam algorithm with shared states.
    """

    def __init__(self,
                 params,
                 lr=1e-3,
                 betas=(0.9, 0.999),
                 eps=1e-3,
                 weight_decay=0):
        super(SharedLrSchedAdam, self).__init__(params, lr, betas, eps,
                                                weight_decay)

        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['step'] = torch.zeros(1)
                state['exp_avg'] = p.data.new().resize_as_(p.data).zero_()
                state['exp_avg_sq'] = p.data.new().resize_as_(p.data).zero_()

    def share_memory(self):
        for group in self.param_groups:
            for p in group['params']:
                state = self.state[p]
                state['step'].share_memory_()
                state['exp_avg'].share_memory_()
                state['exp_avg_sq'].share_memory_()

    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()

        lr = sample_lr[int(state['step'][0] // 40000000)]
        group['lr'] = lr

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

                exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
                beta1, beta2 = group['betas']

                state['step'] += 1

                if group['weight_decay'] != 0:
                    grad = grad.add(group['weight_decay'], p.data)

                # 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)

                denom = exp_avg_sq.sqrt().add_(group['eps'])

                bias_correction1 = 1 - beta1**state['step'][0]
                bias_correction2 = 1 - beta2**state['step'][0]
                step_size = group['lr'] * \
                    math.sqrt(bias_correction2) / bias_correction1
                p.data.addcdiv_(-step_size, exp_avg, denom)

        return loss