Source code for catalyst.contrib.nn.optimizers.ralamb

from typing import Callable, Iterable, Optional, Tuple
import math

import torch
from torch.optim.optimizer import Optimizer


class Ralamb(Optimizer):
    """RAdam optimizer with LARS/LAMB tricks.

    Adapted from:
    https://github.com/mgrankin/over9000/blob/master/ralamb.py
    (Apache-2.0 License)
    """

    def __init__(
        self,
        params: Iterable,
        lr: float = 1e-3,
        betas: Tuple[float, float] = (0.9, 0.999),
        eps: float = 1e-8,
        weight_decay: float = 0,
    ):
        """
        Args:
            params: 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, 0.999))
            eps (float, optional): term added to the denominator to improve
                numerical stability (default: 1e-8)
            weight_decay (float, optional): weight decay
                (L2 penalty) (default: 0)
        """
        defaults = {
            "lr": lr,
            "betas": betas,
            "eps": eps,
            "weight_decay": weight_decay,
        }
        self.buffer = [[None, None, None] for ind in range(10)]
        super(Ralamb, self).__init__(params, defaults)

    def __setstate__(self, state):
        """Sets state."""
        super(Ralamb, self).__setstate__(state)

    def step(self, closure: Optional[Callable] = None):
        """Makes optimizer step.

        Args:
            closure (callable, optional): A closure that reevaluates
                the model and returns the loss.

        Returns:
            computed loss

        Raises:
            RuntimeError: Ralamb does not support sparse gradients
        """
        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.float()
                if grad.is_sparse:
                    raise RuntimeError("Ralamb does not support sparse gradients")

                p_data_fp32 = p.data.float()

                state = self.state[p]

                if len(state) == 0:
                    state["step"] = 0
                    state["exp_avg"] = torch.zeros_like(p_data_fp32)
                    state["exp_avg_sq"] = torch.zeros_like(p_data_fp32)
                else:
                    state["exp_avg"] = state["exp_avg"].type_as(p_data_fp32)
                    state["exp_avg_sq"] = state["exp_avg_sq"].type_as(p_data_fp32)

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

                # Decay the first and second moment running average coefficient
                # m_t
                exp_avg.mul_(beta1).add_(1 - beta1, grad)
                # v_t
                exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)

                state["step"] += 1
                buffered = self.buffer[int(state["step"] % 10)]

                if state["step"] == buffered[0]:
                    n_sma, radam_step_size = buffered[1], buffered[2]
                else:
                    buffered[0] = state["step"]
                    beta2_t = beta2 ** state["step"]
                    n_sma_max = 2 / (1 - beta2) - 1
                    n_sma = n_sma_max - 2 * state["step"] * beta2_t / (1 - beta2_t)
                    buffered[1] = n_sma

                    # more conservative since it"s an approximated value
                    if n_sma >= 5:
                        radam_step_size = math.sqrt(
                            (1 - beta2_t)
                            * (n_sma - 4)
                            / (n_sma_max - 4)
                            * (n_sma - 2)
                            / n_sma
                            * n_sma_max
                            / (n_sma_max - 2)
                        ) / (1 - beta1 ** state["step"])
                    else:
                        radam_step_size = 1.0 / (1 - beta1 ** state["step"])
                    buffered[2] = radam_step_size

                if group["weight_decay"] != 0:
                    p_data_fp32.add_(-group["weight_decay"] * group["lr"], p_data_fp32)

                # more conservative since it"s an approximated value
                radam_step = p_data_fp32.clone()
                if n_sma >= 5:
                    denom = exp_avg_sq.sqrt().add_(group["eps"])
                    radam_step.addcdiv_(-radam_step_size * group["lr"], exp_avg, denom)
                else:
                    radam_step.add_(-radam_step_size * group["lr"], exp_avg)

                radam_norm = radam_step.pow(2).sum().sqrt()
                weight_norm = p.data.pow(2).sum().sqrt().clamp(0, 10)
                if weight_norm == 0 or radam_norm == 0:
                    trust_ratio = 1
                else:
                    trust_ratio = weight_norm / radam_norm

                state["weight_norm"] = weight_norm
                state["adam_norm"] = radam_norm
                state["trust_ratio"] = trust_ratio

                if n_sma >= 5:
                    p_data_fp32.addcdiv_(
                        -radam_step_size * group["lr"] * trust_ratio, exp_avg, denom,
                    )
                else:
                    p_data_fp32.add_(-radam_step_size * group["lr"] * trust_ratio, exp_avg)

                p.data.copy_(p_data_fp32)

        return loss


__all__ = ["Ralamb"]