scheduler.py

#!/usr/bin/env python3.7

from typing import Any, Callable, List, Tuple
from operator import add
from functools import partial

from utils import map_, uc_


class DummyScheduler(object):
    def __call__(self, epoch: int, optimizer: Any, loss_fns: List[List[Callable]], loss_weights: List[List[float]]) \
            -> Tuple[float, List[List[Callable]], List[List[float]]]:
        return optimizer, loss_fns, loss_weights


class AddWeightLoss():
    def __init__(self, to_add: List[float]):
        self.to_add: List[float] = to_add

    def __call__(self, epoch: int, optimizer: Any, loss_fns: List[List[Callable]], loss_weights: List[List[float]]) \
            -> Tuple[float, List[List[Callable]], List[List[float]]]:
        assert len(self.to_add) == len(loss_weights)
        new_weights: List[List[float]] = map_(lambda w: map_(uc_(add), zip(w, self.to_add)), loss_weights)

        print(f"Loss weights went from {loss_weights} to {new_weights}")

        return optimizer, loss_fns, new_weights


class StealWeight():
    def __init__(self, to_steal: float):
        self.to_steal: float = to_steal

    def __call__(self, epoch: int, optimizer: Any, loss_fns: List[List[Callable]], loss_weights: List[List[float]]) \
            -> Tuple[float, List[List[Callable]], List[List[float]]]:
        new_weights: List[List[float]] = [[max(0.1, a - self.to_steal), b + self.to_steal] for a, b in loss_weights]

        print(f"Loss weights went from {loss_weights} to {new_weights}")

        return optimizer, loss_fns, new_weights


class MultiplyT():
    def __init__(self, target_loss: str, mu: float):
        self.target_loss: str = target_loss
        self.mu: float = mu

    def __call__(self, epoch: int, optimizer: Any, loss_fns: List[List[Callable]], loss_weights: List[List[float]]) \
            -> Tuple[float, List[List[Callable]], List[List[float]]]:
        def update(loss: Any):
            if loss.__class__.__name__ == self.target_loss:
                loss.t *= self.mu

            return loss

        return optimizer, map_(lambda l: map_(update, l), loss_fns), loss_weights