source/model.py

from __future__ import annotations

import torch
from torch import nn


class TuringSurrogate(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.network = nn.Sequential(
            nn.Conv2d(4, 32, kernel_size=3, padding=1, padding_mode="circular"),
            nn.SiLU(),
            nn.Conv2d(32, 32, kernel_size=3, padding=1, padding_mode="circular"),
            nn.SiLU(),
            nn.Conv2d(32, 16, kernel_size=3, padding=1, padding_mode="circular"),
            nn.SiLU(),
            nn.Conv2d(16, 2, kernel_size=1),
        )

    def forward(
        self,
        state: torch.Tensor,
        feed: torch.Tensor,
        kill: torch.Tensor,
    ) -> torch.Tensor:
        batch, _, height, width = state.shape
        parameters = torch.stack([feed, kill], dim=1)
        parameter_fields = parameters[:, :, None, None].expand(
            batch,
            2,
            height,
            width,
        )
        delta = self.network(torch.cat([state, parameter_fields], dim=1))
        return torch.clamp(state + delta, 0, 1)


def parameter_count(model: nn.Module) -> int:
    return sum(parameter.numel() for parameter in model.parameters())