from __future__ import annotations

import torch
from torch import nn


def mlp(outputs: int) -> nn.Sequential:
    return nn.Sequential(
        nn.Linear(2, 64),
        nn.Tanh(),
        nn.Linear(64, 64),
        nn.Tanh(),
        nn.Linear(64, outputs),
    )


class HamiltonianNetwork(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.energy = mlp(1)

    def forward(
        self, states: torch.Tensor, *, create_graph: bool = True
    ) -> torch.Tensor:
        if not states.requires_grad:
            states = states.requires_grad_(True)
        hamiltonian = self.energy(states).sum()
        gradient = torch.autograd.grad(
            hamiltonian,
            states,
            create_graph=create_graph,
        )[0]
        return torch.stack([gradient[:, 1], -gradient[:, 0]], dim=1)


class VectorFieldNetwork(nn.Module):
    def __init__(self) -> None:
        super().__init__()
        self.network = mlp(2)

    def forward(self, states: torch.Tensor) -> torch.Tensor:
        return self.network(states)


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