aurora/model/decoder.py

"""Copyright (c) Microsoft Corporation. Licensed under the MIT license."""

from datetime import timedelta

import torch
from einops import rearrange
from torch import nn

from aurora.batch import Batch, Metadata
from aurora.model.fourier import levels_expansion
from aurora.model.perceiver import PerceiverResampler
from aurora.model.util import (
    check_lat_lon_dtype,
    init_weights,
    unpatchify,
)

__all__ = ["Perceiver3DDecoder"]


class Perceiver3DDecoder(nn.Module):
    """Multi-scale multi-source multi-variable decoder based on the Perceiver architecture."""

    def __init__(
        self,
        out_surf_vars: tuple[str, ...],
        out_atmos_vars: tuple[str, ...],
        patch_size: int = 4,
        embed_dim: int = 1024,
        depth: int = 1,
        head_dim: int = 64,
        num_heads: int = 8,
        mlp_ratio: float = 4.0,
        drop_rate: float = 0.0,
        perceiver_ln_eps: float = 1e-5,
    ) -> None:
        """Initialise.

        Args:
            surf_vars (tuple[str, ...]): All supported surface-level variables.
            atmos_vars (tuple[str, ...]): All supported atmospheric variables.
            patch_size (int, optional): Patch size. Defaults to `4`.
            embed_dim (int, optional): Embedding dim.. Defaults to `1024`.
            depth (int, optional): Number of Perceiver cross-attention and feed-forward blocks.
                Defaults to `1`.
            head_dim (int, optional): Dimension of the attention heads used in the aggregation
                blocks. Defaults to `64`.
            num_heads (int, optional): Number of attention heads used in the aggregation blocks.
                Defaults to `8`.
            mlp_ratio (float, optional): Ratio of MLP hidden dimension to embedding dimensionality.
                Defaults to `4.0`.
            drop_rate (float, optional): Drop-out rate for input patches. Defaults to `0.0`.
            perceiver_ln_eps (float, optional): Layer norm. epsilon for the Perceiver blocks.
                Defaults to `1e-5`.
        """
        super().__init__()

        self.patch_size = patch_size

        self.embed_dim = embed_dim
        self.out_surf_vars = out_surf_vars
        self.out_atmos_vars = out_atmos_vars
        
        if out_surf_vars:
            self.surf_heads = nn.ParameterDict(
                {name: nn.Linear(embed_dim, patch_size**2) for name in out_surf_vars}
            )
        
        if out_atmos_vars:
            self.atmos_heads = nn.ParameterDict(
                {name: nn.Linear(embed_dim, patch_size**2) for name in out_atmos_vars}
            )

            self.level_decoder = PerceiverResampler(
                latent_dim=embed_dim,
                context_dim=embed_dim,
                depth=depth,
                head_dim=head_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                drop=drop_rate,
                residual_latent=True,
                ln_eps=perceiver_ln_eps,
            )

            self.atmos_levels_embed = nn.Linear(embed_dim, embed_dim)

        self.apply(init_weights)

    def deaggregate_levels(self, level_embed: torch.Tensor, x: torch.Tensor) -> torch.Tensor:
        """Deaggregate pressure level information.

        Args:
            level_embed (torch.Tensor): Level embedding of shape `(B, L, C, D)`.
            x (torch.Tensor): Aggregated input of shape `(B, L, C', D)`.

        Returns:
            torch.Tensor: Deaggregate output of shape `(B, L, C, D)`.
        """
        B, L, C, D = level_embed.shape
        level_embed = level_embed.flatten(0, 1)  # (BxL, C, D)
        x = x.flatten(0, 1)  # (BxL, C', D)
        _msg = f"Batch size mismatch. Found {level_embed.size(0)} and {x.size(0)}."
        assert level_embed.size(0) == x.size(0), _msg
        assert len(level_embed.shape) == 3, f"Expected 3 dims, found {level_embed.dims()}."
        assert x.dim() == 3, f"Expected 3 dims, found {x.dim()}."

        x = self.level_decoder(level_embed, x)  # (BxL, C, D)
        x = x.reshape(B, L, C, D)
        return x

    def forward(
        self,
        x: torch.Tensor,
        batch: Batch,
        patch_res: tuple[int, int, int],
        lead_time: timedelta,
    ) -> Batch:
        """Forward pass of MultiScaleEncoder.

        Args:
            x (torch.Tensor): Backbone output of shape `(B, L, D)`.
            batch (:class:`aurora.batch.Batch`): Batch to make predictions for.
            patch_res (tuple[int, int, int]): Patch resolution
            lead_time (timedelta): Lead time.

        Returns:
            :class:`aurora.batch.Batch`: Prediction for `batch`.
        """
        surf_vars = self.out_surf_vars
        atmos_vars = self.out_atmos_vars
        atmos_levels = batch.metadata.atmos_levels

        # Compress the latent dimension from the U-net skip concatenation.
        B, L, D = x.shape

        # Extract the lat, lon and convert to float32.
        lat, lon = batch.metadata.lat, batch.metadata.lon
        check_lat_lon_dtype(lat, lon)
        lat, lon = lat.to(dtype=torch.float32), lon.to(dtype=torch.float32)
        H, W = lat.shape[0], lon.shape[-1]

        # Unwrap the latent level dimension.
        x = rearrange(
            x,
            "B (C H W) D -> B (H W) C D",
            C=patch_res[0],
            H=patch_res[1],
            W=patch_res[2],
        )
        surf_preds = None
        atmos_preds = None
        
        if surf_vars:
            # Decode surface vars. Run the head for every surface-level variable.
            x_surf = torch.stack([self.surf_heads[name](x[..., :1, :]) for name in surf_vars], dim=-1)
            x_surf = x_surf.reshape(*x_surf.shape[:3], -1)  # (B, L, 1, V_S*p*p)
            surf_preds = unpatchify(x_surf, len(surf_vars), H , W, self.patch_size)
            surf_preds = surf_preds.squeeze(2)  # (B, V_S, H, W)

        if atmos_vars:
            # Embed the atmospheric levels.
            atmos_levels_encode = levels_expansion(
                torch.tensor(atmos_levels, device=x.device), self.embed_dim
            ).to(dtype=x.dtype)
            levels_embed = self.atmos_levels_embed(atmos_levels_encode)  # (C_A, D)

            # De-aggregate the hidden levels into the physical levels.
            levels_embed = levels_embed.expand(B, x.size(1), -1, -1)
            x_atmos = self.deaggregate_levels(levels_embed, x[..., 1:, :])  # (B, L, C_A, D)

            # Decode the atmospheric vars.
            x_atmos = torch.stack([self.atmos_heads[name](x_atmos) for name in atmos_vars], dim=-1)
            x_atmos = x_atmos.reshape(*x_atmos.shape[:3], -1)  # (B, L, C_A, V_A*p*p)
            atmos_preds = unpatchify(x_atmos, len(atmos_vars), H, W, self.patch_size)
        
        return surf_preds, atmos_preds