import importlib from typing import Callable import typeguard from hydra.utils import instantiate from jaxtyping import Float, jaxtyped from omegaconf import DictConfig from pdeinvbench.utils.types import ( HIGH_RESOLUTION_PDE_SPATIAL_SIZE, PDE, PDE_SPATIAL_SIZE, ) @jaxtyped(typechecker=typeguard.typechecked) def get_function_from_string(string: str) -> Callable: """ Converts a function specified as a string to an actual function object. Used in hydra configs. """ module_name, function_name = string.rsplit(".", 1) # Import the module dynamically module = importlib.import_module(module_name) # Get the function object function = getattr(module, function_name) return function def resolve_pde_resolution(cfg: DictConfig) -> None: """ Simple utility method which checks if we are using the high resolution data. If we are, it updates the types::PDE_SPATIAL_SIZE dict. Currently only works with the inverse setting. Assumes keys cfg:high_resolution[bool] and cfg.data.downsample_factor[int] exist. """ assert "high_resolution" in cfg, "No key 'high_resolution' found in hydra config" assert ( "data" in cfg and "downsample_factor" in cfg.data ), "No key 'data' or 'data.downsample_factor' found in hydra config" high_resolution: bool = cfg.high_resolution downsample_factor: int = cfg.data.downsample_factor pde: PDE = instantiate(cfg.data.pde) if high_resolution: assert ( pde in HIGH_RESOLUTION_PDE_SPATIAL_SIZE ), f"Could not find {pde} in high resolution PDE size mapping." resolution: list[int] = ( HIGH_RESOLUTION_PDE_SPATIAL_SIZE[pde] if high_resolution else PDE_SPATIAL_SIZE[pde] ) if ( downsample_factor == 0 ): # Ensures that dynamic setting works without downsampling downsample_factor = 1 new_resolution: list[float] = [res / downsample_factor for res in resolution] # only allow downsampling to an integer factor for res in new_resolution: assert ( int(res) == res ), f"Downsample factor leads to non-integer resolution {res}" new_resolution: list[int] = [int(res) for res in new_resolution] PDE_SPATIAL_SIZE[pde] = new_resolution