fzimmermann89

class AdjointGridSample(torch.autograd.Function):
    @staticmethod
    def forward(
        ctx: torch.autograd.function.FunctionCtx,
        y: torch.Tensor,
        grid: torch.Tensor,
        xshape: Sequence[int],
        interpolation_mode: Literal['bilinear', 'nearest', 'bicubic'] = 'bilinear',
        padding_mode: Literal['zeros', 'border', 'reflection'] = 'zeros',
        align_corners: bool = True,

fzimmermann89

import torch
from scipy.spatial.transform import Rotation
from typing import Callable, Literal
import einops
import itertools
from torch import Tensor


class MatrixMultiplication(torch.autograd.Function):
    """Helper to do a matrix multiplication if we know the adjoint of the matrix"""

fzimmermann89

import dataclasses
from typing import Type, Callable, Any


def _rapply(obj:Any, functions_per_types: dict[Type | list[Type], Callable]):
    """ Apply callables to all fields of a dataclass and keys of a dictionary, recursively.
    Which callable is used is determined by the type. By default, the function will recurse into dataclasses.fields and dictionaries, respectively, returning new dataclass instances or dictionaries, respectively. This can be disabled by, for example, adding {dict:lambda x:x} to functions_per_types, as these take precedence.

    Example:
    >>> _rapply({'a':1, 'string':'not changed'}, {int:lambda x:x+1})

fzimmermann89

import math
from pathlib import Path
from datetime import datetime
import pypulseq as pp

### SETTINGS ###

timestamp = datetime.now().strftime("%y%m%d-%H%M%S")
filename = Path(f"profile_{timestamp}.seq")

fzimmermann89

import warnings
import torch
import numpy as np

def _filter_separable(x, kernels, axis):
    """Apply the separable filter kernels to the tensor x along the axes axis.

    Does zero-padding to keep the output the same size as the input.

    Parameters

fzimmermann89

import warnings
from typing import Sequence
import torch
import numpy as np
# fzimmermann89, felix.zimmermann@ptb.de, 2024
def sliding_window(x:torch.Tensor, window_shape:int|Sequence[int], axis:None|int|Sequence[int]|=None):
    """Sliding window into the tensor x.

    Returns a view into the tensor x that represents a sliding window.

fzimmermann89

fzimmermann89

import torch
from torch import nn
from typing import Tuple, Callable


class LUT(nn.Module):
    def __init__(self, f: Callable, dx: float, xrange: Tuple[float, float], mode: str = "linear"):
        """
        LUT of values of a function
        f: function to use, does not need to be differentiable

fzimmermann89

import numpy as np


### Requested pixel2q and q2pixel ####
def pixel2q(pixel, E_ev, detz_m, pixelsize_m=75e-6):
    """
    returns q in reciprocal nm
    E_ev: photon Energy in eV
    detz_m: detector distance in m
    pixelsize_m: detector pixelsize in m

fzimmermann89