cwindolf

# adapted from https://matplotlib.org/3.1.1/gallery/statistics/confidence_ellipse.html
import numpy as np
import matplotlib.pyplot as plt
import colorcet
from matplotlib.patches import Ellipse
import matplotlib.transforms as transforms


def cluster_scatter(xs, ys, ids, ax=None, n_std=2.0):
    ax = ax or plt.gca()

cwindolf

import torch
import torch.nn.functional as F


def translate(input, shifts, mode='bilinear', padding_mode='zeros'):
    """Translate a batch of images by a batch of xy shifts

    Arguments
    ---------
    input : torch.Tensor NCHW

cwindolf

import numpy as np
from .triginterp_inner import triginterp_inner


def triginterp(x, y, xi):
    """Trigonometric interpolation with zero-padding (lol, not periodic). Easy to change."""
    P = np.zeros_like(xi, dtype=np.float64)
    triginterp_inner(
        x.astype(np.float64), y.astype(np.float64), xi.astype(np.float64), P
    )

cwindolf

"""
Implements the nonparametric simultaneous confidence interval described
in Gao et al. [1], which modifies the Mandel-Betensky simultaneous CI [2]
such that it is admissible. Notation in the code below comes from [1].

Takes as input an array of bootstrap samples / replicates, returns lower
and upper bounds at 1-alpha confidence.

[1]: https://www.mdpi.com/2073-8994/13/7/1212
[2]: https://pubmed.ncbi.nlm.nih.gov/26452746/

cwindolf

import tempfile
import imageio
from IPython.display import Image

def ipygif(z, fps=16, width=200, height=200):
    f = tempfile.NamedTemporaryFile(prefix="ipygif", suffix=".gif")
    z = z - z.min()
    z *= 255.0 / z.max()
    imageio.mimwrite(f.name, z.astype(np.uint8), fps=fps)
    return Image(filename=f.name, embed=True, width=width, height=height)

cwindolf

import numpy as np

def zroll1d(arr1d, x):
    """np.roll, but not toroidal (backfills with 0s), and only 1d."""
    if x == 0:
        return arr1d
    elif x > 0:
        return np.pad(
            arr1d[:-x],
            ((x, 0),),

cwindolf

import numpy as np


def argmedian(p, axis=None, check_pmf=True):
    """Compute the median of a pmf or many pmfs

    Here, `p` represents one or many probability mass functions, and
    the index to a median will be returned for each pmf.

        - If `axis=None`, the flattened version of `p` will be considered

cwindolf

# Translation of the Octave implementation (GPL), which is
# copyright 2014 Benjamin Eltzner. For that code, and license, see:
# hg.code.sf.net/p/octave/image/file/tip/inst/normxcorr2.m
import numpy as np
from scipy.signal import correlate, convolve


def normxcorr(template, x, mode="full", method="auto", assume_centered=False):
    """normxcorr: Normalized cross-correlation

cwindolf

<snippet>
 <content><![CDATA[
\$\$\begin{aligned}
$1
\end{aligned}\$\$
]]></content>
 <tabTrigger>ali</tabTrigger>
 <scope>text.html.markdown,text.html.markdown.multimarkdown</scope>
</snippet>

cwindolf

import time

class timer:
    def __init__(self, name="timer", format="{:0.1f}"):
        self.name = name
        self.format = format
 
    def __enter__(self):
        self.start = time.perf_counter()
        return self