Last active
October 19, 2023 03:56
-
-
Save norabelrose/459ee8c2e6f3c57165aabdb1367a30ee to your computer and use it in GitHub Desktop.
Ryan Greenblatt's cumulant estimation code
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from typing import Optional | |
import torch | |
def get_all_the_cumulants( | |
x: torch.Tensor, y: torch.Tensor, z: torch.Tensor, w: torch.Tensor, weights_in: Optional[torch.Tensor] = None | |
): | |
if weights_in is not None: | |
weights = weights_in | |
weights = weights / weights.sum() | |
adj_val = (weights ** 2).sum() # I don't remember why this is right, so use at your own risk : ) | |
else: | |
s = x.size(0) | |
weights = torch.full((s,), 1 / s, dtype=x.dtype, device=x.device) | |
adj_val = torch.tensor(1 / s, dtype=x.dtype, device=x.device) | |
vals = [x, y, z, w] | |
all_means = [torch.einsum("i x, i -> x", v, weights).unsqueeze(0) for v in vals] | |
c_x, c_y, c_z, c_w = [v - m for v, m in zip(vals, all_means)] | |
x_mean = all_means[0].squeeze(0) | |
uncorrected_cov = torch.einsum("b i, b j, b -> i j", c_x, c_y, weights) | |
corrected_cov = uncorrected_cov / (1 - adj_val) | |
uncorrected_third_cum = torch.einsum("b i, b j, b k, b -> i j k", c_x, c_y, c_z, weights) | |
corrected_third_cum = uncorrected_third_cum / ((1 - adj_val) * (1 - 2 * adj_val)) | |
centered_4th_mom = torch.einsum("b i, b j, b k, b l, b -> i j k l", c_x, c_y, c_z, c_w, weights) | |
first_pair = torch.einsum( | |
"i j, k l -> i j k l", | |
torch.einsum("b i, b j, b -> i j", c_x, c_y, weights), | |
torch.einsum("b k, b l, b -> k l", c_z, c_w, weights), | |
) | |
snd_pair = torch.einsum( | |
"i k, j l -> i j k l", | |
torch.einsum("b i, b k, b -> i k", c_x, c_z, weights), | |
torch.einsum("b j, b l, b-> j l", c_y, c_w, weights), | |
) | |
thrd_pair = torch.einsum( | |
"i l, j k -> i j k l", | |
torch.einsum("b i, b l, b -> i l", c_x, c_w, weights), | |
torch.einsum("b j, b k, b -> j k", c_y, c_z, weights), | |
) | |
pair_sum = first_pair + snd_pair + thrd_pair | |
uncorrected_4th_cum = centered_4th_mom - pair_sum | |
corrected_4th_cum = (1 / ((1 - adj_val) * (1 - 2 * adj_val) * (1 - 3 * adj_val))) * ( | |
(1 + adj_val) * centered_4th_mom - (1 - adj_val) * (pair_sum) | |
) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment