albertbuchard

"""
This example demonstrates that fitting a weighted linear mixed-effects model using lmer(weights = …) 
is algebraically equivalent to pre-multiplying the outcome and all predictors (including intercepts) 
by the square root of the weights and fitting an unweighted model on the transformed data. 
The test validates that this "pre-weight and refit" strategy yields identical fitted values and 
is especially useful when packages like clubSandwich prohibit prior weights in robust variance estimation. 
Includes realistic longitudinal synthetic data with fixed and random effects, dropout, and continuous covariates.
"""
from typing import List

albertbuchard

Active learning in one line Keep a model $p_\theta(y\mid x,\mathcal D)$ that already knows something ($\mathcal D$); then ask for the single measurement $(x^*,y^*)$ that, once answered, will help most.

---

1. Value of Information (VoI)

Concept	Formal core	Intuition for the next query
Decision-centric	$\text{VoI}(x)=\mathbb E_{y\sim p_\theta(y\mid x,\mathcal D)}\big[\Delta U(\theta\mid x,y)\big]$	“If I were given the label of $x$, how much would my future utility rise on average?”

albertbuchard

import matplotlib.pyplot as plt
import networkx as nx
from causallearn.utils.KCI.KCI import KCI_CInd, KCI_UInd
from npeet_plus import mi, mi_pvalue
from tqdm import tqdm

from src.causal_discovery.static_causal_discovery import (
    run_causal_discovery,
    visualize_causal_graph,
)

albertbuchard

#!/usr/bin/env python3

import fnmatch
import os

"""
This Python script, textrepo, concatenates all files within a specified repository into a single text file
while respecting .gitignore patterns and additional specified ignore patterns. It prints the formatted content
to both a specified output file and standard output. This is useful for reviewing all content within a repository
in a structured format, excluding unwanted files and directories such as node_modules, dist, build, and others.

albertbuchard

from concurrent.futures import ProcessPoolExecutor

import numpy as np
from scipy.signal import convolve2d
from tqdm import tqdm

nsims = int(1e6)
nevents = int(1e5)

albertbuchard

tailast() {
    # Get the directory from the argument or use the current directory as default
    local dir="${1:-.}"

    # Find the most recently modified file in the directory without descending into subdirectories
    local latest_file=$(find "$dir" -maxdepth 1 -type f -exec stat --format='%Y %n' {} \; | sort -n | tail -1 | awk '{print $2}')

    # Check if a file was found
    if [[ -z "$latest_file" ]]; then
        echo "No files found in $dir"

albertbuchard

import math
import torch.optim as optim
import torch
from torch import nn


class MineWrapper(nn.Module):

    def __init__(self, stat_model, moving_average_rate=0.1, unbiased=False):
        super(MineWrapper, self).__init__()

albertbuchard

class DistributedIdentity:
    """
    Singleton class to hold distributed identity information.
    Handles SLURM, torchrun, and local runs.

    Looks for the following environment variables:
    - RANK
    - WORLD_SIZE
    - LOCAL_RANK

albertbuchard

import torch
import torch.nn as nn

class Sparsemax(nn.Module):
    def __init__(self, dim=-1):
        super(Sparsemax, self).__init__()
        self.dim = dim

    def forward(self, x):
        # Move the dimension to apply Sparsemax to the front

albertbuchard

setup='''
import numpy as np
import torch

V_nat = [[1, 2], [3, 4]]
U_nat = [[2, -1, 0, 0, 0, 0],
         [5, 2, 8, -1, 0, 0]]


def compute_using_fleuret_1():