gaurav soodoku

Adaptive LassoCV: fixed-data reproducibility experiment

This experiment evaluates the library implementation in scikit-learn PR #35288, commit 3ac07a78ca6180035c699ef5f2ea97766431dd1b in the finite-sample fork. It is supporting evidence, not a benchmark added to the upstream source tree.

Protocol

The six prespecified cells cross independent Gaussian features or Toeplitz correlation 0.7**abs(i-j) with strong, weak, or null signals. The coefficient vectors begin with [1, -.8, .6], [.25, -.2, .15], or zeros, respectively; the remaining coefficients are zero. Each cell uses one fixed dataset with 80 training rows, 10 features, independent unit Gaussian noise, and a fixed 2,000-row test set. Dataset seeds are 20261101 + cell.

Each cell has 200 independent pairs (400 executions). Each execution fits ordinary single randomized 5-fold CV, fixed 10/50/100/500 repetitions, and adaptive CV using the actual LassoCV API. Methods within an exec

Personality Change

Personality psychology reports "moderate stability" of traits across adulthood, but what does this mean for individuals? We simulated how likely someone is to obtain similar Big Five personality test results over a 30-year span, critically accounting for measurement error.

Method: We generated 10,000 individuals with Big Five traits following a multivariate normal distribution with empirical intercorrelations (e.g., Neuroticism with Extraversion: -0.22). We modeled personality at Time 2 using the test-retest formula: New_Score = Old_Score × r + error × √(1-r²), adding measurement error (reliability = 0.82, per Gnambs 2014) at both time points. We defined "meaningful change" as shifts exceeding 0.5 standard deviations. Crucially, we established a baseline scenario with perfect stability (r=1.0) to isolate measurement error effects, finding that 33% appear to change due to measurement error alone. We then examined: Soldz & Vaillant's (1999) 30-year correlations (E: 0.19, N: 0.20, O: 0.3

To go with Andrew's https://statmodeling.stat.columbia.edu/2017/05/19/continuous-hinge-function-bayesian-modeling/

Also includes alternate smoothers outside of logistic.

Used Claude (but verified).

Scraping the abstracts for the discussion papers by i4replication.org.

	# -- coding: utf-8 --
	"""sentiment_ribeiro_style_checklist.ipynb

	Automatically generated by Colab.

	Original file is located at
	https://colab.research.google.com/drive/1uyBg9-sidauCSSO6dcztl552HVSAgkUJ
	"""

	"""

	data {
	int<lower=0> N;
	int<lower=0> D;
	matrix[N, D] x;
	array[N] int<lower=0, upper=1> y;

	real<lower=0> h;
	int<lower=0> B;
	array[B] vector[D] beta0;

	#!/usr/bin/env python3
	import argparse
	import json
	import requests
	import sys
	from urllib.parse import quote

	def parse_arguments():
	"""Parse command line arguments"""
	parser = argparse.ArgumentParser(description="List Dataverse datasets owned by a user and convert to JSON")

	#!/usr/bin/env python3
	"""
	Dataverse File Uploader

	A script to upload files to Harvard Dataverse or other Dataverse instances.
	"""

	import os
	import argparse
	import requests