vankesteren

library(lavaan)

# sim_dat function from my blog post
# https://erikjanvankesteren.nl/blog/multivariate_regression
sim_dat <- function(N = 100, P = 5, Q = 2, varprop = 0.5, rho_X = 0, rho_B = 0, rho_E = 0, exact = TRUE) {
  # Create X matrix
  X <- matrix(rnorm(N*P), N)
  RX <- matrix(rho_X, P, P)
  diag(RX) <- 1
  CRX <- chol(RX)

vankesteren

import gzip

x1 = "Japan’s Seiko Epson Corp. has developed a 12-gram flying microrobot."
x2 = "The latest tiny flying robot has been unveiled in Japan."
x3 = "Michael Phelps won the gold medal in the 400 individual medley."

def ncd(x: str, y: str) -> float:
    '''
    Returns the normalized compression distance between two strings.

vankesteren

# density ratio estimation using ulsif with laplace kernel
library(kernlab)
library(rmutil)
set.seed(45)

# two samples
x <- matrix(rlaplace(10000))
y <- matrix(rnorm(10000, sd = 0.707))

# kernel stuff

vankesteren

# Figuring out whether this logspline thing is any good
library(logspline)

# mixture
d1 <- function(x) 0.6*dnorm(x) + 0.3*dlnorm(x, sdlog = 0.1) + 0.1*dnorm(x, 2)
r1 <- function(n) sample(c(rnorm(round(n*.6)), rlnorm(round(n*.3), sdlog = 0.1), rnorm(round(n*.1), 2)))

set.seed(123)
x <- r1(100)
dd <- density(x, n = 1000)

vankesteren

library(torch)
tot_obs <- 1000
x <- c(rnorm(.2*tot_obs), rnorm(.2*tot_obs, 5, 2), rnorm(.3*tot_obs), rnorm(.1*tot_obs, 5, 2), rnorm(.2*tot_obs))
x_torch <- torch_tensor(matrix(x))
plot(x, type = "l")


# Create a very simple lstm, with only one hidden node and linear activation for the output
SimpleLSTM <- nn_module("simplelstm",
  initialize = function(obs_size, hidden_size) {

vankesteren

# setup with seeds for reproducibility
library(torch)
set.seed(45)
torch_manual_seed(45)

# True distribution is normal(1, 3), we sample 10000 points
N <- 500
y <- rnorm(N, 1, 3)

# we need to create torch tensor of this data to use in torch

vankesteren

#include <RcppArmadillo.h>
// [[Rcpp::depends(RcppArmadillo)]]

// [[Rcpp::export]]
double rbf_kern(const arma::vec xi, const arma::vec xj, const double& gamma) {
  return exp(-gamma*norm(xi-xj));
}

// [[Rcpp::export]]
arma::mat rbf_gram_cpp(const arma::mat& X, const double& gamma) {

vankesteren

# Plot & interpolate BBB vote percentage in Utrecht
# using open data from data.utrecht.nl
# last edited 20230320 by 
library(tidyverse)
library(sf)
library(curl)
library(glue)
library(jsonlite)
library(pbapply)
library(ggspatial)

vankesteren

Sector_score,LOCATION_ID,PARA_A,Score_A,Risk_A,PARA_B,Score_B,Risk_B,TOTAL,numbers,Score_B,Risk_C,Money_Value,Score_MV,Risk_D,District_Loss,PROB,RiSk_E,History,Prob,Risk_F,Score,Inherent_Risk,CONTROL_RISK,Detection_Risk,Audit_Risk,Risk
3.89,23,4.18,0.6,2.508,2.5,0.2,0.5,6.68,5,0.2,1,3.38,0.2,0.676,2,0.2,0.4,0,0.2,0,2.4,8.574,0.4,0.5,1.7148,1
3.89,6,0,0.2,0,4.83,0.2,0.966,4.83,5,0.2,1,0.94,0.2,0.188,2,0.2,0.4,0,0.2,0,2,2.554,0.4,0.5,0.5108,0
3.89,6,0.51,0.2,0.102,0.23,0.2,0.046,0.74,5,0.2,1,0,0.2,0,2,0.2,0.4,0,0.2,0,2,1.548,0.4,0.5,0.3096,0
3.89,6,0,0.2,0,10.8,0.6,6.48,10.8,6,0.6,3.6,11.75,0.6,7.05,2,0.2,0.4,0,0.2,0,4.4,17.53,0.4,0.5,3.506,1
3.89,6,0,0.2,0,0.08,0.2,0.016,0.08,5,0.2,1,0,0.2,0,2,0.2,0.4,0,0.2,0,2,1.416,0.4,0.5,0.2832,0
3.89,6,0,0.2,0,0.83,0.2,0.166,0.83,5,0.2,1,2.95,0.2,0.59,2,0.2,0.4,0,0.2,0,2,2.156,0.4,0.5,0.4312,0
3.89,7,1.1,0.4,0.44,7.41,0.4,2.964,8.51,5,0.2,1,44.95,0.6,26.97,2,0.2,0.4,0,0.2,0,3.2,31.774,0.4,0.5,6.3548,1
3.89,8,8.5,0.6,5.1,12.03,0.6,7.218,20.53,5.5,0.4,2.2,7.79,0.4,3.116,2,0

vankesteren

Efficient Gaussian Kullback-Leibler divergence in R

Erik-Jan van Kesteren

Introduction

In this document, I create a small R function for computing the Kullback-Leibler divergence between two (multivariate) normal (Gaussian) distributions. The general formula for the divergence is as follows: