khakieconomics

functions {
  matrix L_cov_exp_quad_ARD(matrix x,
                            real alpha,
                            vector rho,
                            real delta) {
                              int N = rows(x);
                              matrix[N, N] K;
                              real sq_alpha = square(alpha);
                              for (i in 1:(N-1)) {
                                K[i, i] = sq_alpha + delta;

khakieconomics

# First load the libraries and define a function to create a 

library(tidyverse); library(rstan)

set.seed(78)
ard_Sigma <- function(features, rho, alpha) {
  noise <- 1e-8
  features <- as.data.frame(features)
  P <- ncol(features)
  if(length(rho) != P) {

khakieconomics

functions {
  // B function
  vector B(vector x, vector t, int i, int j_p1);
  
  vector B(vector x, vector t, int i, int k) {
    vector[rows(x)] out;
    vector[rows(x)] a_i1j1;
    vector[rows(x)] a_ij1;
    
    

khakieconomics

# This script checks that the b-spline function in gist https://gist.github.com/khakieconomics/2272cd7f0d61950852622198b26a2d02
# produces something approximating the function in the splines package. 

library(rstan)
expose_stan_functions("b_spline_function.stan")

probs <- runif(1000)
probs <- probs[order(probs)]

khakieconomics

functions {
  // B function
  vector B(vector x, vector t, int i, int j_p1);
  
  vector B(vector x, vector t, int i, int k) {
    vector[rows(x)] out;
    vector[rows(x)] a_i1j1;
    vector[rows(x)] a_ij1;
    
    

khakieconomics

functions {
  vector inverse_mills(vector z) {
    vector[rows(z)] out;
    
    for(i in 1:rows(z)) {
       out[i] = exp(normal_lpdf(z[i] | 0, 1)) / (Phi(z[i]));
    }
    return(out);
  }
}

khakieconomics

library(rstan); library(tidyverse)

# A utility function
extract_pars <- function(mle_fit, pars) {
  unlist(lapply(pars, function(n) mle_fit$par[grepl(pattern = n, x = names(mle_fit$par))]))
}

# Simulate fake data ------------------------------------------------------

khakieconomics

library(tidyverse); library(lfe)

some_data <- expand.grid(time = 1:10, individual = 1:200) %>% 
  left_join(data_frame(time = 1:10, time_effects=  rnorm(10))) %>% 
  left_join(data_frame(individual = 1:200, individual_effects=  rnorm(200))) %>% 
  mutate(treatment = sample(0:1, n(), replace = T),
         outcome = time_effects + individual_effects + 1 * treatment + rnorm(n())) %>% 
  group_by(time) %>% 
  mutate(demeaned_outcome = outcome - mean(outcome),
         demeaned_treatment = treatment - mean(treatment)) %>% 

khakieconomics

library(tidyverse); library(rstan)

# Some fake data

Individuals <- 1000
Sims <- 1000
Months <- 20


initial_values <- data.frame(customer_id = 1:Individuals, initial_value = rnorm(Individuals))

khakieconomics

---
title: "Demo of mouse-over"
output: 
  flexdashboard::flex_dashboard:
    orientation: columns
    vertical_layout: fill
---

```{r setup, include=FALSE}
library(flexdashboard)