tvladeck

cat orders_2018-05-01.csv | head | sed -e 's/,,/, ,/g' | column -s, -t | less -#5 -N -S

tvladeck

set.seed(20180226)
block_mining_times <- rpois(1000, 10) # lambda = 10
max_time <- sum(block_mining_times)
cumulative_times <- cumsum(block_mining_times)

rand_times <- runif(1000, min = 0, max = max_time)

deltas_between_blocks <- 
  rand_times %>% 
  sapply(function(t){

tvladeck

# We simulate draws from two different timing distributions representing different processes
payment_failures  <- rexp(1000, rate = 1/90)
intentional_churn <- rexp(1000, rate = 1/120)

### proportion of payment failures to intentional churn at 90 day cutoff

pf_90 <- (payment_failures < 90) %>% sum
if_90 <- (intentional_churn < 90) %>% sum 

pf_90 / (if_90 + pf_90)

tvladeck

# draw from an exponential distribution with mean 90
payment_failures  <- rexp(1000, rate = 1/90)

# number of payment failures after 90 days
pf_90 <- (payment_failures < 90) %>% sum

# number of payment failures after 60 days (assuming same distribution)
pf_60 <- (payment_failures < 60) %>% sum

# number of 90 day payment failures observed relative to 60 day payment failures

tvladeck

library(corrplot)

playstation_data <- matrix(0, nrow = 6, ncol = 5)
rownames(playstation_data) <- c("Connoisseurs", "Connectors", "Rivals", "Indulgers", "Escapers", "Dabblers")
colnames(playstation_data) <- c("Periphery", "Hit", "Mid-Core", "Core", "Occasionals")

playstation_data[1, ] <- c(10, 5, 3, 101, 18)
playstation_data[2, ] <- c(38, 10, 72, 18, 68)
playstation_data[3, ] <- c(10, 110, 22, 28, 15) 
playstation_data[4, ] <- c(80, 10, 9, 44, 70)

tvladeck

# normalize sequences
tx_sample <-
  tx_sample %>%
  mutate(
    order_date = order_date - first_date
  )
  
# create TraMineRextras object
tx_seq <- seqecreate(
  id = tx_sample$customer_id,

tvladeck

korresp_som <- trainSOM(a, scaling = "chi2", type = "korresp", dimension = c(2,2))

row_num <- korresp_som$parameters$the.grid$dim[[1]]
col_num <- korresp_som$parameters$the.grid$dim[[2]]

korresp_clusters <- 
  korresp_som$clustering %>% 
  as.data.frame %>% 
  set_colnames(c("cluster")) %>% 
  mutate(entity = rownames(.)) %>% 

tvladeck

indexation <- function(A)
{
  A / ((rowSums(A)/sum(rowSums(A))) %*% 
         t(colSums(A)/sum(colSums(A))) * 
         sum(A))
}

tvladeck

model {

  for(i in 1:n_obs){
    
    hrc_votes[i] ~ dbin(p[i], n_votes[i])
    p[i] ~ dbeta(alpha[i], beta[i])
    
    # reparameterization of the beta distribution taken from this site
    # http://bit.ly/2i880Oj
    

tvladeck

# Betas #

samples <- coda.samples(jags, c('beta_age',
                                'beta_percent_white',
                                'beta_uninsured',
                                'beta_unemployed',
                                'beta_percent_degree',
                                'beta_income',
                                'beta_health'), 10000)