##############################################################################
#
# Purpose:
# Use pilot project data to calculate power of a full study through simulation
#
# Parts:
# (0) - Setup
# (1) - Get the pilot data and clean it
# (2) - Run the model on the pilot data and extract effects
# (3) - Set up and run the simulation
# ====> Set variables at the arrows <====
#
##############################################################################
rm(list=ls())
set.seed(424242) 

library(readr)
library(ggplot2)

# (1) - Get the pilot data and clean it
#source('~/Research/tor_wikipedia_edits/handcoded_edits/inter_coder_reliability_ns0.R')
#source ('/data/users/mgaughan/kkex_data_110823_3')
data1 <- read_csv('../power_data_111023_mmt.csv',show_col_types = FALSE)
data2 <- read_csv('../inst_all_packages_full_results.csv')
#d$nd <- to_logical(d$not.damaging, custom_true=c("Y")) 
#levels(d$source) <- c("IP-based Editors", "New Editors", "Registered Editors", "Tor-based Editors")
python_labeled <- as.numeric(data2$up.fac.mean[match(paste('python',tolower(data1$pkg), sep = "-"), data2$pkg)])
same_labeled <- as.numeric(data2$up.fac.mean[match(tolower(data1$pkg), data2$pkg)])
data1$up.fac.mean <- pmin(python_labeled, same_labeled, na.rm=TRUE)
data1$milestones <- as.numeric(data1$milestones > 0)
# (2) - Run the model on the pilot data
data1$formal.score <- data1$mmt / (data1$milestones/data1$age)
table(data1$milestones)
hist(data1$old_mmt) #inequality of participation
hist(data1$formal.score)
hist(data1$age/365)
kmodel1 <- lm(up.fac.mean ~ mmt, data=data1)
summary(kmodel1)
kmodel1 <- lm(up.fac.mean ~ old_mmt, data=data1)
summary(kmodel1)
kmodel1 <- lm(up.fac.mean ~ formal.score, data=data1)
summary(kmodel1)
hist(data1$formal.score)
cor.test(data1$formal.score, data1$up.fac.mean)
cor.test(data1$mmt, data1$up.fac.mean)
cor.test(data1$milestones, data1$up.fac.mean)
cor.test(data1$age, data1$up.fac.mean)

g <- ggplot(data1, aes(x=formal.score, y=up.fac.mean)) +
  geom_point() +
  geom_smooth()
g

data2 <- subset(data1, (data1$age / 365) < 14 )
hist(data2$age)
g <- ggplot(data2, aes(x=formal.score, y=up.fac.mean)) +
  geom_point() +
  geom_smooth()
g

data2$yearsOld <- data2$age / 365

kmodel2 <- lm(up.fac.mean ~ mmt + milestones + yearsOld, data=data2)
summary(kmodel2)

#pilotM <- glm(up.fac.mean ~ ((mmt) / (milestones/age)),   # give the anticipated regression a try
#                  family=gaussian(link='identity'), data=data1)
summary(pilotM) #we expect effect sizes on this order

pilot.b0 <- coef(summary(pilotM))[1,1] 
pilot.b1 <- coef(summary(pilotM))[2,1]
pilot.b2 <- coef(summary(pilotM))[3,1]
pilot.b3 <- coef(summary(pilotM))[4,1] 


# (3) - Set up and run the simulation

source('powerAnalysis.R') #my little "lib"

#====>
nSims <- 5000 #how many simulations to run
n <- 100 #a guess for necessary sample size (per group)
#makeData(10) #DEBUGGING CODE -- you can uncomment this if you want to see it work
#<====

#print("Levels are:")
#print(levels(d$source))
powerCheck(n, nSims)

#Sample values
powerCheck(50, 100) 
powerCheck(80, 1000) 
powerCheck(200, 5000)