# Setup

library(jsonlite)
library(tidyverse)

train <- read.csv('../input/train.csv')
test <- read.csv('../input/test.csv')

# All JSON columns must be turned into proper arrays before flattening.
# i.e. with a separator "," between each object and wrapped up with "[]" 

json2df <- function(x) { 
  sprintf('[%s]', paste(x, collapse = ',')) %>%
    fromJSON(flatten = T)
}

# Since there aren't too many columns we could manually create a new dataframe!
# Column names may be checked with names(train).

train.df <- data.frame(
  
  fullVisitorId = train$fullVisitorId, # ID
  channelGrouping = train$channelGrouping,
  date = train$date,
  sessionID = train$sessionId,
  socialEngagementType = train$socialEngagementType,
  visitId = train$visitId,
  visitNumber = train$visitNumber,
  visitStartTime = train$visitStartTime,
  
  json2df(train$device), # device (JSON)
  json2df(train$geoNetwork), #geoNetwork (JSON)
  json2df(train$totals), # totals (JSON)
  json2df(train$trafficSource) # trafficSource (JSON)
  
)

# Do the same for the test set.

test.df <- data.frame(
  
  fullVisitorId = test$fullVisitorId, # ID
  channelGrouping = test$channelGrouping,
  date = test$date,
  sessionID = test$sessionId,
  socialEngagementType = test$socialEngagementType,
  visitId = test$visitId,
  visitNumber = test$visitNumber,
  visitStartTime = test$visitStartTime,
  
  json2df(test$device), # device (JSON)
  json2df(test$geoNetwork), #geoNetwork (JSON)
  json2df(test$totals), # totals (JSON)
  json2df(test$trafficSource) # trafficSource (JSON)
  
)

# Save them as new csv files:

write.csv(train.df, "traindf.csv", row.names = F)
write.csv(test.df, "testdf.csv", row.names = F)

# Any suggestions for improvement are much appreciated!