North Shore Data Modeling Contest Submission

Final submission for NSDMC

# Load these Libraries
library(dplyr)
library(caret)
library(e1071)
library(knitr)
library(rmarkdown)
library(quanteda)

Import Data

# Import raw data given to us
rawdata <- read.csv("NSDMC_training_data.csv - NSDMC_training_data.csv.csv")

# Scramble Data
scramble <- sample(1:22984, size = 1500)
rawdata <- rawdata[scramble,]

Make a corpus

# Convert text data into Class character
rawdata$Subject <- as.character(rawdata$Subject)

# Create a corpus
datacorp <- corpus(rawdata$Subject)

# Check Data 
datacorp[10]

##                                       text10 
## "Coupons from Walmart, Target & Whole Foods"

Tokenize data

# Tokenize the corpus
tokens <- tokens(datacorp, remove_punct = TRUE)

# Word Stem the tokens
newtokens <- tokens_wordstem(tokens)
 
# Lowercase the tokens
newtokens2 <- tokens_tolower(newtokens)

# See what data looks like 
newtokens2[10]

## tokens from 1 document.
## text10 :
## [1] "coupon"  "from"    "walmart" "target"  "whole"   "food"

Create a DFM

# Create a dfm
mydfm <- dfm(newtokens2, remove = stopwords())
 
# Trim DFM
trimdfm <- dfm_trim(mydfm, min_termfreq = 15, min_docfreq = 8)
 
# Turn DFM into a dataframe
mydf <- data.frame(trimdfm)
 
# Drop "document" variable
mydf <- select(mydf, -document)

# Append variables on to that dataframe
mydf2 <- cbind(rawdata$class_labels, mydf)

# Rename the rawdata$class_labels variable to "class_labels"
names(mydf2)[1] <- "class_labels"

Splitting Data

# Split mydf2 
traindf <- mydf2[1:500,]
testdf <- mydf2[501:1500,]

Train Model on training set

# Get our features and response variable
train_features <- select(traindf, -class_labels)
train_response <- traindf$class_labels

Model xgboost

# Create Model
mod <- train(x = train_features, y = train_response, method = "xgbTree")

# Make predictions on test data
preds <- predict(mod, testdf)
tbl <- table(preds, testdf$class_labels)

# Function to find percent accuracy
percentage <- function(tbl){
  (tbl[1,1] + tbl[2,2] + tbl[3,3]) / 1000*100
}
percentage(tbl)

## [1] 54.2

# Save Model
saveRDS(mod, "xgbmodel.Rds")