In nikkyxiong/coffee_rating_and_nyc_demographics:

knitr::opts_chunk$set(echo = TRUE)

library(readr)
library(dplyr)
library(broom)

nyc_demo <- read_csv("nyc_zip_demo.csv")
nyc_demo_more <- read_csv('raw_data/nyc_zip_demographics.csv')
nyc_demo <- left_join(nyc_demo, nyc_demo_more,by = c("Zipcode" = "ZIPCODE"))
coffee_zip <- read_csv("coffee_zip.csv")
df <- left_join(coffee_zip, nyc_demo) %>% 
    mutate(Women_pct = 100 - Men_pct)

linear regression

coffee_lm <- df

coffee_lm$Name[coffee_lm$Name == "Dunkin'"] <- 0
coffee_lm$Name[coffee_lm$Name == "Starbucks"] <- 1
coffee_lm$Name[coffee_lm$Name == "Blue Bottle"] <- 2

coffee_sb <- coffee_lm %>% filter(Name == 1)
coffee_dd <- coffee_lm %>% filter(Name == 0)
coffee_bb <- coffee_lm %>% filter(Name == 2)

Supervised Machine Learning

coffee_ml <- coffee_lm %>% 
  filter(coffee_lm$Name == 1 | coffee_lm$Name == 0)

library(caret)

coffee_ml$Name <- factor(coffee_ml$Name, 
                     labels = c("Starbucks", "DD"), 
                     levels = 1:0) 
set.seed(12345)
in_train <- createDataPartition(y = coffee_ml$Name, 
                                p = 0.8, list = FALSE)
training <- coffee_ml[ in_train, ]
testing  <- coffee_ml[-in_train, ]

logit

logit <- glm(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
             data = training, family = binomial(link = "logit"))

y_hat_logit <- predict(logit, newdata = testing, type = "response")
z_logit <- factor(y_hat_logit > 0.5, 
                  levels = c(TRUE, FALSE), 
                  labels = c("Starbucks", "DD"))

confusionMatrix(z_logit, reference = testing$Name)

LDA

LDA <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
             data = training, method = "lda", 
             reProcess = c("center", "scale"))
z <- predict(LDA, newdata = testing)
confusionMatrix(z, testing$Name)

QDA

QDA <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
             data = training, method = "qda", 
             preProcess = c("center", "scale"))

z_QDA <- predict(QDA, newdata = testing)

confusionMatrix(z_QDA, reference = testing$Name)

PLSDA

ctrl_PLSDA <- trainControl(method = "repeatedcv", repeats = 2, 
                     classProbs = TRUE, summaryFunction = twoClassSummary)
PLSDA_grid <- expand.grid(.ncomp = 1:2)

PLSDA <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
               data = training, method = "pls", 
               preProcess = c("center", "scale"), metric = "ROC", 
               trControl = ctrl_PLSDA, tuneGrid = PLSDA_grid)

z_PLSDA <- predict(PLSDA, newdata = testing)

confusionMatrix(z_PLSDA, reference = testing$Name)

Nearest Shrunken Centroids

ctrl_NSC <- trainControl(method = "repeatedcv", repeats = 3, 
                     classProbs = TRUE, summaryFunction = twoClassSummary)
grid_NSC <- data.frame(.threshold = 0:25)

NSC <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
             data = training, method = "pam", 
             preProcess = c("center", "scale"), metric = "ROC", 
             trControl = ctrl_NSC, tuneGrid = grid_NSC)

z_NSC <- predict(NSC, newdata = testing) 
confusionMatrix(z_NSC, reference = testing$Name)

Boosting

gbmControl = trainControl(method="cv", 
                          number=5, returnResamp = "all")

gbm = train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot,
            data=training, method="gbm",
            distribution="bernoulli", 
            trControl=gbmControl, verbose=F, 
            tuneGrid=data.frame(.n.trees=seq(100, 1000, 
                                              by = 100), 
                                 .shrinkage=c(0.01, 0.1), 
                                 .interaction.depth=1:2, 
                                 .n.minobsinnode=1:5),
             na.action = na.omit)

y_hat_gbm <- predict(gbm, newdata = testing, na.action = na.pass)

confusionMatrix(y_hat_gbm, reference = testing$Name)

Random Forest

library(randomForest)
rf<- randomForest(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
                  data=training,
                  importance = TRUE,
                  na.action = na.omit)

y_hat_rf <- predict(rf, newdata = testing,
                 type = "response", na.action = na.pass)

confusionMatrix(y_hat_rf, reference = testing$Name)

Neural Network Model

nnetGrid <- expand.grid(.decay = c(0, 0.01, .1),
                        .size = c(1:10))
ctrl_nn <- trainControl(method = "cv", number = 10)

nn <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot, 
            data = training, method = "nnet",
            trControl = ctrl_nn, tuneGrid = nnetGrid,
            preProcess = c("center", "scale"), trace = FALSE)

y_hat_nn <- predict(nn, newdata = testing)
confusionMatrix(y_hat_nn, reference = testing$Name)

nn <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot + Median_Income, 
            data = training, method = "nnet",
            trControl = ctrl_nn, tuneGrid = nnetGrid,
            preProcess = c("center", "scale"), trace = FALSE)

y_hat_nn <- predict(nn, newdata = testing)
confusionMatrix(y_hat_nn, reference = testing$Name)

nn <- train(Name ~ Avg_rating + Avg_review_count +
               White_tot + Asian_tot + Black_tot, 
            data = training, method = "nnet",
            trControl = ctrl_nn, tuneGrid = nnetGrid,
            preProcess = c("center", "scale"), trace = FALSE)

y_hat_nn <- predict(nn, newdata = testing)
confusionMatrix(y_hat_nn, reference = testing$Name)

nn <- train(Name ~ Avg_rating + Avg_review_count, 
            data = training, method = "nnet",
            trControl = ctrl_nn, tuneGrid = nnetGrid,
            preProcess = c("center", "scale"), trace = FALSE)

y_hat_nn <- predict(nn, newdata = testing)
confusionMatrix(y_hat_nn, reference = testing$Name)

MARS model

ctrl_mars <- trainControl(method = "cv", number = 10)
marsGrid <- expand.grid(.degree = 1:3, .nprune = 1:10)

mars <- train(Name ~ Avg_rating + Avg_review_count + Population +
               White_tot + Asian_tot + Black_tot,  
              data = training, method = "earth", 
              trControl = ctrl_mars, tuneGrid = marsGrid)

y_hat_mars <- predict(mars, newdata = testing)
confusionMatrix(y_hat_mars, reference = testing$Name)