Finding John’s location depending on the day type (week day or weekend)

• Load the naivebayes package.

• Use naive_bayes() with a formula like y ~ x to build a model of location as a function of daytype.

• Forecast the location of John during a morning weekday, using predict() with the weekday_morning object as the newdata argument.

• Repeat the prediction for a weekend_morning location.

Solution

# prepare the environment 
library(tidyverse)
library(readr)
locations_data    <- readr::read_csv(here::here("/content/post/ml-nb-predictloc/_data/locations_john_morning.csv"))
# find out the day type if it's weekday
weekday_morning <- locations_data %>% dplyr::select(daytype) %>% dplyr::filter(daytype == "weekday") %>% slice(n())
# find out the day type if it's a weekend
weekend_morning <- locations_data %>% dplyr::select(daytype) %>% dplyr::filter(daytype == "weekend") %>% slice(n())

# Load the naivebayes package
library(naivebayes)

# Build the location prediction model
locmodel <- naive_bayes(location ~ daytype, data = locations_data)


# Predict location on a week day
predict(locmodel, weekday_morning)

## [1] class
## Levels: campus class dorm tutoring

# Predict location on a weekend
predict(locmodel, weekend_morning)

## [1] dorm
## Levels: campus class dorm tutoring

Find the “raw” probabilities of John being at those locations

The naivebayes package offers several ways to peek inside a Naive Bayes model.

Typing the name of the model object provides the a priori (overall) and conditional probabilities of each of the model’s predictors. If one were so inclined, you might use these for calculating posterior (predicted) probabilities by hand.

Alternatively, R will compute the posterior probabilities for you if the type = "prob" parameter is supplied to the predict() function.

Using these methods, examine how the model’s predicted 9am location probability varies from day-to-day. The model locmodel that you fit in the previous exercise is in your workspace.

Solution

• Print the locmodel object to the console to view the computed a priori and conditional probabilities.
  
• Use the predict() function similarly to the previous exercise, but with type = "prob" to see the predicted probabilities for John’s location on a week day’s morning.
  
• Compare these to the predicted probabilities for John’s location on a weekend morning.

# The 'naivebayes' package is loaded into the workspace
# and the Naive Bayes 'locmodel' has been built

# Examine the location prediction model
locmodel

## 
## ================================== Naive Bayes ================================== 
##  
##  Call: 
## naive_bayes.formula(formula = location ~ daytype, data = locations_data)
## 
## --------------------------------------------------------------------------------- 
##  
## Laplace smoothing: 0
## 
## --------------------------------------------------------------------------------- 
##  
##  A priori probabilities: 
## 
##     campus      class       dorm   tutoring 
## 0.10989011 0.42857143 0.45054945 0.01098901 
## 
## --------------------------------------------------------------------------------- 
##  
##  Tables: 
## 
## --------------------------------------------------------------------------------- 
##  ::: daytype (Bernoulli) 
## --------------------------------------------------------------------------------- 
##          
## daytype      campus     class      dorm  tutoring
##   weekday 1.0000000 1.0000000 0.3658537 1.0000000
##   weekend 0.0000000 0.0000000 0.6341463 0.0000000
## 
## ---------------------------------------------------------------------------------

# Obtain the predicted probabilities for a week day morning
predict(locmodel, weekday_morning, type = "prob")

##         campus class      dorm   tutoring
## [1,] 0.1538462   0.6 0.2307692 0.01538462

# Obtain the predicted probabilities for a weekend morning
predict(locmodel, weekend_morning, type = "prob")

##            campus       class      dorm     tutoring
## [1,] 0.0003838772 0.001497121 0.9980806 3.838772e-05