[MSR] RTVS - A First Look at R

/*******************************************************************************************************************
-- Title : [MSR] RTVS - A First Look at R
-- Reference : microsoft.com
-- Key word : microsoft r installed.packages library search ggplot2 data head tail class str nrow ggplot 
                  scale_x_log lm linear model regression model summary coef exp r-squred 회귀분석 회귀 분석
                  선형모델 선형 모델 rtvs ls
*******************************************************************************************************************/

-- Chart

-- Microsoft R

# **************************************************
# -- Packages used in this set of examples
# **************************************************
 
# Package      | Use
# ----------   | ----------
# ggplot2      | Plots
 
 
# ------------------------------
# -- Looking at Packages    
# ------------------------------
 
# You can extend the functionality of R by installing and loading packages.
# A package is simply a set of functions, and sometimes data
# Package authors can distribute their work on CRAN, https://cran.r-project.org/,
# in addition to other repositors (e.g. BioConductor) and github
# For a list of contributed packages on CRAN, see https://cran.r-project.org/
 
# -- List all available installed packages on your machine.
installed.packages()
 
# -- List all "attached" or loaded packages.
search()      
 
# -- You "attach" a package to make it's functions available, 
#    using the library() function.
library(foreign)
 
# -- You can get help on a package using:
library(help = foreign)
 
# -- To install a new package, use install.packages()
#    Install the ggplot2 package for it's plotting capability.
if (!require("ggplot2")){
  install.packages("ggplot2")
}

# -- Then load the package.
library("ggplot2")

# -- Notice that package:ggplot2 is now added to the search list.
search()


# **************************************************
# -- A Simple Regression Example 
# **************************************************

# -- Look at the data sets that come with the package.
data(package = "ggplot2")$results

# -- ggplot2 contains a dataset called diamonds. Make this dataset available using the data() function.
data(diamonds, package = "ggplot2")

# -- Create a listing of all objects in the "global environment". Look for "diamonds" in the results.
ls()

# -- Now investigate the structure of diamonds, a data frame with 53,940 observations
str(diamonds)

# -- Print the first few rows.
head(diamonds) 

# -- Print the last 6 lines.  
tail(diamonds)

# -- Find out what kind of object it is.
class(diamonds)

# Look at the dimension of the data frame.
dim(diamonds)


# ------------------------------
#  -- Plots in R
#     R has three systems for static graphics: base graphics, lattice and ggplot2.
# ------------------------------

# -- Create a random sample of the diamonds data.
diamondSample <- diamonds[sample(nrow(diamonds), 5000),]
dim(diamondSample)
nrow(diamondSample)
head(diamondSample)

# -- Set the font size so that it will be clearly legible.
theme_set(theme_gray(base_size = 18))

# -- In this sample you use ggplot2.
ggplot(diamondSample, aes(x = carat, y = price)) + geom_point(colour = "blue")
ggplot(diamondSample, aes(x = color, y = depth)) + geom_point(colour = "red")

# -- Add a log scale.
#    데이터가 몰려 보기 어려울 때 로그 변환으로 자료를 퍼트림.
ggplot(diamondSample, aes(x = carat, y = price)) + geom_point(colour = "blue") + scale_x_log10()

# -- Add a log scale for both scales.
ggplot(diamondSample, aes(x = carat, y = price)) + geom_point(colour = "blue") +
  scale_x_log10() + scale_y_log10()


# **************************************************
# -- Linear Regression in R 
# **************************************************

# -- Build a simple regression model
model <- lm(log(price) ~ log(carat) , data = diamondSample)       

# -- Look at the results.     
summary(model)                                        # R-squared = 0.9334, i.e. model explains 93.3% of variance

# -- Extract model coefficients.
coef(model)
coef(model)[1]
exp(coef(model)[1])                                   # exponentiate the log of price, to convert to original units

# -- Show the model in a plot.
ggplot(diamondSample, aes(x = carat, y = price)) + geom_point(colour = "blue") +
       geom_smooth(method = "lm", colour = "red", size = 2) +
       scale_x_log10() +
       scale_y_log10()


# ------------------------------
# -- Regression Diagnostics 
# ------------------------------

# -- Look at some model diagnostics.
#    check to see Q-Q plot to see linearity which means residuals are normally distributed
par(mfrow = c(2, 2))                                  # Set up for multiple plots on the same figure.
plot(model, col = "blue") 
par(mfrow = c(1, 1))                                  # Rest plot layout to single plot on a 1x1 grid


# ------------------------------
# -- The Model Object 
# ------------------------------

# -- Look at the model object. 
str(model)
model$coefficients                                    # note this is the same as coef(model)

# -- Now fit a new model including more columns
model2 <- lm(log(price) ~ log(carat) + ., data = diamondSample) 
                                                      # Model log of price against all columns
# -- R-squared = 0.9824, i.e. model explains 98.2% of variance.
#    i.e. a better model than previously
summary(model2)

# -- Create data frame of actual and predicted price
predicted_values <- data.frame(actual = diamonds$price, 
                               predicted = exp(predict(model, diamonds)) 
                              )                       # anti-log of predictions

head(predicted_values)

# -- Create plot of actuals vs predictions
ggplot(predicted_values, aes(x = actual, y = predicted)) + 
       geom_point(colour = "blue", alpha = 0.01) +
       geom_smooth(colour = "red") +
       coord_equal(ylim = c(0, 20000)) +              # force equal scale
       ggtitle("Linear model of diamonds data")