library(RCurl)

# urlfile <- 'https://archive.ics.uci.edu/ml/machine-learning-databases/gisette/GISETTE/gisette_train.data'

# x <- getURL(urlfile, ssl.verifypeer = FALSE)

# BD_Raw<-read.table(textConnection(x), sep = '', header = FALSE, stringsAsFactors = FALSE)

BD_RAW <- read.csv("C:/STAT/BRIDGE/Module-3/Week-3/PCA_BIG_DATA/PCA_BIG_DATA/BD_RAW.csv")

print(dim(BD_Raw))

# View(BD_Raw)

getwd()

# write.table(BD_Raw,"C: [---- FOLDER NAME HERE ---] /BD_RAW.csv",sep=",")

#

# Checking for Features / Columns with --  . 

#  -- 1) few unique values relative to other values within feature

#  -- 2) Near Zero variance or high frequency of most common value or mode value 

#  -- before we conduct PCA [ Principal Components Analysis ] this can be considered as a First Step for Feature Reduction 

#

library(caret)

#

Ftr_ZeroVAR<-nearZeroVar(BD_Raw, saveMetrics = TRUE)

class(Ftr_ZeroVAR)

#-- Uncomment ---# ?class

## here - nearZeroVar is a function of library(caret)

##-- Uncomment ---# ?nearZeroVar

# nearZeroVar diagnoses predictors that have 

# one unique value (i.e. are zero variance predictors) 

# or predictors that are have both of the following

# characteristics: they have very few unique values

# relative to the number of samples and the ratio of

# the frequency of the most common value to the frequency 

# of the second most common value is large

print(paste('Range:',range(Ftr_ZeroVAR$percentUnique))) 

# printing the "Range" of the nearZeroVar 

max(Ftr_ZeroVAR$percentUnique)

min(Ftr_ZeroVAR$percentUnique)

# We can see that the RANGE of VARIANCE is between - 

# Min == 0.0166 or Near Zero Variance. 

# Max == 8.6 or NOT Near Zero Variance. 

print(head(Ftr_ZeroVAR))

str(Ftr_ZeroVAR)

#

print(paste('Column count prior removal:',ncol(BD_Raw)))

##-- Uncomment ---# Alternative Code to PRINT Command - xx <- paste('Column count prior removal:',ncol(BD_Raw))

##-- Uncomment ---# xx

##-- Uncomment ---# This Code will Not Print ---- paste('Column count prior removal:',ncol(BD_Raw)

# We count Number of Columns / Features in RAW Big Data 

# We remove features with less than 0.2% variance 

# this is done by checking within vector - percentUnique  - which shows values for 

# percentage of unique data points out of the total number of data points

# if the % unique value < 0.2% we drop the Feature / Column. 

# Printing dimensions / number of Features - % unique value > 0.2% , 0.5% 

dim(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 0.2,])

dim(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 0.5,])

# SKIP FOR WEBINAR -1 -- Printing dimensions / number of Features - % unique value > 

# dim(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 1,])

# dim(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 8,])

# dim(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 8.6,])

# Initial BIG DATA had - 5000 Features or Columns and 6000 ROWS or Observations 

# if we choose 0.5% as the Cutoff we drop off - (5000 - 4123 ) == 877 Features , we shall have 

#  6000 ROWS or Observations in  4123 Columns or Feature - Simple Feature Reduction before PCA . 

BD_nzv <- BD_Raw[c(rownames(Ftr_ZeroVAR[Ftr_ZeroVAR$percentUnique > 0.5,])) ]

print(paste('Feature count after cutoff of Near Zero Var Features:',ncol(BD_nzv)))

# Now we conduct PCA using the Dimensions / Features Reduced data frame - BD_nzv

# We check the Dimensions of BD_nzv

print(dim(BD_nzv))

library(doParallel)

===========================R CONSOLE DUMP ================================ WEBINAR -1 =============== #No_Jargon