R Manual Data Management.Rmd

---
title: "Data Analysis Manual for R"
author: "Abu Nayeem"
date: "September 9, 2014"
output: html_document
---

##### Introduction

This Manual is meant to consolidate my knowledge in data analytics. In addition, I am happy to share data techniques and strategies to anyone. Many manuals focus on small datasets, the strategies here will try to focus on Big Data. There are different types of "big"" datasets; 1) many observations [i.e. rows], which is not too bad; 2) many columns which are more difficult as you need to assess if columns are useful or not or it is worth tinkering with. In addition, simple subsetting operations now become more tedious and time-consuming.  

##### Unresolved Questions


##### Update History
MAC OS
R version 3.1.1 (2014-07-10)
R Studio version 0.98
Number of cores: 2 [Use command on terminal: system_profiler | grep -i 'Cores']

##### R markup tools

Link: [Link](http://groupware.les.inf.puc-rio.br/har)
Bold text: **bold** 
Bullet point: *
* bullet
special designation: `rpart`
Italic: *Italic*

chunk commands:
```{r name} 
# you can name name it 
```
echo=FALSE -run but hides code great for graphs
messages=FALSE -run but does not show message of loading
warnings=FALSE - run but does not show warnings

autodep=FALSE

eval= FALSE -does not run the chunk code, still shows
results='hide' -run but does not show the output results
results='hold' -show output once all chunk written down 
fig.show='hold' -hold figure output until end of code


#####1) Preparation

A) Set working directory: 

A good default is create a working directory for the specifc folder. Make sure in quotes and ./ is attached to document. The code below state if not in the designated directory the create a new directory regardless of the starting point.
```{r, message=FALSE}
if (!getwd() == "./out-of-box-samples") {
    dir.create("./out-of-box-samples")
    setwd("./out-of-box-samples")
}
```

Suppose you want to create main directory & sub-directory for output
```{r, eval=FALSE}
dir.create(file.path(mainDir, subDir), showWarnings = FALSE)
setwd(file.path(mainDir, subDir))
```

B) Clear Work Space- No clutter please
```{r, message=FALSE}
rm(list = ls(all = TRUE))
date() #set date
```

C1) Install packages that may me missing on computer
```{r}
if (!library("data.table")) {
  install.packages("data.table")
}

if (!library(data.table)) {
  install.packages("data.table")
}
```

C2) Load Appropiate packages- Load the most important packages that you use frequently 
```{r, message=FALSE}
library(plyr) # load plyr first, then dplyr 
library(lubridate) # a must program in handling dates
library(dplyr) # for fancy data table manipulations and organization; Note it loads plyr as well
library(caret) # my preferred machine learning package in R; includes ggplot2
library(R.utils) # a program occasionally used to extract weird file types
library(reshape2) # for splitting column entries
library(DataCombine) # used to create lagged or forwarded varaiable
library(data.table) #useful for flexible data tablefeatures
```

D) Scripting: Prepare a script page to jot down all the steps. I recommend having it as a markdown file in R because you can shrink code chunks. Other statistical packages have their own scripting procedure. Certainly don't use a text file. 

E) Data Dictionary: For raw data and supplementary analyzes, a data/coding dictionary is crucial in making the dataset both understandable and accessible to other users. KNOW the size of the data and if your computer can handle opening it up

F) Optimizing Code/ Debugging : For big datasets, we want the code to run as fast as possible but still be readable; R by default only uses a single core when operating, you need to get access to new libraries to use multiple cores. Finally parallel processing allows access to use multiple cores
```{r, message=FALSE}
system.time(names(mtcars)) # gives some values of the time it takes to execute an action; Note you can take a huge chunk of code as well
summaryRprof() # gives time slots where most of the computing going to, its helpful for long functions
print(object.size(x), units = "auto") #memory use for any particular object
```

Debuging Example
```{r}
fun1 <- function(x) {
     x <- x * 2
     fun2(x)
}
fun2 <- function(x) {
     fun3(x)
}
fun3 <- function(x) {
     apply(x, 1, mean)
}
```    

1) We can use ``debug()`` to step through a function line by line
2) After an error occurs, we can use ``traceback()`` to look at the *call stack*
3) More helpfully, if we set ``options(error = recover)`` before running code, we can go into the function in which the error occurred
4) We can insert ``browser()`` inside a function and R will stop there and allow us to proceed with debugging statements
5) You can temporarily insert code into a function (including built-in functions) with ``trace(fxnName, edit = TRUE)``

G) Constructing your own functions: Unless you are in the cutting edge or face redundant operations or handle similar datasets, then it worth the time to write. R package functions provide greater protection against faulty code

H) General Stylized Rules:

Google R Stylized Rules: http://google-styleguide.googlecode.com/svn/trunk/Rguide.xml
Unit Testing Rules for functions: http://master.bioconductor.org/developers/how-to/unitTesting-guidelines/
Git Setup: file://localhost/Users/abunayeem/Desktop/R%20Directory/paciorek-r-bootcamp-2013%20(1)/modules/module9_workflows.html
For R as a programming language [it is recommended python handle it]: file://localhost/Users/abunayeem/Desktop/R%20Directory/paciorek-r-bootcamp-2013%20(1)/modules/module10_advanced.html

###2) Extracting Data

A) Pre-Research:
Know what data and/or question you are interested in. This is particulatly important when datasets are absolutely massive

B) Extraction: 
You want your code be able to replicated anywhere either fixed file on library or the alternative online access [github repository]. There are numerous extraction options. It is helpful to use the Rstudio interactive feature of importing data to see a subset of the data, which can inform you of the dividers, missing variable, missing a column or not Note: By giving data.table more info in advance you reduce computing time. 

```{r, message=FALSE}
url <- "XXXXXhttps://d396qusza40orc.cloudfront.net/repdata%2Fdata%2FStormData.csv.bz2"
download.file(url, "repdata-data-StormData.csv.bz2", method="curl") 
#"curl" is needed for mac
bunzip2("repdata-data-StormData.csv.bz2",dest="repdata-data-StormData.csv", overwrite=TRUE, exdir="./Repository Important")
#Set location (exdir) to working directory
portion<-file("repdata-data-StormData.csv","r") 
#make it to readable file for extraction, so we can extract certain elements; useful for big files
StormData<-read.csv(text = grep("^[1,2]/2/2007",readLines(portion),value=TRUE), header=FALSE, sep=",", comment.char="") 
#we are looking for the text that match and extract those rows. This works for time, another way is take explicit number of rows;
#other features include skip, nrows, stringsAsFactors, na.strings,and ColClasses, the last item is to store the data with less RAM 
colnames(Power)<-t(Columns[1]) 
#the transposed matrix of one is appeneded to primary dataset as column names
```

    Extraction of zip with multiple files:
```{r}
temp <- tempfile()
download.file("https://d396qusza40orc.cloudfront.net/getdata%2Fprojectfiles%2FUCI%20HAR%20Dataset.zip ",temp)
unzip(temp, list = TRUE) # This shows the list of files name!
YTest <- read.table(unzip(temp, "UCI HAR Dataset/test/y_test.txt"))
unlink(temp)
```

    SQL Method: Extraction
```{r, message=FALSE}
library('sqldf')
file <- "household_power_consumption.txt"
dates <- "SELECT * from file WHERE Date = '1/2/2007' OR Date = '2/2/2007'"
# getting the records from the file filtering with the SQL sentences
data <- read.csv.sql(file, sql=dates, sep=";")
```

    Strategy: Take a small sample and the figure out the classes of the columns and repsond accordingly 
```{r, message=FALSE}
file<-"household_power_consumption.txt"
a<-read.table(file,nrows=10) # create a smaller subset of data
classes<-sapply(a,class) # separate the classes
b<-read.table(file, colClasses=classes) #now classes are given much more in advance
read.table(file, colClasses=c("character","character",rep("numeric",times=7))) # if you know the class of the tables then this is better
```

    Loading previous workspace: `source(data.R)`

### Data Cleaning/ Presentation

Assume you were given raw data. How do you clean the data to make it more accessible? 

A) What is the task given to you? Depending on the question you want to answer you may only clean a small portion of the entire dataset. However, to redo analysis with same raw dataset, we need to reclean once again. Thus a global cleaning process is recommended. However, computing power may be limited and priorities may be placed elsewhere.

B) Basic Overview
```{r, message=FALSE}
Cars<- tbl_df(mtcars) #this type of dataframe computationally faster and accessible 
str(Cars) # show columns associated class
names(Cars) #column names
summary(Cars) # get a feel if there is missing data
dim(Cars)# the dimensions of your data
CC<-Cars[1:3,] ;CC$A<-c("Tom","DOG","poop") ;CC$B<-c("ToM","DOG","PoOp") # easy dataframe to work with
table(CC$A) #this gives quick frequency of a variable
head(CC); tail(CC)
```

C) Handling Time

    Lubridate [Convert character to date variable]
```{r, message=FALSE}
library(Lubridate)
ymd("1989-05-17") #it is very smart and maneurable in handling dates as order don't matter!
```
    
    Convert Date and Time together
```{r, message=FALSE}
DateTime <-paste(X$Date,X$Time)
X$DateTime <-strptime(DateTime, "%d/%m/%Y %H:%M:%S")
```
    Find day of the week
```{r, message=FALSE}
weekdays(FullActivity$date) # it takes the date and use an internal calendar to figure out what day of the week it is
```


D) Removing Duplicate or Weird Observations 

    Removal of duplicate columns :
```{r, message=FALSE}
CC[, !duplicated(colnames(CC))] #[Assumed they have same values in entries]
```

    For Character Entries:
    The criteria of removal should follow some guidelines
    1) When there is no unique ID, then a identical datapoint can be valid. An index can help to show a unique observation rather than an error  
    2) If a primary ID exists do the following, prior to hitting unique 
```{r, message=FALSE}
duplicated(CC) # this highlights which variables are completely identical
a<- duplicated(CC, by=key(ID)) # spot unique data by the unique only rather than all the variables
CC[a,] #make sure you compare the variables and see any anomalies
CC[-a,] #I recommend choosing which of the identical data is worth removing
```

    3) For Numerical Entries [same methods as above] :
    Use the summary feature, to spot any oddities on the numeric vectors which include unreasonable outliers and contradictory values such as negative values [example: price]
```{r, message=FALSE}
summary(CC)
select(CC, wt>0, mpg<20) # this subsetting remove observations from the dataset, so be wary what you remove
```

E) Converting Classes: Classes should be changed early upon extraction but if not we can do the following

    Example: the number of cylinders (cyl) in mtcars data may be better suited as a factor since the number of cylinders is discrete.
```{r, message=FALSE}
CC$Cyl<- as.character(CC$Cyl) # convert integer to character; do your manipulation with entries
CC$Cyl<- as.factor(CC$Cyl) # convert character to factor, notice the summary feature provides useful information
CC$Cyl<- as.numeric(CC$Cyl) # factor to integer, the circle of class is complete
```

    Suppose we face the issue that the orginal factor variables are number strings. You need to be really careful in converting it to numeric
```{r}
CC[, 1] <- round(as.numeric(as.character(CC[, 1])), 3) # this works for both factors, but also notice I used the round function which makes it more pleasant to see data
```
    Note: 
    1) When creating mixed vectors, R chooses the lowest value of the vectors (i.e. character) to determine the class; also take note of list and vector class. This is important especially if introducing a string entry to a numeric column 
    2) Watch out for missing values when cconverting classes
    3) For some factor variables may be ordered, so assign it when converting as shown below:
    4) Factor variables can serve as dummy varaibles
```{r, message=FALSE}
CC$A<- as.character(CC$A) # convert integer to character; do your manipulation with entries
CC$C<- factor(CC$A) # this creates new variable but the default level is based alphabethic 
CC$D<- as.factor(CC$A, levels=c("poop", "dog"))# the new variable is now ordered with poop at highest level
```

F) Upper and lower Case of strings. This is useful if implementing string manipulation

    Column Names: we want lowercase all the letters and then capitalize it with a function
```{r, message=FALSE}
names(CC)<- tolower(names(CC)) #lower case all strings on columns
capFirst <- function(s) {
    paste(toupper(substring(s, 1, 1)), substring(s, 2), sep = "")} 
# this function takes a string and capitalize only the first letter
names(CC)<-capFirst(names(CC)) # Implement Capitalization Function
```

    Single Row Entries: 
```{r, message=FALSE}
CC$A<- tolower(CC$A) # lower case all strings in the column
CC$A<-capFirst(CC$A) # capitalize function works in this case too
```

    Multiple Row Entries: converting one column variable at a time is tedious. The code checks multiple columns and then lower case them accordingly. 
```{r, message=FALSE}
CC<-data.frame(lapply(CC, function(v) {
  if (is.character(v)) return(tolower(v))
  else return(v)
}))
```
    #I have not figured a good way to capitalize all variables at once, though probably not that useful

    Removing empty space in column names: When calling on column IDs the empty space can cause problems not only in R but in other languages and functions.
```{r, message=FALSE}
names(CC)<-gsub("","_",names(CC))
```

G) Handling Missing Values [Make sure you find what previous text file described missing values while importing data]. Data that does not comply should become missing data. Remember missing data can be ignored in certain cases. We may want to keep missing data if they hold some importance or could be filled in later

    Spotting:
```{r}
CC$test<-c(2,5,NA)
CC$dumm<-c(3,4,5)
colSums(is.na(CC))>0 #checks if each column has missing values
```
   
   Case 1: We have a lot of incorrect entries in a column, instead of handling each entry separately. We can treat each incorrect entry as missing value
```{r, message=FALSE}
a<-grep("[^t]",CC$B,value=FALSE) # find row index that does have those strings
CC[a,13]<-NA #we then set those values NA
```
    
    Case 2: Suppose we want to get rid of NA entries based from a specific column.
```{r, message=FALSE}
filter(CC,!is.na(B)) # keeps non-NA entries based on that column
```

    Case 3: Suppose you want to get rid of NA entries in all columns
```{r, message=FALSE}
good<-complete.cases(CC) # looks in entire dataframe for missing values
CC[good,] # the subset leaves only entries that has no missing values at all
```

    Case 4: Suppose we want to get rid of columns with missing values
```{r}
library(data.table)
CC <- as.data.table(CC)
CC[,which(unlist(lapply(CC, function(x)!all(is.na(x))))),with=F]  # this gets rid of ALL columns with ONLY NA values
a<- which(colSums(is.na(CC)) > 0) # the WHICH function creates a numerical index for the logical test
CC[,-a] #notice there is a negative sign which is to remove the columns that had spotted NA values 
```
NOTE: the which method is needed for logical vectors involving column index, but row index can take logical vectors fine


Note: NaN is missing integers and NA change class type; NaN and NA are Not the same

H) How to shift columns around/ and extract row names. Note: We want the unique ID to be at leftmost.
```{r, message=FALSE}
select(CC,3,2,1,ncol(CC):4) # I rearranged the first three columns, and kept the remainder the same, but backward order; Note: this is actually subset, so make sure all cols are still there 
CC$names <- rownames(CC) # this extracts the row names
rownames(CC)<-NULL # this deletes the row name. Make sure to extract first.
```

I) Sorting within dataset
```{r, message=FALSE}
arrange(mtcars, drat,desc(q)) # double layer sorting where intial value is ascending and upon that it is descending; Application is useful for multi-level lagged variables
```

J) Sometimes it is useful to create an index, but make sure to sort beforehand if intention to do something
```{r, message=FALSE}
CC<- arrange(CC,Wt) #arrange data frame
CC$Index <-seq(along=nrow(CC)) #create a sequence of values based on # of rows, so it picks up missing values as well
```

K) Rounding Numbers: It makes it much pleasant looking
```{r}
round(Cars$drat, 1) # a single column
Cars[, c(5,6)]<- round(Cars[, c(5,6)], 1) # multiple columns and subsetts
```

### Data Manipulation

A) Meticulous Data-Mining:
    The column names are easy to check if in proper order, but the entries can be jumble mess.

1st level: For essentially all variables you check if the values are correctly specified

    For mispelling of character entries: Use gsub or grep to fix
```{r, message=FALSE}
list(unique(CC$A)) # this shows the unique values and you can spot which ones need to be changed; Use grep
```

    For spotting weird numeric entries:
```{r, message=FALSE}
summary(CC$wt) # check if max or min fit desired criteria [no missing values or negative numbers]
all(data$val>0) # check if there is any negative values on the dataset, it gives only one statement
```

    For finding out frequency of values of strings [place into a factor first]
```{r, message=FALSE}
CC$A<- as.factor(CC$A) # change character vector to a factor
summary(CC$A) #Now we view the frequency of each factor variable
table(Cars$Cyl, useNA="ifany") # General form of looking frequency and adds a column if there is any NA
```

2nd Level: When you believe certain pairs of columns (factors) should be matched together. 

    For example: A teacher (teacher) should be assigned to a several specific classrooms (class)
```{r, message=FALSE}
table(CC$teacher, CC$class) # this creates a cross-sectional frequency table; Note any unusual pairings
```

3rd Level: This is higher level matching and there is probably a sleek function to test it all at once. Most of the time, the 2nd level data cleaning will catch most of the proper matches.

B) Subsetting and filtering 

    Columns:
```{r, message=FALSE}
mtcars[,c(3,4,5,7)] # standard subsetting method
mtcars[,3]; mtcars[3] # this is WRONG. 
select(mtcars,3:5,7) # I only took cols 3 to 5 and 7
#select features also include other arguments like start_with(); ends_with(),contains(),match(),num_range()
select(mtcars,starts_with("d"), more argument) #choose the columns that only start with d
select(mtcars,-starts_with("d")) #choose the columns that not start with d WITH the minus sign
select(mtcars, contains("t")) # looks for columns that has t
```

    Column Subsetting with grep index strategy
```{r, message=FALSE}
a<-grep("^d",names(mtcars),value=FALSE); mtcars[,a]; select(mtcars,a) # grep function allow to choose string, we use that index vector (value=FALSE) to subset
grep("^d",names(mtcars),value=TRUE) # this shows the list of names where there is a match
a<- names(mtcars)>"g" #also creates an index but more limiting
```
    
    Rows:
```{r, message=FALSE}
mtcars[c(2,5,6),] # the standard format of taking three rows
filter(mtcars,drat>3.10, qsec>17) #filter on two conditions solved simultaneously!
filter(mtcars,drat>3.10 | qsec>17) #two conditions with an OR aspect
filter(CC, colA>"g") # you can filter by strings and there is greater than and less than respect to alphabet on FIRST character only
filter((CC, colA>"oo")) # it test the first letter if satisfied as a tie than to second line from first character
a<-grep("oo",CC$A,value=FALSE); CC[a,] # for greater precision [i.e exact matches of strings] you can use grep strategy again
a<-mtcars$wt>2.5 & mtcars$hp>100 # this creates a logical index similar to grep function, but we can use multiple arguments
```

    Rows & Columns: Protip you can combine grep function indexes (or any indexes) at the same time
```{r, message=FALSE}
mtcars[1,5] # the entry value from first row and 5th column; class is a number
mtcars[c(1,10),c(5,6)] # return a data.frame with first row and 5th and 6th column
mtcars[a,b] #each variabe is an index to subset data by columns and rows but they satisfied some condition
```

    Lists:
```{r, message=FALSE}
x[[2]] # returns the elements of a list, rather than the list item itself
```
  
C) Adding new Data; Note r handles merging a lot better than most statistical software

    Columns:
```{r, message=FALSE}
mutate(mtcars,A=1:32,am_gear=(am*gear)/2) # first column A correspond to any vector; am_gear represents an operation to create new column  
```

    Adding lagged/leading columns: 
```{r, message=FALSE}
a<- slide(CC,Var="A",slideBy=-1, NewVar="lag1wt") # uniform lag 1
a<- slide(LagDaySteps,Var="A",slideBy=1, NewVar="lead1wt") # uniform lead 1
Grouplag<-arrange(Grouplag, cyl) # when doing a first layer lag you need to sort
slide(Grouplag,Var="wt",slideBy=-1, GroupVar="cyl", NewVar="lag1wt") # lagged 1 by cyl by group;
```

    Rows: A cumbersome ordeal since the row vector need to have same number of rows as the data frame
    Row merging:
```{r, message=FALSE}
rbind(mtcars,Cars) # number of columns AND column names need to be the same, though order of columns don't matter; Note in many loops rbind is used frequently to build up observations from an empty matrix 
```

    Column merging:
```{r, message=FALSE}
cbind(mtcars,Cars) # number of rows need to be same not to intersting 
```    

    Data.frame merging: No doubt the most crucial merging- we merge on basis of a common ID. [ORDER don't matter]
    Case 1: Unique Merging- this is the case where there is unique ID for both datasets    
```{r, message=FALSE}
merge(mtcars,Cars,by="ID") # since the IDs are unique, the merged matrix consist of only EXACT matches
```      

    Case 2: Unique Merging with keeping non-matched data [i.e. unmatched ID]
```{r, message=FALSE}
merge(mtcars,Cars,by=c("ID","blue"),all.x=TRUE,all.y=TRUE) # IDs can be multipe arguement, but notice the additional values are meant to not throw out unmatched pairs. Many of the values will become NA 
```

    Case 3: Non-Unique Merging: Suppose matrix A has unique ID where matrix B has some duplicate ID but with different values
```{r}    
A<-data.frame(id=1:5,b=6:10)
B<-data.frame(id=c(1,1,2,2,3,3,4,4,6,6),c=1:10) #note it has multiple ids
merge(B,A,by="id") # shows all unique combinations of the id
merge(B,A,by="id",all=TRUE) # include ALL IDs that did not match
```

D) Splitting Existing Columns: when there is excess information in one column

```{r, message=FALSE}
df <- data.frame(country =c("Angola","Belize"),dates = c("1951-1953", "1970-1972")) # Our goal is want to separate the years from start date to end date and place value
df <- cbind(df, colsplit(df$dates, "-", c("start", "end"))) # we take column and split it with "-" for two variables; to get a sequence

#split sequence
ddply(df, .(country), function(x){
            data.frame(
                    country=x$country,
                    yrobs=seq(x$start, x$end),
                    yrstart=x$start,
                    yrend=x$end
            )
        }
)
df
```

E) Replacing variables and strings

    Column Names:
```{r, message=FALSE}
names(CC)<-gsub("q","Quarter", names(CC)); # it finds matching string and replace with that string ; useful shortening names or expanding column names; NOTE this an easier way to change a col
names(CC[7])<-"Quarter"
```

    Row Entries: we can use the index strategy once again
```{r, message=FALSE}
Cars$Disp[Cars$Drat<3] <- 0  # we change the value in the discipline entry if certain condition satisfied
a<-Cars$Drat<3 | blah ; Cars$Disp[a]<-0 # index method can work which means grep can work too
```

F) Grouping and Collpasing Datasets: Grouping is convenient when we want to keep the dataset pristine and able to aggregate values. Rules for collapsing data: 1) Figure out the primary new ID/factor 2) Perform all the duties because you CANNOT go backward  

    Case 1: Grouping upon a factor/group to find mean across groups; Don't use dcast function, it's not too great
```{r, message=FALSE}
CCGroup<-group_by(CC,Cyl) # we create a new function which automatically register Cyl as the "group"
summarize(CCGroup, mean(Wt),sum(Qsec)) # this takes the mean and sum calculations and displays it by group basis
```

    Case 2: Grouping by more than one layer; also collapsing dataset [always arrange appropiately and watch out for missing]
```{r, message=FALSE}
ddply(CC,.(Cyl,Carb),summarise,Aver=mean(Drat)) # it groups cycle and Carb and takes the mean of Drat for each subcategory

# More quick ay doing it but you are limited to one operation
outData<-aggregate(selectedData,by=list(Activity,Subject),FUN=mean,na.rm=TRUE)

#another way to do above
Master.dt[,lapply(.SD, mean),by='participants,activities'] #This takes the mean of EVERY column broken down by participants and activities; Great for big data but not useful for data analytics

# Another way to sum up values respect to a group
xtabs(drat~cyl,mtcars) 

```

    Saving new table externally in some sort of text file:
```{r, message=FALSE}
write.table(TidyData,file="Tidy.txt", row.names=FALSE)
```

G) Handling Missing values: Remember the feature "na.rm=TRUE" exists for many functions!
```{r, message=FALSE}
colSums(is.na(data))
```

H) Loops and Advanced: Avoid if necessary

    Loop Example: MAKE sure it is (i in [SEQUENCE]), a good range is 1:length(unique(CC$Hp))
```{r, message=FALSE}
x=c("a","b","c")
for (i in seq_along(x)) {print(x[i])} # this automatically fits to the length of the vector
```
    
    WHILE using a random walk example: the code simply states if the draws reach 3 or 10 this code stop running
```{r, message=FALSE}
z<-5
while(z>= 3 && z<=10) {
  print(z)
	coin<-rbinom(1,1,0.5)
	if (coin==1) {
		z<- z+1
	} else {
		z<- z -1
	}
}
```

    Repeat: creates an infinite loop until you use break to stop it, it is good when using tests for convergence, but we never know if it will converge or not. Other features inlcude next which skips iterations
```{r, message=FALSE}
repeat {
  x1<- computeEstimate()
	if (abs(x1-x) <tol) { break
	} else {
		x<-x1
	}}
```

I) Functions: Avoid unless necessary or redundant work

    Sample Function: Good base example of a function that updates. Another approach is to continually append the vector for each loop. If we were trying to get a data.frame of observations than very likely we would use rbind. 
```{r, message=FALSE}
colmean <- function(y, removeNA= TRUE) { # removeNA has a default value and y is a matrix
  nc<- ncol(y) #nc is the number of cols in matrix y
	Means <-numeric(nc) # empty vector of size of number of rows
	for (i in 1:nc) { # the loop will be going through every column
		Means[i]<- mean(y,[,i], na.rm=removeNA) #it will take the mean of a column and store it under the varaible Means
	}
	Means # it returns the final product with all means
}
```

    Lexical Scoping in R: The difference between lexical vs dynamic scoping is that the global variable is resticted to parent environment under the lexical scoping

    Example of a DOUBLE layer nested function ; use cautiously. Where each layer needs to be defined if to be used
```{r, message=FALSE}
make.power<- function(n) {
    pow<- function(x) {
		x^(n)
	} pow
}
quad<- make.power(4) # first you need to determine the value of exponent n
quad(8) #second you put a place value
```

    Using lapply (challenging to use), sapply (useful), tapply (restrictive, plyr package better), split (pylr package better)
```{r, message=FALSE}
lapply(CC, mean) # treat the data frame as a list and implment action on each column
sapply(CC[a,], mean) ### same as lapply but puts out on vector. This a quick way of getting column means of every variable and with with index you can restrict rows/columns accordingly
apply(CC[,c(1:11)],2,mean) # these function cannot handle string variables, also two means that columns are kept
apply(CC[,c(1:11)],1,quantile, p=c(0.25,0.75)) # now rows are kept and columns are collapsed, in addition I created a quantile ; for sums and means there are quicker function calls rowMeans (etc),
tapply(CC,A,mean) # for tapply to be used the number of factor observations is identical [USE ddplyr instead]
tapply(CC$Mpg,CC$Hp, mean) # both variables are of same length so it can work
split(CC,Cyl) # creates a list of dataset broken down by factors A
split(CC,list(Carb,Cyl),drop=TRUE) # creates a sublist similiar to dplyr, but try every COMBINATION; drop equal to true is to remove empty combinations
lapply(split(CC,Cyl),mean) # does not handle dataframes too well, so use dplyr
```

J) Useful functions [Not Useful- Cross tabs]

    Sequence:
```{r}
seq(1,10, by=3) # 1-10 increase by 2 ; counts by three starting from 1
seq(1,10, length=6) # it now takes this range and splits it perfectly evenly
quantile(tree$LFBM)
```

    Literal Commands [GREP]:
```{r}    
literal # exact matches
^i think # specify the beginning of sentence
morning$ # specify the ending of sentence
[Bb][Uu][Ss][Hh] # capitalization does not matter
^[Ii] am # combined the two
^[0-9][a-zA-Z] # sentence begin with number and word
[^?.]$ # when symbol inside it says if not ? or period at end then show
9.11 # the dot represents any character
flood| fire | coldfire  # it's an alternative [i.e. or]
^[Gg]ood | bad # only good is restricted to beginning of line
^(good|bad) # transitive property
Hey ( [Ww]\.)?  # the question mark indicates it is optional
(.*) # looking for parenthesis, not concerned about anything
# you can differentiate the number of words apart
# you can find repetitions of a phrase in a sentence
```
    
    Unique Indices: For extractions in code
```{r}
unique(CC$Wt)[1] # the output showed the unique value (I presume alphabetically) 
```    

### Presentations and basic plot displays

A) Plotting: the three major packages is base plot, lattice, and ggplot (favorite); Exporting the plots in different formats do matter so be aware and remember to close the screen browser after savinf a photo. JPEG and PNG (poor resize); pdf (good at resizing)

    Base Plotting: 
```{r, message=FALSE}
par(mfrow=c(2,2)) # for the graphics windows we create two rows and two columns [i.e. 4 graphs will exist on same graphics device ] 
with(Power,{ #the curly brackets enable multiple plots to be implemented at ease as long base data being used
  plot(DateTime, Global_active_power, type="l", ylab="Global Active Power") # Plot 1; type l represent use line graph
  plot(DateTime, Voltage, type="l") # Plot 2
  plot(DateTime,Sub_metering_1, type="l", ylab="Energy Sub Metering",col="grey") #Plot 3; this will have multiple plot on same table
    points(Power$DateTime,Power$Sub_metering_2, type="l",col="red") # within plot 3 but with different datapoints
    points(Power$DateTime,Power$Sub_metering_3, type="l",col="blue")
    legend("topright",lty=1,bty="n",cex=0.4, col=c("grey","red","blue"),legend=c("Sub_metering_1","Sub_metering_2","Sub_metering_3")) # legend where cex determine text size ; lty determine the shape of symbol
  plot(DateTime, Global_reactive_power, type="l")
})
dev.copy(png,filename="plot4.png") # this saves the plot image to an external file
dev.off() #MAKE SURE you close this browser
```

    Qplot: The most simplest plotting package in ggplot2 package
```{r, message=FALSE}
#Plot 1
by_Event_StormData <- arrange(by_Event_StormData, desc(TotMPropDmg)) # arrange the data
qplot(TotMCropDmg,TotMPropDmg,data=by_Event_StormData[c(1:10),], col=Event, xlab="Total Crop Damage ($Million)",ylab="Total Property Damage ($Million)", main="Most Harmful Events to Economy") # notice I used the index to pick certain data; the default in qplot of two variables is scatterplot; col=Event colors my data respect to associated Event. See the first R-pub document
#Plot 2
qplot(date,TotSteps,data=TotalDay, geom="histogram", stat="identity",main="Total Steps Vs Time With NA Values", ylab="Total Steps")
# the geom is crucial in determining what kind of plot you want to produce. see help or online to see all feature available

qplot(drat, wt, data=mtcars, facets ="cyl~.") # three plots in row spearated by cyl
qplot(drat, wt, data=mtcars, facets =".~cyl") # three plots in column separated by cyl

```    

    GGPlot:
```{r}
ggplot(data = , aes(x = , y = , color = , linetype = , shape = , size = ))
plot1 = ggplot(data = mydata.subset, aes(x = year, y = vturn, color = )) + geom_line(aes(color = country))
plot2 = ggplot(data = mydata.subset, aes(x = year, y = vturn, linetype = )) + 
    geom_line(aes(linetype = country))
plot3 = ggplot(data = mydata.subset, aes(x = year, y = vturn, shape = )) + geom_point(aes(shape = country))
grid.arrange(plot1, plot2, plot3, nrow = 3, ncol = 1)
ggsave(filename = , plot = , scale = , width = , height = )
```

    Saving Plots quick code:
```{r}
dev.copy(png, "Blah.png", height = 480, width = 480)
dev.off()
```

B) Tables are extremely useful but we do not want to rely on Excel all the time

    Basic Table: [I take advantage tbl_df]
```{r, message=FALSE}    
CC<-tbl_df(CC) #transform to a special dataframe
CC %>% arrange(desc(Wt)) %>% select(1,2,5,6) # this one of numerous ways to create a table, notice there is chaining involved %>%
```

    Advanced Table: TBA

### Advanced: Tidy Data Set: Three Rules; 

1) Each variable forms a column
Violation: suppose there is two columns to represent gender

2) Each observation forms a row
Violation: suppose there is two IDs that share many things except one column attribute and thus separate 
 
3) Each type of observational unit forms a table
More complex and structural based

A) Problem 1: you have column headers that are values

  grade male female
1     A    1      5
2     B    5      0
3     C    5      2
4     D    5      5
5     E    7      4

Issue: Gender count is in two columns, which is easy to view, but you won't be able to do linear regression well or factor analysis 
Solution: Gender should be in one column, and count on the other column
```{r}
gather(students,sex,count,-grade) # this creates two columns sex and count; it breaks down previous male and female columns into single column varaible; Note the minus sign before grade, which says we want to gather all columns EXCEPT grade
melt(data, id="grade") #this essentially does gather but doesn't allow naming, so you need to name separately; MELT can specify which variables you want to keep and which ones to gather, you can also remove which data is not useful and implement gather
```


grade    sex count
1      A   male     1
2      B   male     5
3      C   male     5
4      D   male     5
5      E   male     7
6      A female     5
7      B female     0
8      C female     2
9      D female     5
10     E female     4

B) Problem 2: column headers that are dual value

  grade male_1 female_1 male_2 female_2
1     A      3        4      3        4
2     B      6        4      3        5
3     C      7        4      3        8
4     D      4        0      8        1
5     E      1        1      2        7

Issue: Gender Count AND Class is on the column variables
Solution: we want separate sex and count into separate variables
```{r}
res<-gather(students2,sex_class,count,-grade)
# we create two columns, where sex_class will have four factors related gender and class and each respective entry is in count
melt(data, id="grade") # same as above
separate(res,sex_class,c("sex","class")) 
# we separate sex_class into sex and class. This operator is smart enough to see "_" exist in both column and entries or you set the separator
```
   grade    sex class count
1      A   male     1     3
2      B   male     1     6
3      C   male     1     7
4      D   male     1     4
5      E   male     1     1
6      A female     1     4
7      B female     1     4
8      C female     1     4
9      D female     1     0
10     E female     1     1
11     A   male     2     3
12     B   male     2     3
13     C   male     2     3
14     D   male     2     8
15     E   male     2     2
16     A female     2     4
17     B female     2     5
18     C female     2     8
19     D female     2     1
20     E female     2     7

C) Problem 3: when variables are stored in both rows and columns

   name    test class1 class2 class3 class4 class5
1  Sally midterm      A   <NA>      B   <NA>   <NA>
2  Sally   final      C   <NA>      C   <NA>   <NA>
3   Jeff midterm   <NA>      D   <NA>      A   <NA>
4   Jeff   final   <NA>      E   <NA>      C   <NA>
5  Roger midterm   <NA>      C   <NA>   <NA>      B
6  Roger   final   <NA>      A   <NA>   <NA>      A
7  Karen midterm   <NA>   <NA>      C      A   <NA>
8  Karen   final   <NA>   <NA>      C      A   <NA>
9  Brian midterm      B   <NA>   <NA>   <NA>      A
10 Brian   final      B   <NA>   <NA>   <NA>      C

Issues: 1) all these missing values, obviously it would be better to create a class variable; 2) the final and midterm grade gives information on two classes
Solution: You may be tempted to put each student on one line that would entail two class variable. The goal is to get as much information of a single class for a single student
```{r}
dat<-gather(dat,class, grade, class1:class5, na.rm = TRUE)
#OR
dat<-gather(dat, class,grade,-(name : test),na.rm = TRUE)
#create new cols class and grade which we mush colnames; na.rm=TRUE makes sure we ignore missing values; NOTE: test and name is not touched
dat<-spread(dat,test,grade) 
# it takes test and seprate it into two columns and put corresponding grades; NOTE this is opposite to gather!
mutate(dat, class= extract_numeric(class))
# the final step replace class entries with ONLY the numeric values which determine the class number
```

    name class final midterm
1  Brian     1     B       B
2  Brian     5     C       A
3   Jeff     2     E       D
4   Jeff     4     C       A
5  Karen     3     C       C
6  Karen     4     A       A
7  Roger     2     A       C
8  Roger     5     A       B
9  Sally     1     C       A
10 Sally     3     C       B

D) Problem 4: multiple observational units are stored in the same table

    id  name sex class midterm final
1  168 Brian   F     1       B     B
2  168 Brian   F     5       A     C
3  588 Sally   M     1       A     C
4  588 Sally   M     3       B     C
5  710  Jeff   M     2       D     E
6  710  Jeff   M     4       A     C
7  731 Roger   F     2       C     A
8  731 Roger   F     5       B     A
9  908 Karen   M     3       C     C
10 908 Karen   M     4       A     A

Issues: There is simply too duplicity where the first three columns combination are unique per ID. The issue is that it may take up a lot of space to each variable when it doesn't provide much insight

Solution: To save database/memory, we put identical characteristics of primary ID into one data.frame while the more dynamic charactersitics on another data frame. For example: Notice that many data libraries have two main dataframes; 1) One is performance with different observations etc., and the other is description of the "primary ID", which can be hospitals, banks, etc.

    Step 1: Save the primary ID characteristics [Used chaining (reduce amount of calling) this time around]
```{r}
student_info <- dat %>%
  select(id, name, sex) %>%  #find the matched charactersitics
  unique() %>% #keep only uniqued values [i.e. remove duplicates]
```
   id  name sex
1 168 Brian   F
3 588 Sally   M
5 710  Jeff   M
7 731 Roger   F
9 908 Karen   M

    Step 2: Save the dynamic charactertics, but make sure you still keep at least one primary ID variable
```{r}
gradebook <- dat %>%
  select(id,class,midterm,final)
```    
   id class midterm final
1  168     1       B     B
2  168     5       A     C
3  588     1       A     C
4  588     3       B     C
5  710     2       D     E
6  710     4       A     C
7  731     2       C     A
8  731     5       B     A
9  908     3       C     C
10 908     4       A     A

E) Problem 5: when a single observational unit is stored in multiple tables

> passed
   name class final
1 Brian     1     B
2 Roger     2     A
3 Roger     5     A
4 Karen     4     A

> failed
   name class final
1 Brian     5     C
2 Sally     1     C
3 Sally     3     C
4  Jeff     2     E
5  Jeff     4     C
6 Karen     3     C

Issue: These two separate dataframes is not needed one table can explain all the information needed.
Solution: We simply create a new column that state passed or failed
```{r}
passed<- passed %>% mutate(status = "passed") # create new col for each data frame
failed<- failed %>% mutate(status = "failed")
rbind_list(passed,failed) # combine the datasets by stacking observations
```
    name class final status
1  Brian     1     B passed
2  Roger     2     A passed
3  Roger     5     A passed
4  Karen     4     A passed
5  Brian     5     C failed
6  Sally     1     C failed
7  Sally     3     C failed
8   Jeff     2     E failed
9   Jeff     4     C failed
10 Karen     3     C failed

F) Problem 6: Suppose you had this mess of a dataset where there is many arguements that need to be tidy up. NOTE: the data belows show the count of individuals in that range

  score_range read_male read_fem read_total math_male math_fem math_total
1     700???800     40151    38898      79049     74461    46040     120501
2     600-690    121950   126084     248034    162564   133954     296518
3     500-590    227141   259553     486694    233141   257678     490819
4     400-490    242554   296793     539347    204670   288696     493366
5     300-390    113568   133473     247041     82468   131025     213493
6     200-290     30728    29154      59882     18788    26562      45350
Variables not shown: write_male (int), write_fem (int), write_total (int)

Issues: 1) Separate male and female; 2) Separate reading, math, writing; 3) how to handle total values
Solution: Use the gather strategy to separate the sex, part, count. In addition, you need to 
```{r}
sat %>%
  select(-contains("total")) %>% #remove total variables because they are just simply operations
  gather(part_sex, count, -score_range) %>% # keep score range- this should create extensive column list
  separate(part_sex, c("part", "sex")) %>% # separate part and sex together
  group_by(part,sex) %>% #its simply not re-ordering, but it allows some group function to be implemented
  mutate(total= sum(count), # create new variables like sum which count part and gender
         prop= count/total # variable for gender proportion at this range
  ) %>% 
```

 score_range  part  sex  count  total       prop
1      700???800  read male  40151 776092 0.05173485
2      600-690  read male 121950 776092 0.15713343
3      500-590  read male 227141 776092 0.29267278
4      400-490  read male 242554 776092 0.31253253
5      300-390  read male 113568 776092 0.14633317
6      200-290  read male  30728 776092 0.03959324
7      700???800  read  fem  38898 883955 0.04400450
8      600-690  read  fem 126084 883955 0.14263622
9      500-590  read  fem 259553 883955 0.29362694
10     400-490  read  fem 296793 883955 0.33575578
11     300-390  read  fem 133473 883955 0.15099524
12     200-290  read  fem  29154 883955 0.03298132
13     700???800  math male  74461 776092 0.09594352
14     600-690  math male 162564 776092 0.20946486
15     500-590  math male 233141 776092 0.30040382
16     400-490  math male 204670 776092 0.26371873
17     300-390  math male  82468 776092 0.10626060
18     200-290  math male  18788 776092 0.02420847
19     700???800  math  fem  46040 883955 0.05208410
20     600-690  math  fem 133954 883955 0.15153939
21     500-590  math  fem 257678 883955 0.29150579
22     400-490  math  fem 288696 883955 0.32659581
23     300-390  math  fem 131025 883955 0.14822587
24     200-290  math  fem  26562 883955 0.03004904
25     700???800 write male  31574 776092 0.04068332
26     600-690 write male 100963 776092 0.13009154
27     500-590 write male 202326 776092 0.26069847
28     400-490 write male 262623 776092 0.33839158
29     300-390 write male 146106 776092 0.18825861
30     200-290 write male  32500 776092 0.04187648
31     700???800 write  fem  39101 883955 0.04423415
32     600-690 write  fem 125368 883955 0.14182622
33     500-590 write  fem 247239 883955 0.27969636
34     400-490 write  fem 302933 883955 0.34270183
35     300-390 write  fem 144381 883955 0.16333524
36     200-290 write  fem  24933 883955 0.02820619