README.Rmd

---
title: "Titanic"
date: "November 12, 2018"
output: md_document
---

```{r setup}
knitr::opts_chunk$set(echo = TRUE)
```

Our Project is about Determining whether a person is dead or was lucky enough to survive from the titanic trajedgy

```{r libraries, include=FALSE}
library(dplyr)
library(ggplot2)
library(stringr)
```

```{r reading data}
train <- read.csv('train.csv')
test  <- read.csv('test.csv')
```
## Looking at our data

```{r train}
str(train)
```

```{r test}
str(test)
```

First thing to notice, Some Features has variables with type **numeric**
or **int** that can be turned to **factor** as it's a **Discrete** Feature. (Survived, Pclass, Sibsb, Parch)

But before that, We combine both train and test in a single DataFrame

Test dataset contains all train dataset Features except the *Survived* feature since it's the Dependent Variable in Our Challenge (that we should predict its value)

So we add the Survived Column, thus we can combine 2 datasets by row
```{r test.survived}
test.survived <- data.frame(Survived = rep("None" , nrow(test)), test[,])
```
```{r dim_train}
dim(train)
```
```{r dim_test}
dim(test.survived)
```
```{r combine}
data.combined <- rbind(train , test.survived)
```

Making sure that these features can be treated as cateorical data
```{r}
  table(data.combined$SibSp)
```
```{r}
  table(data.combined$Parch)
```

So we can convert data-type of those features to be Categorical
```{r}
data.combined$Survived <- as.factor(data.combined$Survived)
data.combined$Pclass <- as.factor(data.combined$Pclass)
data.combined$SibSp <- as.factor(data.combined$SibSp)
data.combined$Parch <- as.factor(data.combined$Parch)
```


### Plotting is the Best way to analyse your data

Well the data is Skewed to people didn't make it. 
```{r}
ggplot(data.combined[which(data.combined$Survived != "None"),],aes(x= Survived))+
       geom_bar(fill = "#1ec97f")+
       ggtitle("Survival Ratio")

```

The Same previous plot, according to classes
```{r}
ggplot(data.combined[data.combined$Survived != "None",], aes(x = Pclass, fill = Survived))+
  geom_bar( position = "dodge")+
  xlab("Pclass")+
  ylab("Total Count")+
  labs(fill = "Survived")+
  ggtitle("Survival across Classes")

```

First class has the Highest probability for Survivnig
While the third class is absolute opposite
Second class fifty-fifty

which means *Class* is an important Feature, that we should take into consideration while building our model

Moving to *Names* Feature

```{r}
head(data.combined$Name)
```

`"Mr", "Mrs", "Miss", "Master"`
these names may help us in our analysis, or we can create a new Feature depends on these `titles`

let's see if that hypothese is true

We extract "Misses" only from all the passengers

► `str_detect()`
takes : 1- vector of string (Name column) -- 
        2- pattern  (the "Miss" string)
Outputs: Logical vector (Miss or Not)

```{r}
misses <- data.combined[which(str_detect(data.combined$Name , "Miss.")) , ]
nrow(misses)
```

Now plotting `misses` to see if it's an effective feature or not

```{r}
ggplot(misses[misses$Survived != "None" ,] ,
       aes(x = Pclass , fill = Survived))+
       geom_bar(position = "dodge")+
       ggtitle("Survival Of Misses across Classes")
```

Well it seems that in
   Class 1, class 2 : we can say All misses Survived
   class 3 : it looks like a coin luck.


So doing same through another `title`


```{r}
Mrs <- data.combined[which(str_detect(data.combined$Name , "Mrs.")) ,]
nrow(Mrs)
```


```{r}
ggplot(Mrs[Mrs$Survived != "None" , ] ,aes(x = Pclass , fill = Survived))+
  geom_bar(position = "dodge")+
  ggtitle("Survival of Mrs across Classes")
```

Looks the same as the `Miss` title

So it may be a good feature to create

first thing we make a method that returns whatever title it is
then add it into a vector 
that we then insert to our `data.combined` dataFrame


►`grep()` method
  takes : pattern first
          string value (Name)
  outputs: 1 (match)
           integer(0) (didn't match)
          
we convert the `integer(0) -> 0` using `length()`;

```{r}
assign_titles <- function(name){
  
  name <- as.character(name)
  
  if(length(grep("Miss." , name) > 0))
    return("Miss.")
  if(length(grep("Mrs."  , name) > 0))
    return("Mrs.")
  if(length(grep("Mr." , name) > 0))
    return("Mr.")
  if(length(grep("Master" , name) > 0))
    return("Master.")
  else
    return("Other")
}

```

Now initalizing the `title` vector
```{r}
titles <- NULL
```

and inserting new titles using our methods
```{r}
for(i in 1:nrow(data.combined)){
  titles <- c(titles , assign_titles(data.combined[i , "Name"]))
}
```
finally, adding it to our dataframe
```{r}
data.combined$titles <- as.factor(titles)
```

Here is our new Feature, with corresponding names, token randomly
```{r}
sample_n(data.combined[, c("Name", "titles")], 5)
```

See what this feature say

```{r}
ggplot(data.combined[which(data.combined$Survived != "None"),],aes(x = titles, fill=Survived))+
  geom_bar(position = "dodge")+
  ggtitle("Survival across titles")
```

which means that `Men` are the most likely to be dead
and `Miss, Mrs` more likely to live
while masters (young men) have equal propability (until now)

Now look at our data with respect to 
`Class, title and Survived` Features

```{r}
ggplot(data.combined[data.combined$Survived != "None", ] , aes(x = titles , fill = Survived))+
  geom_bar()+
  facet_wrap(~Pclass)+
  ggtitle("Survival of Titles w.r.t Class")
```


we can see that 
  *Master*: survived with 100% in class 1 and 2 <br>
  *Misses*: almost same<br>
  *Mrs*   : almost same <br>
  *Mr*    : in class 1 - 50/50<br>
            in class 2 - more probable to be dead<br><br>
  in class 3 the Savage class.<br>
  *Mr* are the most likely to die<br>
  while others have almost 50/50 probability<br><br>
  
  <hr>

Note that the `echo = FALSE` parameter was added to the code chunk to prevent printing of the R code that generated the plot.