import numpy as np 
import pandas as pd 
import seaborn as sns 
import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')

train = pd.read_csv('titanic/train.csv')
test = pd.read_csv('titanic/test.csv')

train['train_test'] = 1
test['train_test'] = 0
test['Survived'] = np.NaN
df = pd.concat([train,test])

df.describe()

df.dtypes

numeric = train.select_dtypes(exclude='object')
numeric.columns

#correlation with prediction/target var

correlation = abs(numeric.corr()*100)
correlation[['Survived']].sort_values(['Survived'], ascending=False)

plt.figure(figsize=(8,6))
sns.heatmap(correlation, mask=correlation<10, cmap='Blues', vmin=10, linewidths=.5, linecolor='k')

fig = plt.figure(figsize=(18,16))

for index, col in enumerate(numeric.columns):
     plt.subplot(4,2,index+1)
     sns.distplot(numeric.loc[:,col].dropna(), kde=False)
     
fig.tight_layout(pad=1.0)

# Outliers 

fig =  plt.figure(figsize=(14,20))
for index , col in enumerate(numeric.columns):
    plt.subplot(4,4, index+1)
    sns.boxplot(y=col, data=numeric.dropna())
fig.tight_layout(pad=1.0)

cat = train.select_dtypes(include=['object']).copy()
cat['Pclass'] = train['Pclass']
cat = cat.drop(['Ticket', 'Cabin', 'Name'], axis=1) #droped knowing it has a lot of possible values
cat.columns

fig = plt.figure(figsize=(10,20))
for index in range(len(cat.columns)):
    plt.subplot(4,2,index+1)
    sns.countplot(x=cat.iloc[:,index],data=cat.dropna())
    plt.xticks(rotation=90)
fig.tight_layout(pad=1.0)

df.isnull().sum().nlargest(40)  

df = df.drop(['Cabin'],axis=1)

#Filling missing Age values based on the median of the name tiltle (Mss. Mr. Mrs. etc)

import re
df['name_titles'] = df['Name'].apply(lambda x: re.findall(',\s(\w*).',x)[0])
df['Age'] = df.groupby('name_titles')['Age'].transform(lambda x: x.fillna(x.median()))

df['Embarked'] = df['Embarked'].fillna('S')
df['Embarked'].unique()

df['Fare'] = df.groupby('Pclass')['Fare'].transform(lambda x: x.fillna(x.mean()))

df.isnull().sum().nlargest(40) 

df['fam'] = df['SibSp']+ df['Parch']

df = df.drop(['PassengerId', 'Name', 'Ticket', 'name_titles', 'Embarked', 'SibSp', 'Parch'], axis=1)
df

#converting Sex and Embarked into Binary 
df['Sex'] = np.where(df['Sex']== 'female', 1,0)
df = pd.get_dummies(df, columns=['Sex'])
df

dfbin = df.copy()
dfbin['Fare'] = pd.qcut(dfbin['Fare'], 5)
dfbin = pd.get_dummies(dfbin, columns=['Fare', 'Pclass', 'fam'])
dfbin

# Separating Train and Test 

train_df = df[df['train_test']==1]
test_df = df[df['train_test']==0]
train_df = train_df.drop('train_test', axis=1)
test_df = test_df.drop('train_test', axis=1)

train_dfbin = dfbin[dfbin['train_test']==1]
test_dfbin = dfbin[dfbin['train_test']==0]
train_dfbin = train_dfbin.drop('train_test', axis=1)
test_dfbin = test_dfbin.drop('train_test', axis=1)

train_dfbin_y = train_dfbin.Survived
train_dfbin_X = train_dfbin.drop(['Survived'], axis=1)

train_df_y = train_df.Survived
train_df_X = train_df.drop(['Survived'], axis=1)

train_df_X

from sklearn.ensemble import GradientBoostingClassifier
from sklearn import model_selection, metrics

model = GradientBoostingClassifier().fit(train_df_X, train_df_y)
accuracy =  round(model.score(train_df_X,train_df_y)*100,2)
cross_val = model_selection.cross_val_predict(GradientBoostingClassifier(n_estimators=150, learning_rate=0.1), train_df_X,train_df_y, cv=10, n_jobs=-1)
acc_cv = round(metrics.accuracy_score(train_df_y,cross_val)*100,2)
feature_ranks = pd.Series((model.feature_importances_)*100, index=train_df_X.columns).sort_values(ascending=False)

print('Acc ',accuracy )
print()
print('Cross val ', acc_cv)
print('Feature rank ')
print(feature_ranks)

import numpy as np 
import pandas as pd 
import seaborn as sns 
import matplotlib.pyplot as plt
plt.style.use('fivethirtyeight')
import math, time, random, datetime

train = pd.read_csv('titanic/train.csv')
test = pd.read_csv('titanic/test.csv')

train.isnull().sum()

df_bin =pd.DataFrame() #to store continuous variables in bins  ex(age bins: 0-10, 11-20, 21-31, etc..)
df_con = pd.DataFrame() # to store continuous variables 

train.dtypes

train.describe()

fig = plt.figure(figsize=(20,1))
sns.countplot(y='Survived', data=train)
print(train.Survived.value_counts())

df_bin['Survived']= train['Survived']
df_con['Survived']= train['Survived']

#Name ticket and Cabin have too many value types, hard to put in bins not so useful.
print(train.Name.value_counts().count())
print(train.Ticket.value_counts().count())
print(train.Cabin.value_counts().count())

sns.distplot(train.Pclass) # its Ordinal Data 

df_bin['Pclass']= train['Pclass']
df_con['Pclass']= train['Pclass']

plt.figure(figsize=(20,1))
sns.countplot(y='Sex', data=train)

# Function to plot a count and a distribution(vs Survived) of each variable 

def plot_count_dist (data, bin_df, label_column, target_colum, figsize=(20,5), use_bin_fd=False):
    if use_bin_fd:
        fig= plt.figure(figsize=figsize)
        plt.subplot(1,2,1)
        sns.countplot(y=target_colum, data=bin_df)
        plt.subplot(1,2,2)
        sns.distplot(data.loc[data[label_column]==1][target_colum], kde_kws={'label':'Survived'})
        sns.distplot(data.loc[data[label_column]==0][target_colum], kde_kws={'label':'Did Not Survived'})
    else:
        fig= plt.figure(figsize=figsize)
        plt.subplot(1,2,1)
        sns.countplot(y=target_colum, data=data)
        plt.subplot(1,2,2)
        sns.distplot(data.loc[data[label_column]==1][target_colum], kde_kws={'label':'Survived'})
        sns.distplot(data.loc[data[label_column]==0][target_colum], kde_kws={'label':'Did Not Survived'})

# Adding Sex and changing to binary form using np.where
df_bin['Sex'] = train['Sex']
df_bin['Sex'] = np.where(df_bin['Sex']== 'female', 1,0)
df_con['Sex'] = train['Sex']

# Adding SibSp = num of Siblings/Spouse
df_bin['SibSp']= train['SibSp']
df_con['SibSp']= train['SibSp']

plot_count_dist(train, bin_df=df_bin, label_column='Survived', target_colum='SibSp', figsize=(20,10))

# Adding Parch = num of Parents/Children
df_bin['Parch']= train['Parch']
df_con['Parch']= train['Parch']
plot_count_dist(train, bin_df=df_bin, label_column='Survived', target_colum='Parch', figsize=(20,10))

# Ading Fare, first variable to be cut into bins 
df_con['Fare'] = train['Fare']
df_bin['Fare'] = pd.qcut(train['Fare'], 5)
plot_count_dist(train,df_bin,'Survived', 'Fare', use_bin_fd=True, figsize=(20,15))

# Embarqued
df_bin['Embarked'] = train['Embarked']
df_con['Embarked'] = train['Embarked']

df_bin = df_bin.dropna(subset=['Embarked'])
df_con = df_con.dropna(subset=['Embarked'])

import re
df_con['name_titles'] = train['Name'].apply(lambda x: re.findall(',\s(\w*).',x)[0])
df_con['Age'] = train['Age']
df_con['Age'] = df_con.groupby('name_titles')['Age'].transform(lambda x: x.fillna(x.median()))

df_bin.head(20)

from sklearn.preprocessing import OneHotEncoder, LabelEncoder, label_binarize

#df_bin.drop('Survived', axis=1, inplace=True)
df_bin_enc = pd.get_dummies(df_bin, columns=['Pclass', 'Sex', 'SibSp', 'Parch', 'Fare', 'Embarked'])
df_bin_enc

df_con_enc = pd.get_dummies(df_con, columns=['Sex', 'Embarked'])
df_con_enc

X_train_bin = df_bin_enc.drop('Survived', axis=1)
y_train_bin = df_bin_enc.Survived
X_train_con = df_con_enc.drop(['Survived','name_titles'], axis=1)
y_train_con = df_con_enc.Survived

from sklearn import model_selection, metrics
from sklearn.svm import LinearSVC
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.linear_model import LinearRegression, LogisticRegression, SGDClassifier
from sklearn.tree import DecisionTreeClassifier

# Function to run different algorithm with cross val 

def fit_algo(algo, X_train, y_train, cv): #cv Cross validate, number of times try 10 
    model = algo.fit(X_train, y_train)
    accuracy =  round(model.score(X_train,y_train)*100,2)
    cross_val = model_selection.cross_val_predict(algo, X_train,y_train, cv=cv, n_jobs=-1)
    acc_cv = round(metrics.accuracy_score(y_train,cross_val)*100,2)
    #feature_ranks = pd.Series((model.feature_importances_)*100, index=X_train.columns).sort_values(ascending=False)
    return accuracy, acc_cv, #feature_ranks

acc_grad_boost, cross_grad_boost, featureranks = fit_algo(GradientBoostingClassifier(n_estimators=300, learning_rate=0.1), X_train_con, y_train_con,10)

print('General Acc is ' , acc_grad_boost)
print()
print('With cross val is ',cross_grad_boost)
print()
print(featureranks)

acc_dTree, cross_dTree = fit_algo( DecisionTreeClassifier(),  X_train_con, y_train_con, 10)

print('General Acc is ' , acc_dTree)
print()
print('With cross val is ',cross_dTree)

accuracy_LSVC, Cross_LSVC =fit_algo(LinearSVC(), X_train_bin, y_train_bin, 10)

print('General Acc is ' , accuracy_LSVC)
print()
print('With cross val is ',Cross_LSVC)

acc_log, cross_log = fit_algo( LogisticRegression(),  X_train_bin, y_train_bin, 10)

print('General Acc is ' , acc_log)
print()
print('With cross val is ',cross_log)

acc_KN, cross_KN = fit_algo(KNeighborsClassifier(),  X_train_bin, y_train_bin, 10)

print('General Acc is ' , acc_KN)
print()
print('With cross val is ',cross_KN)

acc_gaus, cross_gaus = fit_algo(GaussianNB(),  X_train_con, y_train_con, 10)

print('General Acc is ' , acc_gaus)
print()
print('With cross val is ',cross_gaus)

acc_sgdc, cross_sgdc = fit_algo(SGDClassifier(),  X_train_bin, y_train_bin, 10)

print('General Acc is ' , acc_sgdc)
print()
print('With cross val is ',cross_sgdc)

from sklearn.svm import SVC, NuSVC

acc_grad_boost, cross_grad_boost = fit_algo(NuSVC(), X_train_bin, y_train_bin,10)

print('General Acc is ' , acc_grad_boost)
print()
print('With cross val is ',cross_grad_boost)