titanic.py

import pandas as pd ; import numpy as np
import matplotlib.pyplot as plt
import multiprocessing
from sklearn.preprocessing import LabelEncoder, OneHotEncoder
from sklearn.model_selection import train_test_split
import pathlib
from sklearn.preprocessing import MinMaxScaler
from sklearn.ensemble import RandomForestClassifier
import time
import os

os.chdir('/home/coder/work/ensae-reproductibilite-application')
TrainingData = pd.read_csv('data.csv')

TrainingData.head()


TrainingData['Ticket'].str.split("/").str.len()

TrainingData['Name'].str.split(",").str.len()

n_trees = 20
max_depth =None
max_features='sqrt'

TrainingData.isnull().sum()


## Un peu d'exploration et de feature engineering

### Statut socioéconomique

fig, axes=plt.subplots(1,2, figsize=(12, 6)) #layout matplotlib 1 ligne 2 colonnes taile 16*8
fig1_pclass=sns.countplot(data=TrainingData, x ="Pclass",    ax=axes[0]).set_title("fréquence des Pclass")
fig2_pclass=sns.barplot(data=TrainingData, x= "Pclass",y= "Survived", ax=axes[1]).set_title("survie des Pclass")


### Age

sns.histplot(data= TrainingData, x='Age',bins=15, kde=False    )    .set_title("Distribution de l'âge")
plt.show()

## Encoder les données imputées ou transformées.
from sklearn.preprocessing import MinMaxScaler, OneHotEncoder
from sklearn.ensemble import RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer

numeric_features=["Age", "Fare"]
categorical_features=["Embarked", "Sex"]

numeric_transformer = Pipeline(steps=[("imputer", SimpleImputer(strategy="median")),
("scaler", MinMaxScaler()),])

categorical_transformer = Pipeline(steps=[("imputer", SimpleImputer(strategy="most_frequent")),("onehot", OneHotEncoder()),])


preprocessor = ColumnTransformer(
transformers=[
("Preprocessing numerical", numeric_transformer, numeric_features),
(
"Preprocessing categorical",
categorical_transformer,
categorical_features,
),
        ]
    )

pipe = Pipeline(
        [
            ("preprocessor", preprocessor),
            ("classifier", RandomForestClassifier(n_estimators=20)),
        ]
    )



# splitting samples
y = TrainingData["Survived"]
X = TrainingData.drop("Survived", axis = 'columns')

# On _split_ notre _dataset_ d'apprentisage pour faire de la validation croisée une partie pour apprendre une partie pour regarder le score.
# Prenons arbitrairement 10% du dataset en test et 90% pour l'apprentissage.
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1)
pd.concat([X_train, y_train]).to_csv("train.csv")
pd.concat([X_test, y_test]).to_csv("test.csv")

jetonapi = "$trotskitueleski1917"


# Random Forest

from sklearn.preprocessing import LabelEncoder, OneHotEncoder
from sklearn.model_selection import train_test_split
import pathlib
from sklearn.preprocessing import MinMaxScaler
from sklearn.ensemble import RandomForestClassifier


#Ici demandons d'avoir 20 arbres
pipe.fit(X_train, y_train)


#calculons le score sur le dataset d'apprentissage et sur le dataset de test (10% du dataset d'apprentissage mis de côté)
# le score étant le nombre de bonne prédiction
rdmf_score = pipe.score(X_test, y_test)
rdmf_score_tr = pipe.score(X_train, y_train)
print(f"{rdmf_score:.1%} de bonnes réponses sur les données de test pour validation")
from sklearn.metrics import confusion_matrix
print(20*"-")
print("matrice de confusion")
print(confusion_matrix(y_test, pipe.predict(X_test)))