app.py

import streamlit as st, base64
import pandas as pd, seaborn as sns
import os, matplotlib.pyplot as plt
import pickle, numpy as np, xgboost as xgb
from keras.models import load_model
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import classification_report, accuracy_score, confusion_matrix


# image de fond
def add_bg_from_local(image_file):
    with open(image_file, "rb") as image_file:
        encoded_string = base64.b64encode(image_file.read())
    st.markdown(
    f"""
    <style>
    .stApp {{
        background-image: url(data:image/{"png"};base64,{encoded_string.decode()});
        background-size: cover
    }}
    </style>
    """,
    unsafe_allow_html=True
    )
add_bg_from_local('route.png')


fig = plt.figure(figsize=(10, 10))
_, middle, _ = st.columns((2, 3, 2))
with middle:
    st.title(":orange[_Scoring App_]") 
# path du dossier data
#path  = ".\data"

# fonction pour loader le dataset
@st.cache_data
def load_data(file_path):
    return pd.read_csv(file_path)

# convertir dataframe en csv
def  convert_df_to_csv(frame):
    return frame.to_csv(index=False).encode("utf-8")

# fonction principale
st.sidebar.image("picture1.png")
def main():
    st.markdown("<h2 style = 'text-align:center; \
        color:green;'> Classification pour l'octroi de credit </h2>", unsafe_allow_html = True)
    
    # charger le fichier
    uploaded_file = st.sidebar.file_uploader("Importer fichier csv", type=["csv"])
    
    # creation du menu
    menu = ["Home", "Data Exploration", "Data Visualisation", "Make prediction"]
    choice = st.sidebar.selectbox("Select menu", menu)
    
    # charger le jeu de donnees
    data = load_data("loan.csv")
        
    # supprime la colonne Loan_ID
    data.drop("Loan_ID", axis=1, inplace=True)
    if choice == "Home":
        st.write("Nous avons develeopper pour ce projet un model de classification\
            qui permet, sur la base de certaines variables, de determiner si oui ou non\
                il est envisageable d'octroyer un pret bancaire a une tierce personne.")
        
        st.subheader(":orange[__Presentation du jeu de donnee__] :memo:")
        
        st.markdown("Le jeu de donnees comporte 614 lignes et 13 colonnes. **Loan_Status**\
            est la variables a predire (categorielle a deuc classe: **Y** pour le pret a ete \
                octroyer et **N**) pour le contraire. afin d'avoir les reultats les plus\
                    optimaux possibles, nous allons dans un premier temps faire une \
                        ***Analyse exploratoire** de nos donnees. Par suite nous passerons\
                            a la phase de preparation des donnees pour afin finir avec \
                                la phase de creation et optimisation des models.\
                                `Si vous televerser un fichier au format csv, vous avez la\
                                    possibilite de comparer les prediction pour chaque\
                                    model et de telechager le fichier csv correspondant.`")
        #st.image("./images/processor.jpg")
        
    if choice == "Data Exploration":
        st.subheader(":orange[_Data Exploration_] :bar_chart:")
        # afficher les donnees
        st.write(data.head())
        
        # valeurs manquante
        if st.sidebar.checkbox("Valeur Manquante"):
            st.subheader(":orange[Valeur Manquante]")
            na_count = data.isnull().sum().to_frame(name='count')
            na_per = (data.isnull().sum().to_frame(name='percentage %')/data.shape[0]*100).round(2)
            st.write(pd.concat([na_count, na_per], axis=1).sort_values(by='count', ascending=False).T)
        
        # valeur unique par colonne
        if st.sidebar.checkbox("Valeur Unique par colonnes"):
            st.subheader(":orange[Valeur Unique par colonnes]")
            only = data.nunique().sort_values(ascending=False).to_frame(name='count')
            perc = (data.nunique().sort_values(ascending=False).to_frame(name='percentage %')/data.shape[0]*100).round(2)
            dtype = data.dtypes.to_frame(name='dtypes')
            st.write(pd.concat([only, perc, dtype], axis=1).T) 
           
        # statistique sommaire 
        if st.sidebar.checkbox("Statistiques somaire"):
            st.subheader(":orange[Statistiques sommaire]")
            st.write(data.describe())
        
        # matrice de correlation
        if st.sidebar.checkbox("Matrice de correlation"):
            fig = plt.figure(figsize=(7,5))
            st.subheader(":orange[Matrice de correlation]")
            st.write(sns.heatmap(data.corr(), annot=True, vmin=-1, vmax=1, cmap='ocean'))
            st.pyplot(fig)
            plt.show()
    
    if choice == "Data Visualisation":
        st.subheader(":orange[_Data Visualisation_] :chart:")
        if st.sidebar.checkbox("Analyse Univariee"):
            # selection des variables qualitatives
            categorical_columns = data.select_dtypes(include='object').columns.tolist()
            st.write("Liste des variables qaulitatives")
            st.write(categorical_columns)
            fig = plt.figure(figsize=(14, 8))
            sns.set_theme(context='notebook', style='darkgrid', palette='deep', font='sans-serif', font_scale=1, color_codes=True, rc=None)
            for idx, col in enumerate(categorical_columns[:-1]):
                plt.subplot(2, 3, idx+1)
                sns.countplot(data=data, x=col, hue="Loan_Status")
            sns.countplot(data=data, x='Loan_Status')
            st.pyplot(fig)
            plt.show()
            
            # selection des variables quantitatives
            numerical_columns = data.select_dtypes(include='number').columns.tolist()
            st.write("Liste des variables quantitatives")
            st.write(numerical_columns)
            fig = plt.figure(figsize=(15,7))
            for idx, col in enumerate(numerical_columns):
                plt.subplot(2,3, idx+1)
                plt.hist(data[col], density=True)
                sns.kdeplot(data=data, x=col)
                plt.title(col)
            #plt.subplots_adjust(hspace=0.5)
            plt.tight_layout(h_pad=2, w_pad=3., rect=(1,1,2,2))
            st.pyplot(fig)
            plt.show()
        
        if st.sidebar.checkbox("Analyse bivariee"):
            st.subheader(":orange[Analyse bivariee]")
            numerical_columns = data.select_dtypes(include='number').columns.tolist()
            fig = plt.figure(figsize = (14, 8))
            for idx, num_col in enumerate(numerical_columns[:-2]):
                plt.subplot(2, 2, idx+1)
                sns.boxplot(y=num_col, data=data, x='Loan_Status')
            plt.tight_layout(h_pad=2, w_pad=3., rect=(1,1,2,2))
            st.pyplot(fig)
            plt.show()
    
    if choice == "Make prediction":
        st.subheader(":orange[Make prediction] :fleur_de_lis:")
        if uploaded_file is not None:
            data = pd.read_csv(uploaded_file)
            
            # data preprocessing
            from sklearn.impute import SimpleImputer
            try:
                data.drop(["Loan_ID"], axis=1, inplace=True)
            except:
                pass
            # encodage
            data_encoded = pd.get_dummies(data, drop_first=True)
            st.subheader(":orange[Donnees encodees]")
            st.write(data_encoded)
            
            # separation du jeu de donnee
            if data_encoded.shape[1] == 15:
                X, y = data_encoded.drop(["Loan_Status_Y"], axis=1), data_encoded["Loan_Status_Y"]
            else:
                X = data_encoded
            
            # traintement des valeurs manquantes
            sp = SimpleImputer(strategy="most_frequent")
            X = sp.fit_transform(X)
            
            # mis a l'echelle des variables
            std = StandardScaler()
            X = std.fit_transform(X)
            
            # Prediction
                # Random Forest predictor
            if st.sidebar.checkbox("Random Forest"):
                st.subheader(":orange[Random Forest] :sunglasses:")
                rf = pickle.load(open("scoring_rf.pkl", "rb"))
                pred = rf.predict(X)
                pred_proba = rf.predict_proba(X)
                st.subheader(':green[Prediction]')
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data, prediction], axis=1)
                st.write(df)
                 # download frame 
                csv = convert_df_to_csv(df)
                st.download_button("Press to Download",
                                   csv,
                                   "random_forest.csv",
                                   "text/csv",
                                   key='rf_download_csv')

                # Accuracy score
                if data_encoded.shape[1] == 15:
                    st.text("Model report : \n " + classification_report(y, pred))
                    rf_score = accuracy_score(pred,y)
                    st.write(":green[score d'exactitude]")
                    st.write(f"{round(rf_score*100,2)}% d'exactitude")
                st.subheader(':green[Prediction Probability]')
                st.write(pred_proba)
                
                # Linear Discriminant Analysis
            if st.sidebar.checkbox("Discriminant Analysis"):
                st.subheader(":orange[Discriminant Analysis] :sunglasses:")
                lda = pickle.load(open("scoring_lda.pkl", "rb"))
                pred = lda.predict(X)
                pred_proba = lda.predict_proba(X)
                st.subheader(':green[Prediction]')
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data, prediction], axis=1)
                st.write(df)
                 # download 
                csv = convert_df_to_csv(df)
                st.download_button("Press to Download",
                                   csv,
                                   "discriminant.csv",
                                   "text/csv",
                                   key='lda_download_csv')
                #st.text("Model report : \n " + classification_report(y, pred))

                if data_encoded.shape[1] == 15:
                    st.text("Model report : \n " + classification_report(y, pred))
                    # Accuracy score
                    lda_score = accuracy_score(pred,y)
                    st.subheader(":green[score d'exactitude]")
                    st.write(f"{round(lda_score*100,2)}% d'exactitude")
                st.subheader(':green[Prediction Probability]')
                st.write(pred_proba)

                if data_encoded.shape[1] == 15:
                    # matrice de confusion
                    fig = plt.figure(figsize=(2,1))
                    cm = confusion_matrix(y, pred)
                    st.subheader(":green[Matrice de confusion]")
                    sns.heatmap(cm, annot=True, cmap='Dark2')
                    st.pyplot(fig)
                    plt.plot()
                
                # XGBoost
            if st.sidebar.checkbox("XGBoost"):
                st.subheader(":orange[XGBoost] :sunglasses:")
                xg = xgb.XGBClassifier()
                xg.load_model("xg.json")
                pred = xg.predict(X)
                pred_proba = xg.predict_proba(X)
                st.subheader(':green[Prediction]')
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data, prediction], axis=1)
                st.write(df)
                 # download 
                csv = convert_df_to_csv(df)
                st.download_button("Press to Download",
                                   csv,
                                   "xgboost.csv",
                                   "text/csv",
                                   key='xg_download_csv')
                #st.text("Model report : \n " + classification_report(y, pred))
                if data_encoded.shape[1] == 15:
                    st.text("Model report : \n " + classification_report(y, pred))
                    # Accuracy score
                    xg_score = accuracy_score(pred,y)
                    st.subheader(":green[score d'exactitude]")
                    st.write(f"{round(xg_score*100,2)}% d'exactitude")
                st.subheader(':green[Prediction Probability]')
                st.write(pred_proba)
                
                # ANN
            if st.sidebar.checkbox("Neural Network"):
                st.subheader(":orange[Neural Network] :sunglasses:")
                ann = load_model("ann.h5")
                pred_proba = ann.predict(X)
                pred = np.where(pred_proba < 0.5, 0, 1)
                st.subheader(':green[Prediction]')
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data, prediction], axis=1)
                st.write(df)
                 # download 
                csv = convert_df_to_csv(df)
                st.download_button("Press to Download",
                                   csv,
                                   "neural_network.csv",
                                   "text/csv",
                                   key='ann_download_csv')
                #st.text("Model report : \n " + classification_report(y, pred))
                if data_encoded.shape[1] == 15:
                    st.text("Model report : \n " + classification_report(y, pred))
                    # Accuracy score
                    ann_score = accuracy_score(pred,y)
                    st.subheader(":green[score d'exactitude]")
                    st.write(f"{round(ann_score*100,2)}% d'exactitude")
                st.subheader(':green[Prediction Probability]')
                un = pd.DataFrame(pred_proba, columns=['1'])
                zero = pd.DataFrame(np.subtract(1, pred_proba), columns=['0'])
                st.write(pd.concat([zero, un], axis=1).round(2))
            
        else:
            def user_input_features():
                gender = st.sidebar.selectbox('Gender',('Male','Female'))
                married = st.sidebar.selectbox('Married',('Yes','No'))
                depedents = st.sidebar.selectbox('Dependent',(0, 1, 2, "3+"))
                education = st.sidebar.selectbox('Education',('Graduate','Not Graduate'))
                self_employed = st.sidebar.selectbox('Self_employed',('Yes','No'))
                applicanincome = st.sidebar.slider('ApplicanIncome', 150, 81000)
                coapplicanincome = st.sidebar.slider('CoapplicanIncome', 0, 42000)
                loan_amount = st.sidebar.slider('LoanAmount', 0, 800)
                loan_amount_term = st.sidebar.slider('Loan_Amount_Term', 10, 500)
                credit_history = st.sidebar.selectbox('Credi_History', (0, 1))
                property_area = st.sidebar.selectbox('Property_Area', ("Urban", "Rural", "Semiurban"))
                
                if gender == "Male": 
                    gender = 1
                else: 
                    gender = 0
                
                if married == 'Yes':
                    married = 1
                else: 
                    married = 0
                
                depedents_1, depedents_2, depedents_3 = 0,0,0
                if depedents == 1: 
                    depedents_1=1
                elif depedents == 2: 
                    depedents_2=1
                elif depedents > 2 : 
                    depedents_3=1
                
                if education == "Not Graduate": 
                    education=1
                else: 
                    education=0
                
                if self_employed == "Yes":
                    self_employed = 1
                else:
                    self_employed = 0
                    
                property_urban, property_semiurban = 0, 0
                if property_area == "Semiurban":
                    property_semiurban = 1
                elif property_area == "Urban":
                    property_urban == 1
                
                data = { 'ApplicationIncome': (applicanincome - 5403)/6109,
                        'CoapplicationIncome': (coapplicanincome - 1621) / 2926,
                        'LoanAmount': (loan_amount -146)/85,
                        'Loan_Amount_Term': (loan_amount_term - 342)/65,
                        'Credi_History': (credit_history -0.84)/0.35,
                        'Gender_Male': gender,
                        'Married_Yes': married,
                        'Depedents_1': depedents_1,
                        'Depedents_2': depedents_2,
                        'Depedents_3+': depedents_3,
                        'Education_Not_Graduate': education,
                        'Self_Employed_Yes': self_employed,
                        'Property_Area_Semiurban': property_semiurban,
                        'Property_Area_Urban': property_urban
                        }
                features = pd.DataFrame(data, index=[0])
                return features
            data_input = user_input_features()
            
            # Random Forest
            if st.sidebar.checkbox("Random Forest"):
                st.subheader(":orange[Random Forest]")
                rf = pickle.load(open("scoring_rf.pkl", "rb"))
                pred = rf.predict(data_input)
                if pred == 1:
                    st.write(":orange[__Le pret peut etre octroyer__] :white_check_mark:")
                else: 
                    st.write(":red[__Desole,...__] :disappointed:")
                pred_proba = rf.predict_proba(data_input)
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data_input, prediction], axis=1)
                st.write(df)
                st.subheader(":green[probability] :question:")
                st.write(pred_proba)
                
             # Discriminant Analysis
            if st.sidebar.checkbox("Discriminant Analysis"):
                st.subheader(":orange[Discriminant Analysis]")
                lda = pickle.load(open("scoring_lda.pkl", "rb"))
                pred = lda.predict(data_input)
                if pred == 1:
                    st.write(":orange[__Le pret peut etre octroyer__] :white_check_mark:")
                else: 
                    st.write(":red[__Desole,...__] :disappointed:")
                pred_proba = lda.predict_proba(data_input)
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data_input, prediction], axis=1)
                st.write(df)
                st.subheader(":green[probability] :question:")
                st.write(pred_proba)
                
            # XGboost
            if st.sidebar.checkbox("XGBoost"):
                st.subheader(":orange[XGBoost]")
                xg = xgb.XGBClassifier()
                xg.load_model("xg.json")
                pred = xg.predict(data_input)
                if pred == 1:
                    st.write(":orange[__Le pret peut etre octroyer__] :white_check_mark:")
                else: 
                    st.write(":red[__Desole,...__] :disappointed:")
                pred_proba = xg.predict_proba(data_input)
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data_input, prediction], axis=1)
                st.write(df)
                st.subheader(":green[probability] :question:")
                st.write(pred_proba)
                
            # ANN
            if st.sidebar.checkbox("Neural Network"):
                st.subheader(":orange[Neural Network]")
                ann = load_model('ann.h5')
                pred_proba = ann.predict(data_input)
                pred = np.where(pred_proba < 0.5, 0, 1)
                if pred == 1:
                    st.write(":orange[__Le pret peut etre octroyer__] :white_check_mark:")
                else: 
                    st.write(":red[__Desole,...__] :disappointed:")
                loan_status = np.array(['N','Y'])
                prediction = pd.DataFrame(loan_status[pred], columns=['prediction'])
                df = pd.concat([data_input, prediction], axis=1)
                st.write(df)
                st.subheader(":green[probability] :question:")
                un = pd.DataFrame(pred_proba, columns=['1'])
                zero = pd.DataFrame(np.subtract(1, pred_proba), columns=['0'])
                st.write(pd.concat([zero, un], axis=1).round(2))


# lancer l'application
if __name__ == "__main__":
    main()