first commit

app.py

@@ -0,0 +1,14 @@
+import streamlit as st
+
+import eda
+import predictor
+
+navigation = st.sidebar.selectbox(
+    label='Pilih Halaman',
+    options=('EDA', 'Predictor')
+)
+
+if navigation == 'EDA':
+    eda.run()
+elif navigation == 'Predictor':
+    predictor.run()

config.ini

@@ -0,0 +1,8 @@
+[DEFAULT]
+production = True
+
+[DEVELOPMENT]
+URL = http://127.0.0.1:5000/predic
+
+[PRODUCTION]
+URL = https://backend-telco-churn-kidfrom.koyeb.app

eda.py

@@ -0,0 +1,54 @@
+import streamlit as st
+from helpers import query, plot
+from PIL import Image
+
+st.set_page_config(
+    page_title='Telkomsel - PredictorTelkomsel',
+    layout='wide',
+    initial_sidebar_state='expanded'
+)
+
+def run():
+    st.title('Telkomsel Customers Exploratory Data Analysis')
+    
+    st.subheader('Exploratory Data Analysis of Telkomsel Customers')
+    
+    st.write("This page is made by Jason Rich Darmawan Onggo Putra")
+    
+    st.write("Disclaimer: the data set used is not real.")
+    
+    df = query.fetch_all_data()
+    
+    st.write("## Histogram of categorical features")
+    st.pyplot(fig=plot.plot_categorical_features(df=df))
+    
+    st.write("## Pairplot of numerical features")
+    st.pyplot(fig=plot.plot_numerical_features(df))
+    
+    st.write("## Model Layers")
+    image = Image.open("./images/sequential_improved_model.png")
+    st.image(image, caption='Sequential Improved Model')
+    
+    st.write("## Model Strengths and Weaknesses")
+    image = Image.open("./images/sequential_improved_prediction.png")
+    st.image(image, caption='Sequential Improved Model Strengths and Weaknesses')
+    
+    st.markdown(
+        """
+        We will inform management, to use this model for a specific customer segment which is more predictable, according to the model:
+        1. A customer with one year or two year contract.
+        2. An old customer / customer with tenure above 40 / customer with total charges above 4000.
+        3. A customer without internet service.
+        4. A customer with internet service is unpredictable. 
+
+           However, a customer with internet service and 1 related internet service will make the customer more predictable.
+           
+           e.g A customer with tech support / online security / online backup.
+
+        5. A customer that pays with Bank Transfer (automatic)
+        6. A customer with monthly charges below 20.
+
+        We will also inform management, not to use this model for a specific customer segment, which is less predictable according to the model:
+        1. A customer that is paying with Electronic check / Mail check / Credit Card (automatic).
+        """
+    )

helpers/plot.py

@@ -0,0 +1,47 @@
+import matplotlib.pyplot as plt
+import seaborn as sns
+import pandas as pd
+
+def __histplot(features, target, df):
+    """
+    Usage: features parameter should be nominal features.
+    """
+    fig, axes = plt.subplots(4,4, figsize=(12,8), layout='constrained')
+
+    for feature, ax in zip(features, axes.ravel()):
+        sns.histplot(data=df, x=feature, hue=target, multiple='dodge', 
+            stat='percent', common_norm=False, 
+        ax=ax)
+        if feature != 'gender':
+            ax.get_legend().set_visible(False)
+        if feature in [
+            'MultipleLines', 'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 
+            'TechSupport', 'StreamingTV', 'StreamingMovies', 'PaymentMethod',
+            'Contract'
+        ]:
+            ax.set_xticks(ax.get_xticks())
+            ax.set_xticklabels(ax.get_xticklabels(), rotation=15, ha='right')
+            
+    return fig
+
+def plot_categorical_features(df):
+    return __histplot(
+        features=[
+            'gender', 'SeniorCitizen', 'Partner', 'Dependents', 
+            'PhoneService', 'MultipleLines', 'InternetService', 
+            'OnlineSecurity', 'OnlineBackup', 'DeviceProtection', 
+            'TechSupport', 'StreamingTV', 'StreamingMovies', 
+            'PaperlessBilling', 'PaymentMethod', 'Contract'
+        ],
+        target='Churn',
+        df=df
+    )
+    
+def plot_numerical_features(df):
+    df_c = df.copy()
+    df_c['TotalCharges'] = pd.to_numeric(df_c['TotalCharges'], errors='coerce')
+    
+    return sns.pairplot(
+        data=df_c[['tenure', 'MonthlyCharges', 'TotalCharges', 'Churn']], 
+        hue='Churn'
+    )

helpers/query.py

@@ -0,0 +1,12 @@
+import requests
+import pandas as pd
+
+def fetch_all_data() -> pd.DataFrame:
+    URL = "http://127.0.0.1:5000/query"
+    r = requests.get(URL)
+    
+    if r.status_code == 200:
+        res = r.json()
+        return pd.DataFrame(res)
+    else:
+        return ('Error with status code ', str(r.status_code))

predictor.py

@@ -0,0 +1,48 @@
+import streamlit as st
+import requests
+import pandas as pd
+import streamlit_toggle as tog
+from configparser import ConfigParser
+
+str_to_array = lambda items: [item.strip("\" ") for item in items.split(',')]
+
+config = ConfigParser()
+config.read('./config.ini')
+
+if config['DEFAULT']['production'] == 'True':
+    URL = config['PRODUCTION']['URL']
+else:
+    URL = config['DEVELOPMENT']['URL']
+    
+def run():
+    with st.form(key='predictor'):
+        customerID = st.text_input(
+            label="Customer IDs, separate it with a comma"
+        )
+        fetch_customer_data = tog.st_toggle_switch(
+            label="Fetch the customer data",
+        )
+
+        submitted = st.form_submit_button('Predict')
+        
+        if submitted:
+            customerID_final = str_to_array(customerID)
+            print("[DEBUG] customerID:", customerID_final)
+            
+            r = requests.post(
+                URL+"/predict", 
+                json={
+                    "customerID": customerID_final,
+                    "fetch_customer_data": fetch_customer_data
+                }
+            )
+            
+            if r.status_code == 200:
+                res = r.json()
+                st.dataframe(pd.DataFrame(res))
+            else:
+                st.write('Error with status code ', str(r.status_code))
+         
+# If we deploy the app, __name__ is app.       
+if __name__ == '__main__':
+    run()

requirements.txt

@@ -0,0 +1,9 @@
+streamlit
+streamlit_toggle
+
+pandas
+matplotlib
+seaborn
+
+# pillow is the package name for PIL
+pillow