app.py

import streamlit as st
import hopsworks
import xgboost
import plotly.graph_objs as go
import plotly.express as px
import joblib
import math
import pandas as pd
import os

st.title('🔮 Customer Churn Prediction')

st.write(36 * "-")
st.header('\n📡 Connecting to Hopsworks Feature Store...')


def header(text):
    st.write(36 * "-")
    st.write('#### ' + text)

project = hopsworks.login(project = "annikaij", api_key_value=os.environ['HOPSWORKS_API_KEY'])
fs = project.get_feature_store()

header('🪄 Retrieving Feature View...')

feature_view = fs.get_feature_view(
    name="churn_feature_view",
    version=1
    )

st.text('Done ✅')
header('⚙️ Reading DataFrames from Feature View...')

@st.cache_data()
def retrive_data(feature_view=feature_view):
    feature_view.init_batch_scoring(1)
    batch_data = feature_view.get_batch_data()
    batch_data.drop('customerid', axis=1, inplace=True)
    df_all = feature_view.query.read()
    df_all.drop('churn', axis=1, inplace=True)
    return batch_data, df_all


batch_data, df_all = retrive_data()

st.dataframe(df_all.head())
st.text(f'Shape: {df_all.shape}')
header('🔮 Model Retrieving...')


@st.cache_data()
def get_model(project=project):
    mr = project.get_model_registry()
    model = mr.get_model("churnmodel", version=1)
    model_dir = model.download()
    return joblib.load(model_dir + "/churnmodel.pkl")


model = get_model()
st.write(model)


def transform_preds(predictions):
    return ['Churn' if pred == 1 else 'Not Churn' for pred in predictions]


header('📝 Batch Data Prediction...')

st.dataframe(batch_data.head())

predictions = model.predict(batch_data)
predictions = transform_preds(predictions)

df_all['Churn'] = predictions

result_table = df_all[['customerid', 'Churn']]

st.text(f'👩🏻‍⚖️ Predictions for 5 rows:\n {predictions[:5]}')
header('💳 Prediction by Customer Id...')

with st.form(key="Selecting Customer ID"):
    option = st.selectbox(
             'Select a Custimer ID to return a predict.',
             (result_table.customerid.values[:15])
          )
    submit_button = st.form_submit_button(label='Submit')

if submit_button:   
    result = result_table[result_table.customerid == option]['Churn'].values

    st.text(f'👮🏻‍♂️ Customer ID: {option}')
    st.text(f'👩🏻‍⚖️ Prediction: {result}')

header('👨🏻‍🎨 Prediction Visualizing...')


feature_names = batch_data.columns

feature_importance = pd.DataFrame(feature_names, columns=["feature"])
feature_importance["importance"] = model.feature_importances_
feature_importance = feature_importance.sort_values(by=["importance"], ascending=False)

fig_importance = px.bar(
    feature_importance,
    x='feature',
    y='importance',
    title='Feature Importance Plot'
     )

fig_importance.update_xaxes(tickangle=23)
fig_importance.update_xaxes(title="Feature")
fig_importance.update_yaxes(title="Importance")
fig_importance.update_traces(hovertemplate='Feature: %{x} <br>Importance: %{y}') 

st.plotly_chart(fig_importance)


def plot_histogram(data, x_col, title, xlabel, ylabel):

    fig = go.Figure()

    fig = px.histogram(
        data,
        x=x_col,
        color="Churn",
        title=title
    )

    fig.update_xaxes(title=xlabel)
    fig.update_yaxes(title=ylabel)
    fig.update_traces(hovertemplate=xlabel + ': %{x} <br>' + ylabel + ': %{y}')

    return fig


st.plotly_chart(plot_histogram(df_all, 'internetservice', 'Churn rate according to internet service subscribtion', 'Internet service', 'Number of customers'))
st.plotly_chart(plot_histogram(df_all, 'streamingmovies', 'Churn rate according to streaming movies subscribtion', 'Streaming movies', 'Number of customers'))
st.plotly_chart(plot_histogram(df_all, 'streamingtv', 'Churn rate according to internet streaming tv subscribtion', 'Gender', 'Number of customers'))
st.plotly_chart(plot_histogram(df_all, 'gender', 'Churn rate according to Gender', 'Gender', 'Number of customers'))
st.plotly_chart(plot_histogram(df_all, 'totalcharges', 'Distribution of Total Charges according to Churn/Not', "Charge Value", 'Number of customers'))
st.plotly_chart(plot_histogram(df_all, 'paymentmethod', 'Amount of each Payment Method', "Payment Method", 'Total Amount'))
st.plotly_chart(plot_histogram(df_all, 'partner', 'Affect of having a partner on Churn/Not', "Have a partner", 'Number of customers'))

st.success('🎉 📈 🤝 App Finished Successfully 🤝 📈 🎉')