Create an interactive app for business users to consume an ML model in a simple and intuitive way, for example requesting a Chun prediction for one given customer.

This template is for developing a UI to easily consume ML models. By following this template, you can create ML-powered interactive applications that can be easily used by a wide range of business users, and that can deliver real value and insights to end-users.

This tutorial is tailored for the Churn Prediction model, but can be modified to work with other models as well.

Import required modules & load models

from sklearn.feature_extraction.text import TfidfVectorizer
from joblib import load
import pandas as pd

# Loading models to make predictions
model = load('/data_app/model_churn_prediction')
encoder = load('/data_app/encoder_churn_prediction')

Building a UI

The app is designed to allow users to Search records by customer name and obtain churn predictions for a customer. The UI includes a Submit button to get the records and initiate the analysis process.

st.title('Churn Prediction')
st.write("Welcome to Peliqan's Churn Prediction app!")

form = st.form(key='churn-form')
user_input = form.text_input('Search by customer name')
submit = form.form_submit_button('Submit')

Learn more about Streamlit functionality here.

Create a function to get records on search

We are applying the user search to find records in an SQL query:

def getRecords(searchTerm):
    query = f"select * from customers where lower(CustomerName) like lower('{searchTerm}%') limit 5"
    data = pq.load_table(query=query)
    st.dataframe(data) # show searched results
    return data

Write a function to make predictions

Now we will define the function makePrediction that takes a user_input parameter as input. It uses a previously trained encoder and model object to predict the Churn probability of customer:

def makePrediction(user_input):
    df = getRecords(user_input)
    if df.empty:
        return st.error(f"Customer with name {user_input} does not exist.")
    X = df.drop(['Prediction', 'Churn', 'CustomerName'], axis=1)
    # converting categorical columns to numeric
    cat_cols = X.select_dtypes('object').columns
    X[cat_cols] = X[cat_cols].apply(lambda x: encoder[x.name].transform(x))

	# Making prediction
    prediction = model.predict(X)
    proba = model.predict_proba(X)[0][prediction]
    result = pd.DataFrame({'CustomerID':X['CustomerID'], 'Customer Name':df['CustomerName'], 'Prediction': prediction, 'Probability': proba})
    return result

Logic for Submit button

When the submit button is pressed, the makePrediction() function is called with the user input as the argument. The predicted churn score for each found record is extracted from the function's returned result. Depending on the predicted score, the UI displays a message. We also plot a bar chart for Prediction vs Customer name.

if submit:
    result = makePrediction(user_input)
    if type(result) == pd.DataFrame:
        st.header('Predictions')
        st.dataframe(result)
        st.header('Prediction of churn vs Customers')
        st.bar_chart(result, y='Prediction', x='Customer Name')

‣

Expand this to see the full code

from sklearn.feature_extraction.text import TfidfVectorizer
from joblib import load
import pandas as pd

# Loading models to make predictions
model = load('/data_app/model_churn_prediction')
encoder = load('/data_app/encoder_churn_prediction')

st.title('Churn Prediction')
st.write("Welcome to Peliqan's Churn Prediction app!")

form = st.form(key='churn-form')
user_input = form.text_input('Search by customer name')
submit = form.form_submit_button('Submit')

def getRecords(searchTerm):
    query = f"select * from public.customer_churn where lower(CustomerName) like lower('{searchTerm}%') limit 5"
    data = pq.load_table(query=query)
    st.dataframe(data) # show searched results
    return data

def makePrediction(user_input):
    df = getRecords(user_input)
    if df.empty:
        return st.error(f"Customer with name {user_input} does not exist.")
    X = df.drop(['Prediction', 'Churn', 'CustomerName'], axis=1)
    # converting categorical columns to numeric
    cat_cols = X.select_dtypes('object').columns
    X[cat_cols] = X[cat_cols].apply(lambda x: encoder[x.name].transform(x))

	# Making prediction
    prediction = model.predict(X)
    proba = model.predict_proba(X)[0][prediction]
    result = pd.DataFrame({'CustomerID':X['CustomerID'], 'Customer Name':df['CustomerName'],'Prediction': prediction, })#'Probability': proba})
    return result

if submit:
    result = makePrediction(user_input)
    if type(result) == pd.DataFrame:
        st.header('Predictions')
        st.dataframe(result)
        st.header('Prediction of churn vs Customers')
        st.bar_chart(result, y='Prediction', x='Customer Name')

What’s Next