Streamlit for ML

Streamlit turns a Python script into an interactive web app in minutes — the fastest way to build a demo, internal tool, or stakeholder-facing interface for a trained model, without writing any HTML, CSS or JavaScript.

A Complete ML Demo App

import streamlit as st
import joblib
import numpy as np

st.title("Loan Approval Predictor")
st.write("Enter applicant details to get a prediction.")

model = joblib.load("model_pipeline.pkl")

income = st.number_input("Annual Income", min_value=0, value=45000, step=1000)
age = st.slider("Age", min_value=18, max_value=100, value=34)
credit_score = st.slider("Credit Score", min_value=300, max_value=850, value=680)

if st.button("Predict"):
    features = np.array([[income, age, credit_score]])
    prediction = model.predict(features)[0]
    probability = model.predict_proba(features)[0].max()

    if prediction == 1:
        st.success(f"Approved (confidence: {probability:.1%})")
    else:
        st.error(f"Declined (confidence: {probability:.1%})")

# Run with:
# streamlit run app.py

That's the entire app — number inputs, sliders, a button, and conditional styled output, all from a handful of Python lines with no separate frontend code.

Adding Visualizations

import matplotlib.pyplot as plt

feature_importance = model.named_steps["model"].feature_importances_
feature_names = ["income", "age", "credit_score"]

fig, ax = plt.subplots()
ax.barh(feature_names, feature_importance)
st.pyplot(fig)   # Streamlit renders any matplotlib figure directly

When Streamlit Is the Right Tool

Practical Use Cases

• Showing a trained model's behavior to non-technical stakeholders interactively, without building a full frontend
• Internal exploration and debugging tools for a data science team

Common Mistakes

• Using Streamlit as a production API backend for another application — it's built for interactive human use, not machine-to-machine API calls; use Flask or FastAPI for that.
• Loading the model inside code that reruns on every widget interaction instead of using @st.cache_resource to load it once.

# Cache the model load so it only happens once, not on every interaction
@st.cache_resource
def load_model():
    return joblib.load("model_pipeline.pkl")

model = load_model()

Interview Relevance

Q: "When would you use Streamlit instead of building a Flask or FastAPI service?" When the goal is a quick, interactive demo or internal tool for humans to explore a model directly — not a production API meant to be called programmatically by other systems; Streamlit trades API-style flexibility for extremely fast interactive UI development.

Practice Question

Modify the app above to add a file upload widget that lets a user upload a CSV of multiple applicants and see predictions for all of them at once.

Related ML Notes

• Flask ML API
• ML Model Deployment
• SHAP