Create app.py

app.py

@@ -0,0 +1,109 @@
+import streamlit as st
+import pandas as pd
+import joblib
+import shap
+import matplotlib.pyplot as plt
+
+# Load the model
+@st.cache_resource
+def load_model():
+    return joblib.load('best_model.pkl')
+
+model = load_model()
+
+# Streamlit app title and description
+st.title("Retention Probability Predictor")
+st.markdown("""
+Predict the probability of retention based on patient and treatment details. 
+Provide the required inputs below to see the prediction and feature contributions.
+""")
+
+# Sidebar for inputs
+st.sidebar.header("Input Features")
+
+# Input fields for features
+st.sidebar.subheader("Episode Details")
+previous_bup_episodes = st.sidebar.number_input(
+    "Number of Previous Buprenorphine Episodes:", min_value=0, value=20, step=1
+)
+previous_failed_bup_episodes = st.sidebar.number_input(
+    "Number of Previous Failed Buprenorphine Episodes:", min_value=0, value=20, step=1
+)
+
+st.sidebar.subheader("Days Details")
+days_covered = st.sidebar.number_input(
+    "Days Covered in Current Episode:", min_value=0, value=30, step=1
+)
+days_since_start = st.sidebar.number_input(
+    "Days Since Start of the Episode:", min_value=1, value=90, step=1
+)
+
+st.sidebar.subheader("Other Features")
+prior_BUP_days_supply = st.sidebar.number_input(
+    "Prior BUP Days Supply:", min_value=0, value=30, step=1
+)
+YearOfInitiation = st.sidebar.number_input(
+    "Year of Initiation:", min_value=1900, max_value=2100, value=2023, step=1
+)
+Previous_Meth_Episodes = st.sidebar.radio(
+    "Previous Methadone Episodes (Yes/No):",
+    options=[0, 1],  # Binary options
+    format_func=lambda x: "Yes" if x == 1 else "No"  # Display "Yes" or "No" instead of 0/1
+)
+
+# Calculate derived features
+if st.sidebar.button("Predict Retention Probability"):
+    # Calculate Failed Buprenorphine Rate
+    if previous_bup_episodes > 0:
+        Failed_Bup_Rate = previous_failed_bup_episodes / previous_bup_episodes
+    else:
+        Failed_Bup_Rate = 0.0  # No prior episodes, so no failures
+
+    # Calculate Proportion of Days Covered
+    if days_since_start > 0:
+        proportionofDaysCovered = days_covered / days_since_start
+    else:
+        proportionofDaysCovered = 0.0  # Avoid division by zero
+
+    # Prepare input data for prediction
+    input_data = pd.DataFrame({
+        'proportionofDaysCovered': [proportionofDaysCovered],
+        'Failed_Bup_Rate': [Failed_Bup_Rate],
+        'prior_BUP_days_supply': [prior_BUP_days_supply],
+        'YearOfInitiation': [YearOfInitiation],
+        'Previous_Meth_Episodes': [Previous_Meth_Episodes]
+    })
+
+    # Make prediction
+    retention_probability = model.predict_proba(input_data)[:, 1][0]
+
+    # Display the prediction
+    st.subheader("Prediction Result")
+    st.success(f"The predicted retention probability is: **{retention_probability:.2f}**")
+
+    # Display calculated features
+    st.subheader("Calculated Features")
+    st.write(f"**Failed Buprenorphine Rate:** {Failed_Bup_Rate:.2%}")
+    st.write(f"**Proportion of Days Covered:** {proportionofDaysCovered:.2%}")
+
+    # Explain model prediction using SHAP
+    st.subheader("Feature Contribution to Prediction")
+    explainer = shap.Explainer(model, input_data)  # Create a SHAP explainer
+    shap_values = explainer(input_data)
+
+    # Plot feature importance
+    shap.force_plot(
+        explainer.expected_value[1],
+        shap_values.values[0],
+        input_data.iloc[0],
+        matplotlib=True,
+        show=False
+    )
+    st.pyplot(plt.gcf())  # Display SHAP force plot
+
+    st.write("The above visualization shows how each feature contributes to the prediction.")
+
+    # Feature importance bar chart
+    st.subheader("Feature Importance (Bar Chart)")
+    shap.summary_plot(shap_values, input_data, plot_type="bar", show=False)
+    st.pyplot(plt.gcf())