File size: 3,026 Bytes
863da02
 
b7fb521
863da02
 
b7fb521
 
 
 
034860a
b7fb521
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
import numpy as np
import pandas as pd
import joblib
import streamlit as st

st.set_page_config(page_title="Insurance Charges Predictor", page_icon="🩺", layout="centered")

@st.cache_resource
def load_model():
    return joblib.load("src/gb_model.pkl")

def build_features(age, sex, bmi, children, smoker, region, expected_cols=None):
    row = pd.DataFrame([{
        "age": int(age),
        "sex": sex,
        "bmi": float(bmi),
        "children": int(children),
        "smoker": smoker,
        "region": region
    }])

    # Same idea as training: pd.get_dummies(X, drop_first=True)
    X = pd.get_dummies(row, drop_first=True)

    # Align to training columns if we know them
    if expected_cols is not None:
        X = X.reindex(columns=expected_cols, fill_value=0)

    return X

st.title("🩺 Insurance Charges Predictor")
st.write("Enter the details and get an estimated medical insurance charge.")

# Load model
try:
    model = load_model()
    st.caption("Loaded model: `gb_model.pkl`")
except Exception as e:
    st.error(f"Could not load model: {e}")
    st.stop()

# Detect expected columns (works if model was fit with a DataFrame in sklearn)
expected_cols = list(model.feature_names_in_) if hasattr(model, "feature_names_in_") else None
if expected_cols is None:
    st.warning(
        "I could not detect the training columns from the model. "
        "If you get a column mismatch error, see the fix at the bottom."
    )

with st.sidebar:
    st.header("Options")
    log_model = st.checkbox("My model predicts log(charges) (I used np.log1p on charges)", value=False)

c1, c2 = st.columns(2)
with c1:
    age = st.number_input("Age", min_value=0, max_value=120, value=30, step=1)
    bmi = st.number_input("BMI", min_value=10.0, max_value=80.0, value=27.5, step=0.1)
    children = st.number_input("Children", min_value=0, max_value=10, value=0, step=1)

with c2:
    sex = st.selectbox("Sex", ["female", "male"])
    smoker = st.selectbox("Smoker", ["no", "yes"])
    region = st.selectbox("Region", ["northeast", "northwest", "southeast", "southwest"])

if st.button("Predict Charges", type="primary"):
    X_input = build_features(age, sex, bmi, children, smoker, region, expected_cols)

    try:
        pred = float(model.predict(X_input)[0])
        if log_model:
            pred = float(np.expm1(pred))

        st.success(f"Estimated Insurance Charges: **{pred:,.2f}**")

        with st.expander("Show encoded input features"):
            st.dataframe(X_input)

        if hasattr(model, "feature_importances_") and expected_cols is not None:
            fi = pd.DataFrame({
                "Feature": expected_cols,
                "Importance": model.feature_importances_
            }).sort_values("Importance", ascending=False)

            st.subheader("Feature Importance")
            st.dataframe(fi)

    except Exception as e:
        st.error(f"Prediction failed: {e}")

st.markdown("---")
st.caption("Tip: The encoding must match the training data (same dummy columns).")