Predicting house prices

Objectives

• Example end-to-end supervised learning workflow with Ames Housing dataset
• Focus on conceptual understanding of machine learning
• Demonstrate use of Predictive Power Score (PPS) for feature selection
• Demonstrate capabilities of AutoML model comparison with lazypredict

Attribution

Dataset

• Ames Housing dataset paper (original paper)
• Kaggle competition advanced regression techniques (link)

Python libraries

• Altair (docs)
• ydata-profiling (docs)
• Predictive Power Score (PPS, GitHub, blog)
• lazypredict: runs all sklearn estimators and ranks them (GitHub)

import marimo as mo

import altair as alt
import pandas as pd
import ppscore as pps
import warnings
from lazypredict.Supervised import LazyRegressor
from sklearn.model_selection import train_test_split
from ydata_profiling import ProfileReport

warnings.filterwarnings("ignore")

# Altair theme: clean axes
def y_axis():
    return {
        "config": {
            "axisX": {"grid": False},
            "axisY": {
                "domain": False,
                "gridDash": [2, 4],
                "tickSize": 0,
                "titleAlign": "right",
                "titleAngle": 0,
                "titleX": -5,
                "titleY": -10,
            },
            "view": {
                "stroke": "transparent",
                "continuousHeight": 300,
                "continuousWidth": 400,
            },
        }
    }

alt.themes.register("y_axis", y_axis)
alt.themes.enable("y_axis")


def get_descriptions():
    """Parse column descriptions from the Ames Housing data_description.txt."""
    with open("data_description.txt") as f:
        descriptions = {}
        for line in f.readlines():
            if ":" in line and "2nd level" not in line:
                parts = line.split(": ")
                descriptions[parts[0].strip()] = parts[1].strip()
    return pd.Series(descriptions).rename("description")

descriptions = get_descriptions()

Read and explore the data

train = pd.read_csv("train.csv")
test = pd.read_csv("test.csv")
profile = ProfileReport(train, minimal=True, title="Ames Housing Profiling Report")
profile.to_file("ames-housing-profiling-report-minimal.html")

mo.Html(
    '<iframe width="100%" height="800px" src="ames-housing-profiling-report-minimal.html"></iframe>'
)

Feature selection with Predictive Power Score

We use the Predictive Power Score to identify the top 30 features most predictive of SalePrice. This is especially useful for the Ames Housing dataset which has 79 features, many of which are categorical.

# Compute PPS for all numeric features predicting SalePrice
numeric_train = train.select_dtypes(include="number")
pps_scores = pps.predictors(numeric_train, "SalePrice").sort_values(
    "ppscore", ascending=False
)
top30 = pps_scores.head(30)

pps_bar = (
    alt.Chart(top30)
    .mark_bar()
    .encode(
        x=alt.X("ppscore:Q", title="PPS", scale=alt.Scale(domain=[0, 1])),
        y=alt.Y("x:N", sort="-x", title="Feature"),
        tooltip=["x", alt.Tooltip("ppscore:Q", format=".3f")],
    )
    .properties(title="Top 30 features by Predictive Power Score (predicting SalePrice)", height=500)
)
pps_bar

AutoML model comparison with lazypredict

LazyRegressor runs all scikit-learn regressors and ranks them. We train on the top 30 PPS-selected numeric features to keep the comparison fast and interpretable.

top30_features = top30["x"].tolist()
# Remove target from features if present
top30_features = [f for f in top30_features if f != "SalePrice"]

X = train[top30_features].fillna(train[top30_features].median())
y = train["SalePrice"]

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.3, random_state=42
)

reg = LazyRegressor(verbose=0, ignore_warnings=True, custom_metric=None)
models_reg, predictions_reg = reg.fit(X_train, X_test, y_train, y_test)
models_reg_df = models_reg.reset_index().rename(columns={"index": "Model"})
models_reg_df

# Bar chart ranked by R-Squared
r2_chart = (
    alt.Chart(models_reg_df)
    .mark_bar()
    .encode(
        x=alt.X("R-Squared:Q", title="R²", scale=alt.Scale(domain=[0, 1])),
        y=alt.Y("Model:N", sort="-x", title=None),
        color=alt.Color(
            "R-Squared:Q",
            scale=alt.Scale(scheme="blues", domain=[0, 1]),
            legend=None,
        ),
        tooltip=[
            "Model",
            alt.Tooltip("R-Squared:Q", format=".3f"),
            alt.Tooltip("RMSE:Q", format=".0f"),
            alt.Tooltip("Time Taken:Q", format=".2f"),
        ],
    )
    .properties(title="Model comparison — R² (test set)", height=600)
)
r2_chart