🏡 Educational DecisionTreeRegressor from scratch

A fully-transparent, from-scratch regression decision-tree with built-in visualisation

---

✨ Why I created it?

What you get	Why it matters
100 % pure-Python implementation (no Cython, no compiled extensions)	Easy to read, extend and debug. Ideal for teaching, code-review sessions or exploratory research.
Step-by-step, self-contained code	Each method mirrors the math you find in textbooks—great for learning how trees calculate variance reduction & information gain.
Handles both numeric and categorical data	Works out-of-the-box on real-world data sets without heavy preprocessing.
Graphviz-powered visualisation CLI	Instantly export a high-resolution PNG of the trained tree to understand splits at a glance.
sklearn-style API	Familiar .fit() / .predict() workflow plus drop-in benchmarking against scikit-learn.
Config-driven hyper-parameters	Tweak depth, information-gain threshold, category limits and more from a single config.py file or per-run CLI flags.

---

🚀 Quick start

1 · Install prerequisites

# System package needed by Graphviz
# Ubuntu / Debian
sudo apt-get install graphviz

# macOS (Homebrew)
brew install graphviz

# Project dependencies
python -m venv venv           
source venv/bin/activate

pip install -r requirements.txt

2 · Train and evaluate

import pandas as pd
from decision_tree import MyDecisionTreeRegressor
from config import DATA_PATH


df = pd.read_csv(DATA_PATH)
df.columns = df.columns.str.replace(" ", "_")
bool_cols = df.select_dtypes("bool").columns
df[bool_cols] = df[bool_cols].astype(int)
df = df.dropna()

X = df.drop(["Hotel_Name", "Price", "Description"], axis=1)
y = np.log(df["Price"])                


tree = MyDecisionTreeRegressor(max_depth=6,
                               min_samples_split=20,
                               min_information_gain=1e-4)
tree.fit(X, y)

pred = tree.predict(X.head())
print(pred)

3 · Visualise the tree (CLI)

python visualize.py \
    --data          data/hotels.csv \
    --output        my_tree          \
    --max-depth     6               \
    --min-samples-split 20

4 · ??? Profit!!!1!