Why Every Data Scientist Should Learn Python in 2026

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

df = pd.read_csv("loans.csv")
print(df.describe(include="all").T.head(10))
sns.boxplot(data=df, x="grade", y="interest_rate")
plt.title("Interest rate by loan grade")
plt.tight_layout()
plt.savefig("rate_by_grade.png", dpi=150)

import pandas as pd
from sqlalchemy import create_engine

engine = create_engine("postgresql+psycopg2://reader:secret@localhost/bank")
query = """
    SELECT grade, AVG(interest_rate) AS avg_rate, COUNT(*) AS n
    FROM loans
    GROUP BY grade
    ORDER BY grade
"""
summary = pd.read_sql(query, engine)
print(summary)

import pandas as pd
import statsmodels.formula.api as smf

df = pd.read_csv("loans.csv")
model = smf.ols("interest_rate ~ credit_score + loan_amount + C(term)", data=df).fit()
print(model.summary().tables[1])

from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.ensemble import HistGradientBoostingClassifier
from sklearn.model_selection import cross_val_score

num = ["credit_score", "loan_amount", "income"]
cat = ["term", "purpose"]
prep = ColumnTransformer([
    ("num", StandardScaler(), num),
    ("cat", OneHotEncoder(handle_unknown="ignore", sparse_output=False), cat),
])
clf = Pipeline([("prep", prep), ("model", HistGradientBoostingClassifier())])
scores = cross_val_score(clf, df[num + cat], df["defaulted"], cv=5, scoring="roc_auc")
print("ROC AUC:", scores.mean().round(3))

# features.py
import pandas as pd

def clean_loans(df: pd.DataFrame) -> pd.DataFrame:
    out = df.copy()
    out = out.drop_duplicates(subset="loan_id")
    out["income"] = out["income"].clip(lower=0)
    out["term"] = out["term"].str.strip().str.lower()
    return out

# test_features.py
import pandas as pd
from features import clean_loans

def test_clean_loans_removes_duplicates_and_negatives():
    raw = pd.DataFrame({
        "loan_id": [1, 1, 2],
        "income": [50000, 50000, -10],
        "term": [" 36 Months", " 36 Months", "60 months "],
    })
    out = clean_loans(raw)
    assert len(out) == 2
    assert (out["income"] >= 0).all()
    assert set(out["term"]) == {"36 months", "60 months"}

loan-default/
├── data/            # not committed; download script instead
├── notebooks/
│   └── 01_eda.ipynb
├── src/
│   ├── features.py
│   └── train.py
├── tests/
│   └── test_features.py
├── requirements.txt
├── Makefile
└── README.md

setup:
	python -m venv .venv && .venv/bin/pip install -r requirements.txt

test:
	.venv/bin/pytest -q

train:
	.venv/bin/python src/train.py

# Loan default prediction
Question: which loans are likely to default in the first year?
Data: public lending dataset (see data/README.md)
Method: gradient boosting with cross-validated ROC AUC
Code: cleaning in src/features.py (tested), training in src/train.py
Run: make setup && make test && make train
Limits: historical data only; not tested on current applicants