Data Science — Python

Notebook Structure

Organize notebooks in this order:

1. Imports & configuration
2. Data loading
3. Exploratory data analysis (EDA)
4. Feature engineering
5. Modeling
6. Evaluation
7. Conclusions & next steps

Keep notebooks for exploration. Move reusable logic to src/ Python modules with tests.

Data Validation

# Always validate after loading
assert df.shape[0] > 0, "DataFrame is empty"
assert df.isnull().sum().sum() == 0, f"Nulls found: {df.isnull().sum()}"
assert df['price'].between(0, 1_000_000).all(), "Price out of expected range"

# For production pipelines: use pandera
import pandera as pa
schema = pa.DataFrameSchema({
    "price": pa.Column(float, pa.Check.ge(0)),
    "category": pa.Column(str, pa.Check.isin(["A", "B", "C"])),
})
schema.validate(df)

Reproducibility

import random
import numpy as np

SEED = 42
random.seed(SEED)
np.random.seed(SEED)
# sklearn: pass random_state=SEED to all estimators

• Pin all dependency versions in requirements.txt or pyproject.toml.
• Track experiments: MLflow, or a simple experiments/ log with metadata JSON.
• Save models with timestamp + metadata: model_rf_20260115_v1.pkl.

Model Documentation

Document for every model:

• Training data: source, date range, size, preprocessing steps.
• Feature list and engineering decisions.
• Hyperparameters and tuning approach.
• Evaluation metrics on held-out test set.
• Known limitations and failure modes.

Code Quality

# Extract transforms into sklearn Pipeline
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier

pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', RandomForestClassifier(random_state=SEED)),
])

• Write unit tests for data processing functions in tests/.
• Use pathlib.Path — never hardcode file paths.
• Use logging not print in production code.