About

6 Essential Data Science Coding Techniques
This canvas collects six practical coding techniques that every data scientist should know. Each section contains runnable Python code cells, explanatory text, and print statements so you can see the outcomes as you go. The examples progress in difficulty — from beginner transformations to advanced performance optimizations — and are designed to be self-contained.

What’s inside:
- One-Hot Encoding (Beginner): Use scikit-learn’s OneHotEncoder to transform categorical values into model-ready numeric vectors. Example with "city" and "color" features, showing before-and-after outputs.
- Vectorized GroupBy Aggregations with pandas (Intermediate): Learn how to summarize transactional data into customer-level features. Includes totals, averages, recency, distinct counts, and rolling 30-day activity.
- Custom scikit-learn Transformer (Intermediate → Advanced): A reusable transformer (LogStandardizeTransformer) that applies a log transform, optional standardization, and supports inverse transforms. Demonstrates how to plug into Pipeline and ColumnTransformer.
- SQL-in-Pandas (Window Functions) (Intermediate → Advanced): Translating SQL window functions into pandas: rolling averages, ranking, and lag-based percent changes. Shows how groupby, transform, rolling, and shift map to SQL concepts like OVER, RANK, and LAG.
- Custom Loss Function (Advanced): Define an asymmetric loss that penalizes under-predictions more than over-predictions. Implementation for both XGBoost (custom objective) and scikit-learn (custom scorer).
- Vectorization Challenge with NumPy (Advanced): Four independent implementations of pairwise Euclidean distances: Nested loops (baseline), Semi-vectorized, Full broadcasting, Norm identity formula. Benchmark harness included to compare runtime and correctness.

These are the kinds of coding techniques that show up in real projects — preparing categorical data, aggregating events, customizing pipelines, translating SQL logic, aligning model objectives with business goals, and making code efficient enough to scale.