Data Science with Python

Loading a CSV

Selecting columns and rows

Filtering with boolean masks

groupby and aggregate

Missing data (NaN, fillna, dropna)

Merging DataFrames

Pivot tables

apply() — custom per-row logic

Time series: dates and resampling

value_counts and unique

String methods (.str)

Plotting basics

Polars: when pandas is too slow

Capstone: end-to-end sales analysis

DataFrame.melt — wide to long

Rolling windows

Type optimization for memory

GroupBy.transform

Polars vs pandas: syntax differences

🎯 Review: Data Science module 1 recap

SQL from pandas: read_sql

Parquet vs CSV

scikit-learn intro: a 5-line model

Avoiding leakage

Cross-validation

pandas MultiIndex (hierarchical indexing)

Window functions: .rolling and .expanding

Categorical data deep-dive

pd.cut and pd.qcut for binning

Time-series resampling tricks

pandas-on-Spark (pyspark.pandas)

DuckDB on Parquet — analytics without a DB

Plotly Express for interactive charts

seaborn for statistical viz

Plotnine (ggplot in Python)

Streamlit dashboards: first app in 50 lines

Streamlit caching + state

Outlier detection (IQR, Z-score, Isolation Forest)

Feature engineering for tabular ML

Stratified train/test/val splits

🎯 Review: Data Science module 2 recap

Cross-validation strategies

sklearn Pipeline + ColumnTransformer

Hyperparameter tuning

XGBoost / LightGBM intro

Capstone: end-to-end ML pipeline

Rolling, expanding, and EWM windows

Pivot tables and multi-index basics

merge_asof for backward-join on time

qcut for quantile-based bucketing

Target encoding (and leakage caveat)

Time-series lag features via shift()

ARIMA, seasonal decomposition, Prophet

Rolling std as a volatility feature

Time-series CV vs random K-fold

Predict: ROC AUC for perfect ranking

Class imbalance: weights vs SMOTE

Fix the leakage: scale-before-split

GridSearchCV vs RandomizedSearchCV vs Optuna

SHAP — local feature attribution

PCA explained-variance calculation

🎯 Review: Data Science module 3 recap

K-Means: silhouette score from scratch

Outlier detection: IQR rule

A/B test sample size calculator

Multi-armed bandits: epsilon-greedy

Capstone: production-grade pipeline + calibration

Streaming z-score outlier detector

Stratified train/test split

Confusion-matrix metrics from scratch

ROC AUC by Mann-Whitney equivalence

Class imbalance: SMOTE-lite oversampling

Min-max & standard scaling

K-fold cross-validation harness

One-hot vs target encoding

Train/test data leakage — find the bug

Cohen's d — effect size

Bootstrap confidence interval

Sliding-window time-series feature

Gradient descent on linear regression

PCA — top component by power iteration

🏆 Capstone: end-to-end Kaggle-style pipeline

🎯 Review: Data Science module 5 recap

Neuron: weights, bias, activation

Backpropagation: chain rule applied

Optimizers: SGD vs Adam vs AdamW

Embeddings: cosine vs Euclidean

Tokenization: BPE / WordPiece / SentencePiece

Transformer: attention in one paragraph

Fine-tuning: full vs LoRA

Evaluation: BLEU / ROUGE / LLM-as-judge

Class imbalance: production fix

Feature stores: why they exist

Drift detection: data vs concept

MLOps: model registry + versioning

Causal inference: correlation ≠ causation

A/B test: minimum detectable effect

Recommenders: collaborative vs content-based

Time series: stationarity check

Pandas: vectorized vs apply vs iterrows

Polars / DuckDB for big DataFrames

Notebooks → production

🏆 Final capstone: ML production blueprint