The stats, queries, metrics, and model behaviour you should know cold
— type them until they're reflex.
Math · linear algebra, probability, calculus, distributions, entropy
Statistics · distributions, estimation, hypothesis tests, effect sizes
Wrangling · grain, joins, NULLs, window functions, dataframe hygiene
Data · splits, leakage, scaling, encoding, imbalance, missing values
Analytics · metric design, funnels, cohorts, honest charts
Experiments · A/B design, validity threats, causal inference
Time series · stationarity, ARIMA, backtesting, forecast metrics
Classical · linear & logistic regression, trees, SVM, kNN, ensembles
Unsupervised · k-means, hierarchical, DBSCAN, PCA/SVD, t-SNE, anomalies
Evaluation · confusion matrix, precision/recall, ROC-AUC, CV, calibration
Optimization · loss functions, gradient descent, LR, regularization, bias-variance
Neural · perceptron, activations, backprop, init, batchnorm, dropout
Architectures · CNNs, RNN/LSTM, attention, transformers, embeddings, transfer
LLMs · tokenization, pretraining, fine-tuning, RLHF, RAG, decoding, scaling
Python · the NumPy, pandas, scikit-learn, PyTorch and matplotlib lines you type every day
Reinforcement learning · MDPs, Bellman, TD and Q-learning, bandits, policy gradients, PPO
Recommenders · collaborative filtering, factorization, retrieve-then-rank, cold start, feedback loops
MLOps · reproducibility, pipelines, deployment, serving, monitoring, retraining
Responsible ML · bias, fairness definitions, interpretability, privacy, accountability
Vision & edge AI · practical workflows, deployment and evaluation
Terms · every abbreviation in gradflow, drilled in both directions
Everything · mixed endless decks across all layers + fill-in-the-blank