Machine Learning, Deeply Explained.
Not textbook summaries. Every algorithm explained with intuition, math, code, and interview-level depth. Built for engineers who want to actually understand ML.
48/55
Topics · built · in progress
720+
Interview Qs · per topic with follow-ups
480+
Tricky Qs · deep understanding tests
Zero to ML Foundations
4 topicsWhat is Machine Learning?
Rule-based programming vs ML, training, inference, and failure modes.
ML Problem Types
Regression, classification, clustering, anomalies, ranking, recommendations, and forecasting.
Dataset Thinking
Features, labels, splits, leakage, noise, missing data, outliers, and imbalance.
ML Pipeline Overview
Problem to monitoring: the full map of a real ML workflow.
Core ML Thinking
15 topicsOverfitting vs Underfitting
Diagnose whether the model is too simple or too fragile.
Bias-Variance Tradeoff
Core tradeoff lens for choosing model capacity and regularization.
Regularization
Control complexity to improve stability and generalization.
Loss Functions
Define what the model optimizes and what errors are expensive.
Optimization
How parameter updates drive convergence and stability.
Generalization
Measure unseen-data performance, not just training fit.
Data Leakage
Prevent hidden leaks that fake performance.
Model Complexity
Choose capacity to balance fit quality and robustness.
Feature Scaling
Normalize magnitudes for stable optimization and geometry.
Model Interpretability
Explain model behavior and decision drivers clearly.
Error Analysis
Turn failure slices into targeted model improvements.
Model Selection
Select models by quality, latency, and maintainability tradeoffs.
Baseline Models
Anchor progress before advanced modeling.
Evaluation Strategy
Design validation that reflects product reality.
Production Failure Cases
Common deployment failures and prevention patterns.
Foundations
6 topicsLinear Regression
Fit a line. Predict a number. Understand everything.
Logistic Regression
Binary classification via sigmoid probability outputs.
Gradient Descent
The engine behind almost every ML algorithm.
Regularization
Ridge, Lasso, and ElasticNet — fighting overfitting.
Bias-Variance Tradeoff
The fundamental tension at the heart of every ML model.
Maximum Likelihood Estimation
The math behind most loss functions — deeply explained.
Tree-Based Methods
4 topicsDecision Trees
Recursive binary splits on a dataset.
Random Forest
An ensemble of decorrelated trees.
Gradient Boosting
Sequential learners that fix prior errors — XGBoost, LightGBM.
Ensemble Methods
Stacking, blending, and voting — combining models intelligently.
Distance-Based
2 topicsClustering
8 topicsK-Means Clustering
Partition data into k centroids iteratively.
K-Medoid
Robust variant of K-Means using actual data points.
Hierarchical Clustering
Build a tree of merges or splits (dendrograms).
DBSCAN
Density-based spatial clustering with noise detection.
OPTICS
Density-based ordering for variable-density clusters.
BIRCH
Balanced iterative reducing and clustering using hierarchies.
Affinity Propagation
Message-passing clustering without k selection.
Mean Shift
Non-parametric density mode seeking algorithm.
Probabilistic
2 topicsDimensionality Reduction
4 topicsPrincipal Component Analysis
Project high-dimensional data to lower dimensions.
t-SNE
Visualize high-dimensional data in 2D with perplexity-based embeddings.
UMAP
Topology-preserving dimensionality reduction — faster than t-SNE.
LDA (Linear Discriminant Analysis)
Supervised dimensionality reduction maximizing class separability.
Neural Networks
3 topicsNeural Network Basics
Perceptrons, layers, weights — the foundation of deep learning.
Backpropagation
How neural networks learn — chain rule through computation graphs.
Activation Functions
ReLU, sigmoid, tanh, GELU — what they do and why it matters.
Practical ML
3 topicsHandling Imbalanced Data
SMOTE, class weights, undersampling — real-world class imbalance.
Anomaly Detection
Isolation Forest, LOF, and one-class SVM for outlier detection.
Missing Data & Imputation
MCAR, MAR, MNAR — and how to handle each correctly.
Evaluation & Best Practices
4 topicsEvaluation Metrics
MSE, R², Accuracy, F1, AUC-ROC — decoded.
Cross-Validation
K-Fold, Stratified, Leave-One-Out — reliable model assessment.
Feature Engineering
Transform raw data into ML-ready signals.
Hyperparameter Tuning
Grid search, random search, Bayesian optimization with Optuna.