Chooser Page

Topics

Topic pages are the academy's smallest teaching unit. Use them to learn one workflow move, one inspection habit, or one failure pattern fast. Do not use this page like a shelf. Use it to choose the next page that will change your behavior.

Study Plan Examples Tracks Back To Home

Start Here

First-Time Learners

Start with tooling, then honest evaluation, then the first deep-learning mechanics. Do not open the broad survey pages first.

Use As Repair Map

IOAI Learners

Jump directly to the weakest layer: validation, optimization, reliability, review budgets, or representation choice.

Shared Learning Contract

Every topic page inherits the contract below unless it declares an override near the title. This keeps objectives, prerequisites, time, and evidence explicit without copying metadata across 81 files.

Topic family	Objective	Prerequisites	Typical time	Evidence to keep
Foundations	explain the core idea with shapes or tiny numbers, then connect it to training	First Steps; basic Python expressions	25–40 minutes	one worked calculation or runnable output plus a short explanation of the main failure pattern
Tooling and data	inspect or transform data without losing schema, alignment, or uncertainty	foundations contract; basic NumPy/Pandas syntax	30–45 minutes	the requested table, shape, or plot plus answers to at least three practice prompts
Classical ML	fit and compare the smallest admissible model under an honest split	tooling contract; Honest Splits and Baselines for model-selection pages	35–60 minutes	held-out metric or diagnostic, baseline comparison, and one written rejection or keep decision
Deep learning	run and diagnose one controlled training or representation experiment	Loss and Gradients Intuition, MLP Training Baseline, and PyTorch basics	45–75 minutes	curve or diagnostic artifact, selected checkpoint/strategy, and the first failure check performed
Modalities and decisions	choose a workflow lever for the modality or constraint and defend its boundary	honest-validation contract plus the relevant classical or deep-learning family	45–75 minutes	task-specific artifact, weak-slice or failure decomposition, decision, and evidence that would reverse it

For every family, “complete” means you can:

1. state when the move applies and when it does not
2. run the Runnable Example and name the first output worth inspecting
3. complete at least three Practice prompts without reopening the worked pattern
4. explain the page's Failure Pattern using evidence from the run
5. save a short decision note or calculation that another learner can check

Time estimates assume the environment is already installed and exclude optional long training runs. A topic should add a local override only when it needs extra prerequisites, an accelerator, more than 75 minutes, or a different evidence artifact.

Pick Topics By Need

If data inspection is weak, start with:

• Array Shapes and Axis Operations
• Table Inspection
• Grouped Summaries and Slice Checks
• Feature Matrix Construction

If evaluation discipline is weak, start with:

• Honest Splits and Baselines
• Leakage Patterns
• Cross-Validation
• Calibration and Thresholds

If deep-learning mechanics are weak, start with:

• PyTorch Training Loops
• Debugging Deep Learning
• Batch Normalization and Initialization
• Learning Rate Schedulers

If decision quality is weak, start with:

• Baseline-First Task Solving
• Experiments and Ablations
• Imbalanced Metrics and Review Budgets
• Reliability Slices

Default Beginner Sequence

Use this sequence if you want one clear route instead of browsing:

1. What Is a Model?
2. Probability and Linear Algebra Refresher
3. Array Shapes and Axis Operations
4. Table Inspection
5. Grouped Summaries and Slice Checks
6. Plotting for Model Debugging
7. Feature Matrix Construction
8. Honest Splits and Baselines
9. Leakage Patterns
10. Cross-Validation
11. Hyperparameter Tuning
12. Calibration and Thresholds

After that, move into Tracks.

Family Guide

The directory below lists all 81 topic lessons. Use it to verify coverage, then choose by need rather than reading top to bottom.

Foundations

Use this family to make the later code and notation readable:

• What Is a Model?
• Probability and Linear Algebra Refresher
• Loss and Gradients Intuition

Tooling And Data

Use this family to build inspection habits before modeling:

• Array Shapes and Axis Operations
• Table Inspection
• Grouped Summaries and Slice Checks
• Plotting for Model Debugging
• Feature Matrix Construction
• Data Cleaning and Preprocessing

Classical ML Workflow

Use this family when the core issue is split discipline, metric choice, or model comparison:

• Honest Splits and Baselines
• Leakage Patterns
• Cross-Validation
• Hyperparameter Tuning
• Calibration and Thresholds
• SVM Margins and Kernels
• Linear Regression
• Logistic Regression
• K-Nearest Neighbors
• Decision Trees
• Clustering and Low-Dimensional Views
• K-Means
• PCA
• Advanced Clustering and Dimensionality Reduction
• Evaluation Metrics Deep Dive
• Learning Curves and Bias-Variance
• Ensemble Methods
• Dimensionality Reduction
• Feature Selection
• Regression Metrics and Diagnostics
• Semi-Supervised Learning

Deep Learning Workflow

Use this family when loops, optimization, checkpoints, or representation reuse are the real bottleneck:

• PyTorch Training Loops
• MLP Training Baseline
• Backpropagation
• Activation Functions
• Optimizers and Regularization
• PyTorch Optimization Recipes
• Transfer and Fine-Tuning
• Batch Normalization and Initialization
• Learning Rate Schedulers
• Mixed Precision Training
• Convolutional Neural Networks
• Attention and Transformers
• Recurrent Networks and Sequences
• Data Augmentation
• Debugging Deep Learning
• Metric Learning and Retrieval
• Saliency and Explainability
• Self-Supervised and Representation Learning
• Autoencoders and VAEs
• Generative Adversarial Networks
• Diffusion Models
• PEFT and LoRA
• Mixture-of-Experts and Scaling
• Quantization, Distillation, and Deployment
• Graph Neural Networks
• Reinforcement Learning Foundations

Modalities And Decisions

Use this family when the main question is what to compare, what to trust, and what to do next:

• Baseline-First Task Solving
• Text Representations and Order
• Sequential Splits and Lag Features
• Experiments and Ablations
• Vision Augmentation and Shift Robustness
• Audio Windows and Spectral Features
• Vision and Text Encoders
• Detection and Segmentation
• Mock Tasks and Timed Workflows
• Audio Models
• Imbalanced Metrics and Review Budgets
• Selective Prediction and Review Budgets
• Reliability Slices
• Tabular Feature Engineering
• Multi-Modal Fusion
• Object Detection Basics
• Text Generation and Language Models
• Language Modeling Fundamentals
• Tokenization Mechanics
• Encoder-Decoder and Translation
• Density-Based Counting
• Steering Frozen Generative Models
• Black-Box LLM Optimization
• Retrieval-Augmented Generation
• Prompting and Tool Use

Before You Leave A Topic Page

Ask:

1. do I know when to use this move
2. do I know which artifact to inspect first
3. do I know the main trap
4. do I know which example or track comes next

If the answer is no, stop browsing and run the matching example before opening another topic.