Skip to content

Study Plan

Use this page in one of two ways:

  • if you are learning AI for the first time, follow the phases in order
  • if you are preparing for IOAI, use the phases as a gap map and start at the first one that still feels weak

For route-specific entry, use Beginner Path or IOAI Path.

Default Entry By Route

Use the route pages once, then use this page as the map:

  • beginner route: start at Phase 1, then move forward in order
  • IOAI route: start at Phase 2 if validation is weak, or Phase 6 if timed decision-making is the main weakness
  • mixed case: if you can compare models but still mishandle tables, repair Phase 1 first and then jump back to your weak phase

Before Phase 1

You should already be able to:

  • create and use a Python virtual environment
  • run one script from the terminal
  • read a short traceback
  • tell the difference between a topic page, an example, and a track

If that is not true yet, start with Getting Started.

First 90 Minutes

Do this before you worry about the full plan:

  1. read Getting Started
  2. choose Beginner Path or IOAI Path
  3. run one small example from Examples
  4. inspect one artifact carefully
  5. write one short note about the next move

If you are on the IOAI route, do not keep re-choosing the route after this page. Use the phases below as a repair map and enter at the weakest layer.

Phase 1: Tooling And Inspection

Goal: Make arrays, tables, grouped summaries, and plotting feel mechanical enough that the first model does not break your attention.

Start with:

Before moving on, you should be able to:

  • explain row versus column operations
  • inspect missing values and label balance quickly
  • build a simple feature matrix without losing row alignment
  • make one plot that changes what you inspect next

Phase 2: Honest Validation

Goal: Make split discipline, baselines, cross-validation, tuning, calibration, and leakage checks feel non-negotiable.

Continue with:

Before moving on, you should be able to:

  • keep train, validation, and test roles separate
  • compare a learned model to a simple baseline honestly
  • explain when cross-validation helps and when it does not
  • describe one leakage pattern that could fake progress

Phase 3: Classical Model Choice

Goal: Add a few classical patterns that sharpen geometry, clustering, and representation judgment without losing evaluation discipline.

Continue with:

Before moving on, you should be able to:

  • justify linear versus nonlinear boundaries
  • explain what a low-dimensional view can and cannot prove
  • say which clustering result is stable enough to trust

Phase 4: Deep Learning Mechanics

Goal: Make training loops, checkpoints, optimizer choices, and overfitting checks feel controlled instead of mysterious.

Continue with:

Before moving on, you should be able to:

  • describe the difference between training and evaluation mode
  • explain what the best checkpoint means in your workflow
  • name one intervention for overfitting and one for unstable optimization

Phase 5: Transfer And Representation Reuse

Goal: Learn when to freeze, probe, fine-tune, or switch representation families instead of retraining everything from scratch.

Continue with:

Before moving on, you should be able to:

  • justify frozen features versus fine-tuning
  • compare at least two representation choices on one task
  • say what evidence would make you stop adding model complexity

Phase 6: Decisions Under Constraint

Goal: Practice model choice, operating points, and workflow discipline under time, budget, or leaderboard pressure.

Continue with:

Before moving on, you should be able to:

  • defend a stop-or-continue call
  • choose a threshold or review budget policy explicitly
  • explain why a public gain or small validation jump may still be untrustworthy

Suggested Weekly Rhythm

For steady progress, one week should usually include:

  1. one topic page
  2. one example run
  3. one track session
  4. one clinic or question pack
  5. one short written note about the weakest point in the workflow

That rhythm is better than reading many pages without producing evidence.