Route Guide

IOAI Path

Use this route if you already code and want stronger competition discipline: honest validation, restrained iteration, operating-point choices, and decisions you can defend under time pressure.

Study Plan Decision Clinics Solved Questions Track Guide

Who It Is For

Competition-Minded Learners

Use this path when you can already run Python code and want better model judgment, not slower setup help.

First Goal

Stop Trusting The Easy Score

The early IOAI work is mostly evaluation discipline: fixed splits, strong baselines, leakage checks, and decisions that survive pressure.

Exit Rule

Repair One Weak Layer

This path works when you use it as a repair map. Pick the weakest layer first instead of rereading everything in order.

Use This Route If

• you already code comfortably in Python
• you can run a baseline without step-by-step setup help
• you care about leaderboard restraint, hidden evaluation, and policy decisions
• you want practice that feels closer to competition or real ML work

First 90 Minutes

1. read Study Plan
2. read Honest Splits and Baselines
3. read Leakage Patterns
4. run academy/.venv/bin/python academy/examples/validation-baseline-comparison/baseline_comparison.py
5. do Public/Private Restraint
6. write one stop-or-continue note before you move to a full track

Default Entry Sequence

If you do not know where to start, use this order:

1. run academy/.venv/bin/python academy/examples/validation-baseline-comparison/baseline_comparison.py
2. do Public/Private Restraint cold
3. open scikit-learn Validation and Tuning
4. run academy/.venv/bin/python academy/labs/sklearn-validation-and-tuning/src/validation_tuning_workflow.py
5. do Validation, Leakage, and Model Choice
6. move to Mock Tasks and Timed Workflows only after the validation story feels stable

Default Weekly Loop

Use one repeatable loop instead of random page hopping:

1. repair one weak topic
2. run one matching example
3. do one clinic before seeing the reveal
4. spend one longer session on a track
5. finish with one question pack or timed sheet

Pick The Next Move By Weakness

If validation discipline is weak:

• scikit-learn Validation and Tuning
• Validation, Leakage, and Model Choice

If your decisions under pressure are weak:

• Mock Tasks and Timed Workflows
• Decision Clinics

If operating points and queue policy are weak:

• Imbalanced Triage and Review Budgets
• Selective Prediction and Review Budgets

If representation or transfer choices are weak:

• ResNet, BERT, and Fine-Tuning
• Representation Reuse and Embedding Transfer

If you need a default judgment ladder instead of a diagnosis:

1. Decision Clinics
2. scikit-learn Validation and Tuning
3. Mock Tasks and Timed Workflows
4. Timed Checkpoint Sheets

A Good IOAI Session

A good session should leave behind:

• one explicit baseline
• one comparison that you actually trust
• one rejected tempting move
• one weak slice or hidden-risk note
• one short decision about what to do next

If you only collect scores, you are not getting the real value of this route.