Quiz Settings and Questions
Configure scoring, answer each scenario, and submit your completed assessment.
Formula Used
Adjusted Points = Correct Answers − (Incorrect Answers × Penalty)
Adjusted Score = Adjusted Points ÷ Total Questions × 100
Calibration = Average confidence alignment × 100
The calculator first measures correct answers against all available questions. It then subtracts your selected penalty for each incorrect response. Unanswered questions receive no negative penalty.
Confidence calibration rewards justified certainty and sensible caution. Correct high-confidence answers receive stronger calibration credit. Incorrect high-confidence answers reduce calibration quality.
How to Use This Calculator
- Enter an optional participant name.
- Select practice or exam mode.
- Choose your wrong-answer penalty and pass mark.
- Answer each question and record confidence honestly.
- Submit the form to calculate detailed results.
- Review weak categories before attempting another session.
Your progress is stored inside the current browser. Clearing storage removes saved answers and settings. Printed reports exclude the answer form automatically.
Understanding Bayesian Optimisation
Bayesian optimisation searches expensive objective functions using selected experiments. It avoids exhaustive testing when evaluations consume money or time. Each observation improves a probabilistic model of the objective.
A surrogate model predicts performance across unexplored parameter combinations. Gaussian processes remain popular because they quantify uncertainty directly. Tree based surrogates often handle conditional or mixed parameter spaces.
The acquisition function converts predictions into the next testing decision. Expected Improvement balances promising means against uncertainty. Upper Confidence Bound adjusts exploration through a confidence parameter.
Acquisition, Noise, and Transformations
Probability of Improvement favors locations likely to exceed the incumbent. Its behavior depends strongly on the chosen improvement margin. Larger margins encourage broader searches beyond currently competitive regions.
Posterior uncertainty usually shrinks near previously evaluated configurations. Sparse regions retain wider uncertainty and stronger exploration potential. This mechanism guides experiments without requiring objective derivatives.
Noise complicates comparisons because repeated evaluations can produce outcomes. Systems model observation variance or average repeated trials. Robust acquisition rules prevent noisy winners from dominating selections.
Hyperparameter transformations can greatly influence surrogate quality. Log scaling suits learning rates and regularization strengths. Categorical encoding should preserve meaningful differences without imposing false distances.
Designing Efficient Experiments
Initial designs create the evidence needed for reliable modeling. Random sampling works, but space filling designs improve coverage. Latin hypercube sequences distribute early observations across dimensions.
Constraints separate feasible solutions from attractive but unusable candidates. Constrained acquisition multiplies utility by estimated feasibility probability. This approach prioritizes improvements that also satisfy operational requirements.
Batch optimisation proposes several experiments before receiving new outcomes. Naive selection may repeat nearly identical candidate configurations. Diversity penalties or fantasy observations improve parallel recommendation quality.
Multiobjective problems seek tradeoffs rather than one universal winner. Pareto fronts display configurations that remain nondominated across objectives. Hypervolume improvement measures progress toward better balanced solution sets.
Validation and Search Diagnostics
Stopping rules protect budgets when additional experiments offer little value. Common signals include tiny improvements and persistently weak acquisition values. Maximum evaluations provide a safeguard against endless searching.
Reliable validation requires separating optimiser decisions from final assessment data. Reusing test results can bias configuration choices and reported performance. Nested evaluation gives stronger evidence for model selection claims.
Diagnostics reveal whether the search process behaves sensibly. Plot observed outcomes, uncertainty, acquisition values, and incumbent progress. Sudden instability often indicates scaling, noise, or surrogate misspecification.
This calculator grades conceptual reasoning across advanced optimisation scenarios. Negative marking discourages unsupported guessing during exam mode. Category scores identify topics needing deeper study and practice.
Frequently Asked Questions
1. What does this quiz calculator measure?
It measures advanced knowledge across surrogate models, acquisition functions, constraints, noise, validation, batching, and multiobjective optimisation. It also reports adjusted scoring, category accuracy, completion time, and confidence calibration.
2. How does negative marking work?
The selected penalty is subtracted for every incorrect answer. Unanswered questions receive no penalty. The adjusted score cannot fall below zero.
3. What is confidence calibration?
Confidence calibration compares certainty with correctness. High confidence helps when answers are correct. It reduces calibration quality when answers are incorrect.
4. Which mode should beginners choose?
Practice mode is better for guided learning. It displays answer review after submission. Exam mode supports stricter assessment and optional explanations.
5. Are unanswered questions counted?
Yes. Unanswered questions reduce both raw and adjusted percentages because totals remain fixed. They do not trigger the wrong-answer penalty.
6. Is progress saved automatically?
Yes. Answers, confidence selections, and quiz settings are stored locally. The clear button removes saved progress from the current browser.
7. Why use a pass mark setting?
The pass mark lets teachers and learners choose a suitable standard. Higher thresholds support certification practice. Lower thresholds suit early diagnostic attempts.
8. Can the result be printed?
Yes. Use the print button inside the result panel. The question form is hidden automatically during printing.
9. How should weak categories be improved?
Review the lowest category scores first. Study related formulas and practical examples. Then retake the quiz with a stronger penalty and higher pass mark.