Warning

This page is located in archive.
Go to the latest version of this course pages.

The report and presentation should in the first place reasonably evaluate and compare implemented algorithms. The algorithms should be compared w.r.t. their efficiency, i.e., the ability to generate high-quality solutions as fast as possible in terms of the number of executed fitness function evaluations.

Although the material described in both your report and your presentation is the same, their focus differs. The report should contain more detailed description of the algorithms and their results. The presentation should cover only the main ideas behind the algorithms and the main results achieved with the algorithms.

General structure:

- Problem statement (brief)
- Description of the chosen representation
- Description of the fitness function (i.e., the common fitness function as well as your own one, if you used a different one in your algorithm)
- Description of the algorithm
- Operators
- Perturbation operators for local search
- Crossover and mutation for EA
- Special features of advanced EA, e.g. the way the local search is incorporated into EA in a memetic algorithm

- Parameter values used in the experiments
- Pseudocode???

- Reasonable comparison of algorithms:
- Report:
- Most importantly, compare your three algorithms (e.g., local search, evolutionary algorithm, specialized EA).
- You can include also results of your colleague, but do not forget to cite him!
- Tables and graphs with statistically evaluated experimental results.
- Graphs with the median of the best fitness function value as a function of the number of fitness function evaluations.
- Graphs with the median of the best common fitness value as a function of the number of fitness function evaluations.

- Presentation:
- Each student contributes with his/her best algorithm.
- Algorithms are compared similarly as in the report.

- Discussion of results and final conclusions

The time slot for each presentation will be approximately **10 minutes**. So be concise!

The group presentation should contain:

- Collective part: group of students solving the same problem prepares the introductory slides with description of the problem solved, and final slides presenting the achieved results and comparisons, and concluding slides.
- Individual parts: slides with a brief and clear description of the representation, operators and evaluation function used by algorithms of the individual members of the group.

Put all parts into a single presentation. We have to strictly stick with the schedule, so we need to minimize the overhead related to switching between presentations. Individual parts of the presentation should be presented by individual students. One of the students will present the collective parts.

When comparing individual algorithms, you should ensure that all the algorithms were run in similar conditions. For most of the tasks it is sufficient to compare the achieved results

- on the same instances of the task and
- after the same number of fitness function evaluations.

The most important feature of an optimization algorithm is the best solution achieved after certain number of fitness evaluations (or after a certain “time”), the so-called best-so-far (BSF) solution. (The average quality of solutions in the population is not that important.) To reduce the uncertainty of the results of stochastic algorithms, we shall perform several (tens) runs of the same algorithm on the same problem. This way we get for all numbers of evaluations (for all “time” instants) one BSF solution from each of N runs. Based on this data set, it is possible to compute for each “time” instant the statistics like median/mean, interquartile range/standard deviation/min-max range.

The results shall be presented in tables and/or graphs. If you present the quality of achieved solution, make sure the following is clear:

- after how many evaluations the quality was achieved,
- whether you present mean, median or another statistic,
- how many runs were performed.

In addition to the mean or median, present also a measure of uncertainty (variability). Use the following pairs:

- mean and standard deviation, or
- median and interquartile range, or
- median and min-max range.

Instead of a boring table, it is often better to use picture:

- About box plots (40 years of box plots)
- To present the differences in achieved solutions after certain number of evaluations, you can use
- function boxplot from MATLAB Statistics toolbox, or
- function notBoxPlot available at MATLAB File Exchange, or
- function boxplot from matplotlib package for Python, or
- function boxplot in R, etc.

- To present the differences in many “time” instants accross the whole optimization process, you can use
- function errorbar in MATLAB, or
- function errorbar from matplotlib package for Python, or
- function errorBarGraph in R, etc.

courses/a0m33eoa/en/semestral_tasks/guidelines.txt · Last modified: 2020/11/16 11:54 by xposik