Experimenting with algorithm that maximise the profit for Combinatorial Auction problem
An auxiliary report will be done based on the results of the following experiments:
We have the test datasets that can be found here, in total we hage 630 files with auction data.
A special dataset can be found here that is a small subset of the all the datasets that contains some small and large instances, it is mainly used for fast iterations.
Raw data for our experiments can be found here, informations about the experiments can be found in the above reports.
- experiment_01 is a result of running out LP model with no timeout
- 1_minute_time_out_full_run_binary_operation is result of running our unproved LP model with 1 minute timeout
- 5_minute_time_out_full_run_binary_operation is result of running our unproved LP model with 5 minute timeout
- GreedyAverageItemsPrice_full_run
- GreedyBigBet_full_run
- GreedyNumberOfItems_full_run
- aoc_firs_implementation_5sec
- 10_sec_groupe1_dataset_on_all_solutions_1_retry result of running all models one time over the groupe1 dataset with a 10 second timeout
- aoc_second_5_seconds result of running the improved ACO on all datasets
- 5_min_custom_datasets_all_retry_1 result of running on the custom dataset all solutions with a 5 minute timeout
- parial_results this a partial runs that were termianted
Every solution is based on this base_model.
We have a few solutions:
- LP model that can be found here
- A few greedy implementation that can be found here
- GreedyBigBet - is ordering bids by the amount they offer
- GreedyNumberOfItems - is ordering bids by the amount if items the bid contains
- GreedyAverageItemsPrice - is ordering bids by the average price/item
- Ant Colony Optimization that can be found here - this is the first solution, we can still improve it
- Ant Colony Optimization that can be found here that uses 1000 ants to encourage exploration
A utility tool to generate Markdown with tables based on a summary, can be found here.
Also a tool that compared results to one another can be found here, it is already liked to existing experiment results, you can run it and see how differing experiments compare with each other and offer static data on the results.
Team Members:
- Glodeanu Irina-Elena
- Micu Matei-Marius