This repository contains the experimental data used in the paper “Boosting Binary Optimization via Binary Classification: A Case Study of Job Shop Scheduling” by O.V. Shylo and H. Shams. Here you can also find the code for generating the probability dominance plots.
The data is located in the folder data, while the code for generating the plots is in the jupyter notebook file generate-plots.ipynb. The file analyze-regression.ipynb provides the code for generating the accuracy plots for the logistic regression.
Below is the snapshot of the typical log file that we use to compare algorithms.
algo expid problem runid mpirank seed obj time epoch 0 gta gtacurve ta11.txt 10 0 1168868254 1395 6 0 1 gta gtacurve ta11.txt 10 0 1168868254 1394 9 1 2 gta gtacurve ta11.txt 10 0 1168868254 1393 10 1 3 gta gtacurve ta11.txt 10 0 1168868254 1392 12 1 4 gta gtacurve ta11.txt 10 0 1168868254 1391 12 1
The columns are described in the following table.
| column | description |
|---|---|
| algo | algorithm name |
| expid | name of the experiment |
| problem | problem name |
| runid | id of the run (there may be two distinct runs with the same id) |
| mpirank | not used |
| seed | random initialization seed (unique for each distinct run) |
| obj | objective value of the solution that was found |
| time | total seconds after the start, when the above objective was found |
| epoch | epoch number, when the above objective was found |
For the gta algorithm, the experiment id gtacurve corresponds to the runs that were limited to 200 epochs. The experiment id gtacurvetime corresponds to the runs that were limited to 30 minutes. The experiment gtacurve0001 corresponds to the last experiment described in the paper (changed theta-min to 0.0001).
For the stabu algorithm, the experiment id tabu0 corresponds to the runs that were limited to 200 epochs. The experiment id tabutime corresponds to the runs that were limited to 30 minutes.
All the performance log files are located in the data/performance folder. There is a subfolder for each algorithm, each experiment id and each problem instance.
For example, the folder data/performance/gta/gtacurve/ta11/, contains all the logs by the algorithm gta with experiment id gtacurve, when used on the problem ta11.
Below is the snapshot of the typical log file that we use to plot the optimal regression parameters.
algo expid problem runid ... time epoch muopt accuracy 0 stabu tabu0 ta48.txt 141 ... 9 1 0.009183 0.714767 1 stabu tabu0 ta48.txt 141 ... 19 2 0.012534 0.713215 2 stabu tabu0 ta48.txt 141 ... 29 3 0.020770 0.734721 3 stabu tabu0 ta48.txt 141 ... 39 4 0.025433 0.749809 4 stabu tabu0 ta48.txt 141 ... 48 5 0.033475 0.750561 [5 rows x 9 columns]
The columns are described in the following table.
| column | description |
|---|---|
| algo | algorithm name |
| expid | name of the experiment |
| problem | problem name |
| runid | id of the run (there may be two distinct runs with the same id) |
| seed | random initialization seed (unique for each distinct run) |
| time | total seconds after the start, when the optimal logreg parameter was calculated |
| epoch | epoch number, when the optimal logreg parameter was calculated |
| muopt | optimal logreg parameter found at the above time/epoch stamps) |
| accuracy | accuracy of the logistic regression at the above time/epoch |
All the regression log files are located in the data/onlineregression folder. There is a subfolder for each algorithm, each experiment id and each problem instance.
For example, the folder data/onlineregression/stabu/tabu0/ta11/, contains all the logistic regression logs by the algorithm stabu with experiment id tabu0, when used on the problem ta11.