Installation and running shortest path notebook comments

[akazachk]

I encountered some issues (now resolved) when running the Shortest Path notebook on my laptop running MacOS Monterey 12.2.1 with Python 3.9.10 installed via Homebrew.

After installation, I went to the shortest path notebook.

• I had an issue with plt.style.use("science") --- SciencePlots is a dependency, but I did not see it listed, so I had to install that.
• I also recommend making the comments in the file a bit more detailed. Examples:
  • In the "Init Data" section, it may help to add more explanation of the parameters for genData in the cell comments, such as that costs are calculated using deg. Or to simply link to the documentation.
  • When calling optdataset, the current comments can be more informative (e.g., "get training data set, by computing optimal solution and optimal values for each sample cost vector in c_train", or whatever else is done there --- are the features used?).
  • After training the SPO+ loss, there are two plots, and both are commented as # draw plot. It took a moment to identify what the second plot is showing.
• I think the title + axes need to be relabeled in the second plot, if I understood correctly.
• When comparing solutions after training for the SPO+ loss, it may help to plot the values of the pred vs actual cost, and pred vs actual solution, to have a visual for what is now currently given as text. Maybe just for the first sample, or maybe there is a nice way to aggregate the data for all the samples.
• Just like the small example in the SPO paper (with 1-d features), it would be good to have a direct comparison of predictions with linear regression vs SPO+. Same for decisions.

A final comment: the documentation is great and helped a lot when I had questions! Thank you.

[LucasBoTang]

Hi Aleksandr! Thank you for your helpful suggestion!

• I removed the SciencePlots requirement
• I updated the new notebook
• The y-axis of plots was relabeled
• Plotting the values of the pred vs. actual cost/solution is a good idea. I will add this visualization later.
• For more experiments and figures in the paper, you can see the notebook in the develpment branch.

[LucasBoTang]

Hi Aleksandr! I upload ./notebooks/02 2D knapsack Visualization.ipynb, which try to visualize and compare true solution and predicted solution.

[LucasBoTang]

Hi Aleksandr! New notebooks are in. Hope it will be more clear and helpful :)

• 01 Optimization Model: Build up optimization solver
• 02 Optimization Dataset: Generate synthetic data and use optDataset
• 03 Training and Testing: Train and test end-to-end predict-then-optimize for different approaches
• 04 2D knapsack Solution Visualization: Visualize solutions for knapsack problem
• 05 Warcraft Shortest Path: Use the Warcraft terrains dateset to train shortest path