hovinh / hdb-simulation Goto Github PK

Due to the current bottleneck (slow speed) in Integer Programming Solver, I can only run with a small number of agents and flats. Another idea is scaling down every project in sale launches within 1 years to 1 block with 10 flats each and then rerun the simulation.

Data: . I have provided number of apartments (flats) and number of blocks in 15 projects in 9 months. Your job is scale them to 15 blocks with 10 flats each and rerun the simulation in .

Steps:

• Modify actual_flats_per_block.txt file.
• Modify actual_location.txt file.
• Modify actual_ethnic.txt file.
• Run main.py with compare_random_utility(10) and compare_location_utility(10) uncommented.

Refer to > Time sheet, the time it takes to run a small scale experiment is quite big already. I need a way to optimize this bottleneck.

The implementation follows the tutorial .

Suggestion or PR should be an alternative way which uses PuLP package or any Python-supported package.

I have implemented 2 ways to twist the value of variance in Location utility model with ethnicity factor(defined in class):

• fixed it with a constant value.
• adapt to value of the mean, follows this equation variance = mean * (1-mean)/2. Please note the required range of value of these 2 parameters in the comment. In addition, the raw mean with no normalization always having value > 3.0.

The question is, how can one determine a good variance? Currently, I have no clear measurement or criterion in mind.