Run an experiment that simulates the same conditions as Dibene.

[hansyuan]

To do so, we need the following:

• double coverage
• location of the bases
• ambulance selection algorithm. best travel time vs best coverage? one that chooses between the two based on case severity?
• case severity?

[hansyuan]

@mcdeoliveira, cc: @thl024

Mauricio, I have a question about running the experiment based on the Dibene paper's scenarios.

To choose ambulances, we have best time and best coverage. What does the RCT do right now?

Sort of related -- I've been wanting to implement the tension between best time and best coverage based on the severity of the case (most severe -> best travel time, etc).

[hansyuan]

The use of Best Travel Time, Best Coverage, and Weighted Time vs Coverage for the experiment

[hansyuan]

k-means best base selection algorithm:

• made a minor adjustment where I allow each base to be the first chosen base in the set of chosen bases. Usually, the first chosen base is the one that has the most demand points < r1. This implementation means a thousand sets are calculated and then the best coverage is chosen.
• TODO: consider the secondary coverage in the decision!
• results: 92% to 93% single coverage for 8 bases. not a huge improvement here, but.

double coverage algorithm:

• implemented as a metric algorithm.
• still need to also implement this in the k-means base selection algorithm as part of the decision as to which set of bases we want to use for the simulator.

case severity was used as part of considering the demand points. unweighted vs weighted.

[hansyuan]

bases, primary coverage %, secondary coverage %
10, 93, 79
14, 99, 89 (I think?)
15, 100, 96

[hansyuan]

TODO, the two double coverage implementations are inconsistent

[hansyuan]

The secondary coverage SHOULD be implemented correctly, but the coverages (both primary and secondary) are inconsistent. I think it's because the kd-tree travel times is using the wrong locations.

[hansyuan]

1. When we generate random cases, also save them so that we can run the same cases under varying policies and algorithms.

2. Allow multiple simulations to run and save the output with different names so that we can analyze multiple different scenarios.

3. Implement a class that will choose between best time and least disruption based on the severity.

[hansyuan]

#25

Would be helpful as well

[hansyuan]

The original bases and ambulances are chosen. But there is an error in the coverage. Don't know why.

Going to look at the analysis and plotting the metrics first, though

[hansyuan]

Things left:

• case severity for a total of 3 dispatch policies
• (3 policies) x (2-3 case densities) to run multiple different cases