Yet some more experiment of decentralized vs centralized search
2018-10-03
1 Introduction
The goal was to procude data that would allow to compare the centralized and decentralized algorithms performance when the number of item nodes or number of similarity nodes increases. The problem is that number of nodes and links in the graph are obviously related – so they are not independent variables. Furthermore, number of links is not a direct parameter of the simulation. Rather similarity
links get created during a simulation and depend on similarityConnectThreshold
parameters which regulates how similar item values should be in order to be connected by an explicit link, and the item values are initialized randomly. In short, number of links is an emergent variable, therefore the following run was meant to try out several parameter values (based on the knowledge from previous experiments) in order to come up with graphs and link numbers that are comparable.
No. | expermentId | Comment |
---|---|---|
1 | EXP09-05-02-15-an9jYp | agentNumbers=[400, 500, 600] : chainLengths=[10] : randomWorksNumberMultipliers=[1] : maxDistances=[10] : similaritySearchThresholds=[1] : similarityConnectThresholds=[0.9, 0.93, 0.96, 0.99] : message=[pre-generated smallWorld graph of agents with known diameters < 10] |
2 | EXP09-05-11-49-gcWcHs | Same as above, only agentNumbers=[700, 800] |
3 | EXP09-07-02-54-hATxwM | Same as above, agentNumbers=[700, 800], randomWorksNumberMultipliers=[2] |