UNIVERSAL DISTRIBUTED GRAPH PROCESSING METHOD AND SYSTEM BASED ON REINFORCEMENT LEARNING

摘要

Disclosed are a universal distributed graph processing method and system based on reinforcement learning. The method comprises: defining a distributed data processing center on the basis of a graph theory so as to form a distributed graph, and by means of a preset graph cutting model and a preset graph processing model, cutting the distributed graph in a reinforcement learning manner and on the basis of a preset constraint condition; allocating a learning automaton to each vertex; finding the most suitable data processing center for the vertex by means of training, wherein the probability of each vertex in all data processing centers obeys a certain probability distribution, and during each iteration process, the whole system comprises five steps of action selection, vertex migration, score calculation, reinforcement signal calculation and probability update; and determining that the iteration is ended when the maximum number of iterations is reached or the constraint condition is converged. A distributed graph processing model formed by means of the universal distributed graph processing method provided in the present application is a universal distributed graph model, and for different optimization objectives, only different score calculation solutions and different weight vectors need to be designed.