Semantic Pill 14

Semantic Pill 14



  • Map Reduce is a methodology to process large data sets via parallel and distributed tools and algorithms. Conceptually the idea is not new: most sorting techniques applied in conventional computing used similar procedures, especially when challenged with large data sets. It has two main procedures, namely: 1) map o mapping that by filtering, masking and sorting data sets are open in streams, and 2) reduce, synthesizing and summarizing those streams. MapReduce also refers to a similar methodology used by Google. Hadoop is one of the implementations of this idea. See also the mother idea-paradigm “divide and conquer algorithms”;


Below a MDP, Markovian Decision Process schema of an “entity”, either real or artificial is depicted, with three possible “states” S0, S1 and S2, and only two possible “actions”, a0 and a1 to state change no matter the state. In order that these sort of automatons represent “alive” entities should exist an associated probability P(a) [s, s’] that state change from s to s’ at time (t+1) by executing action (a). In order to evolve - or at least to have a reason of existence - we should associate to this automaton a “Reward” function R(a) [s, s’] when state changes from s to s’ due to action (a). These rewards are associated to learning. It also defines the 4-tuple [S A P R], State, Action, Probability, and Reward.





Mapreduce Google Rank Examples from



Fuente: MDP from Wikipedia



Additional information