A question about Bayesian calculation formula
Most of the Bayesian algorithm that circulated on the network is mostly similar to this.
But the formulas and the author's examples have conflicts: the formula is as follows: A Event ---- mail is spam; T1, T2 ......tn represents the Token string P (A | Ti) Indicates in the message When the Token string Ti appears, the email is the probability of spam. Set P1 (Ti) = (Ti in HashTable_Good) P2 (Ti) = (TI) = (value in HashTable_BAD), P (A | Ti) = P1 (Ti) / [(P1 (Ti) P2 ( Ti)];
The author's example is as follows: For example: a spam A and a non-spam b in which "legal" words containing "leg" is generated by mail A, which records in the hash table F: 1 time L: 1 time G: 1 time Calculated in this table: The probability of the probability of 0.3 rounds is 0.3 The probability of genesis is 0.3 Generate HashTable_Good according to the mail B, the hash table The record is: method: 1 Law: 1 Calculated in this table: The probability of the probability of 0.5 law is 0.5 Comprehensive consideration of two hash table, there are four token strings: f L g When the "Act" appears in the law, the probability of the spam is: P = 0.3 / (0.3 0.5) = 0.375 "wheel": P = 0.3 / (0.3 0) = 1 When "Power": P = 0.3 / (0.3 0) = 1 P = 0 / (0 0.5) = 0 when "Law" appears
The above contradiction is:
P (a | ti) = p1 (ti) / [(p1 (ti) p2 (ti)];
P1 (Ti) of the upper black body is P1 (Ti) = (Ti is in Hashtable_Good)
And: p = 0.3 / (0.3 0.5) = 0.3750.3 is inde: P2 (Ti) = (TI value in HashTable_ Bad)
The article reprint is much, I don't know who the original author is! I don't know how to see this problem. Originally wanted to find English Bayes algorithm, it found that Bayesian applications are so broad, which may also be used. After all: By analyzing the past facts, how beautiful you can predict future things! Author: Tian Chunfeng in pairs Festival: February 22
