4.5 Symbol Table Applications

Transcription

1 Set ADT 4.5 Symbol Table Applications Set ADT: unordered collection of distinct keys. Insert a key. Check if set contains a given key. Delete a key. SET interface. addkey) insert the key containskey) is given key present? removekey) remove the key iterator) return iterator over all keys Q. How to implement? Java library: java.util.hashset. Set Client: Remove Duplicates More Set Applications Remove duplicates. [e.g., from commercial mailing list] Read in a key. If key is not in set, insert and print it out. Application Purpose Key Spell checker Identify misspelled words Word Browser Highlight previously visited pages URL public class DeDup { public static void mainstring[] args) { SET<String> set = new SET<String>); while StdIn.isEmpty)) { String key = StdIn.readString); if set.containskey)) { set.addkey); System.out.printlnkey); Chess Detect repetition draw Board position Spam blacklist Prevent spam IP addresses of spammers 3 4

2 Vectors and Matrices Inverted Index Inverted index. Given a list of documents, preprocess so that you can quickly find all documents containing a given query. Ex 1. Book index. Ex. Web search engine index. Key. Query word. Value. Set of documents. 6 Inverted Index Implementation Inverted Index Implementation cont) public class InvertedIndex { public static void mainstring[] args) { ST<String, SET<String>> st; st = new ST<String, SET<String>>); use SET to avoid duplicates // create inverted index for String filename : args) { In in = new Infilename); while in.isempty)) { String word = in.readstring); if st.containsword)) st.putword, new SET<String>)); SET<String> set = st.getword); set.addfilename); // read queries from standard input In stdin = new In); while stdin.isempty)) { String query = in.readstring); if st.containsquery)) System.out.println"NOT FOUND"); else { SET<String> set = st.getquery); for String filename : set) System.out.printfilename + " "); System.out.println); 7 8

3 Inverted Index Extensions. Ignore stopwords: the, on, of, etc. Multi-word queries: set intersection AND) set union OR). Record position and number of occurrences of word in document. Sparse Vectors and Matrices 9 Vectors and Matrices Sparsity Vector. Ordered sequence of N real numbers. Matrix. N-by-N table of real numbers. Def. An N-by-N matrix is sparse if it has ON) nonzeros entries. Empirical fact. Large matrices that arise in practice are usually sparse. [ ], b = ["1 ] [ ] a = a + b = " a o b = 0 # "1) + 3 # ) + 15 # ) = 36 Matrix representations. D array: space proportional to N. Goal: space proportional to number of nonzeros, without sacrificing fast access to individual elements. A = # 0 1 1& 4 ", B = $ ' # 1 0 0& 0 0, A + B = $ 0 0 3' # 1 1 1& 6 " $ ' Ex. Google performs matrix-vector product with N = 4 billion # 0 1 1& 4 " ) $ ' #"1& $ ' = # 4& $ 36' 11 1

4 Sparse Vector Implementation Sparse Vector Implementation cont) public class SparseVector { private final int N; private ST<Integer, Double> st = new ST<Integer, Double>); public SparseVectorint N) { this.n = N; public void putint i, double value) { if value == 0.0) st.removei); else st.puti, value); public double getint i) { if st.containsi)) return st.geti); else return 0.0; a[i] = value a[i] // return a b public double dotsparsevector b) { SparseVector a = this; double sum = 0.0; for int i : a.st) if b.st.containsi)) sum += a.geti) * b.geti); return sum; // return c = a + b public SparseVector plussparsevector b) { SparseVector a = this; SparseVector c = new SparseVectorN); for int i : a.st) c.puti, a.geti)); for int i : b.st) c.puti, b.geti) + c.geti)); return c; Sparse Matrix Implementation Sparse Matrix Implementation cont) public class SparseMatrix { private final int N; private SparseVector[] rows; public SparseMatrixint N) { this.n = N; rows = new SparseVector[N]; for int i = 0; i < N; i++) rows[i] = new SparseVectorN); public void putint i, int j, double value) { rows[i].putj, value); public double getint i, int j) { return rows[i].getj); N-by-N matrix of all zeros A[i][j] = value A[i][j] // return b = A*x public SparseVector timessparsevector x) { SparseMatrix A = this; SparseVector b = new SparseVectorN); for int i = 0; i < N; i++) b.puti, rows[i].dotx)); return b; // return C = A + B public SparseMatrix plussparsematrix B) { SparseMatrix A = this; SparseMatrix C = new SparseMatrixN); for int i = 0; i < N; i++) C.rows[i] = A.rows[i].plusB.rows[i]); return C; 15 16

5 A Plan for Spam A Plan for Spam Bayesian spam filter. Filter based on analysis of previous messages. User trains the filter by classifying messages as spam or ham. Parse messages into tokens alphanumeric, dashes, ', $) Build data structures. Hash table A of tokens and frequencies for spam. Hash table B of tokens and frequencies for ham. Hash table C of tokens with probability p that they appear in spam. double h =.0 * ham.freqword); double s = 1.0 * spam.freqword); double p = s/spams) / h/hams + s/spams); bias probabilities to avoid false positives Reference: 18 A Plan for Spam Identify incoming as spam or ham. Find 15 most interesting tokens difference from 0.5). Combine probabilities using Bayes law. which data structure? p 1 " p " L " p 15 p1 " p " L" p15) + 1 p1) " 1 p ) " L" 1 p15) ) Declare as spam if threshold > 0.9. Details. Words you've never seen. Words that appear in ham corpus but not spam corpus, vice versa. Words that appear less than 5 times in spam and ham corpuses. Update data structures. 19