Scalable Mining of Large Diskbased Graph Databases


 Morris Copeland
 2 years ago
 Views:
Transcription
1 Sclle Mining of Lrge Disksed Grph Dtses Chen Wng Wei Wng Jin Pei Yongti Zhu Bile Shi Fudn University, Chin, {chenwng, weiwng1, 2465, Stte University of New York t Bufflo, USA & Simon Frser University, Cnd, ABSTRACT Mining frequent structurl ptterns from grph dtses is n interesting prolem with rod pplictions. Most of the previous studies focus on pruning unfruitful serch suspces effectively, ut few of them ddress the mining on lrge, disksed dtses. As mny grph dtses in pplictions cnnot e held into min memory, sclle mining of lrge, disksed grph dtses remins chllenging prolem. In this pper, we develop n effective inde structure, ADI (for djcency inde), to support mining vrious grph ptterns over lrge dtses tht cnnot e held into min memory. The inde is simple nd efficient to uild. Moreover, the new inde structure cn e esily dopted in vrious eisting grph pttern mining lgorithms. As n emple, we dpt the wellknown gspn lgorithm y using the ADI structure. The eperimentl results show tht the new inde structure enles the sclle grph pttern mining over lrge dtses. In one set of the eperiments, the new disksed method cn mine grph dtses with one million grphs, while the originl gspn lgorithm cn only hndle dtses of up to 3 thousnd grphs. Moreover, our new method is fster thn gspn when oth cn run in min memory. Ctegories nd Suject Descriptors: H.2.8 [Dtse Applictions]: Dt Mining Generl Terms: Algorithms, Performnces. Keywords: Grph mining, inde, grph dtse, frequent grph pttern. 1. INTRODUCTION Mining frequent grph ptterns is n interesting reserch prolem with rod pplictions, including mining struc This reserch is supported in prt y NSF grnt IIS 381 nd Ntionl Nturl Science Foundtion of Chin (No. 6338). All opinions, findings, conclusions nd recommendtions in this pper re those of the uthors nd do not necessrily reflect the views of the funding gencies. Permission to mke digitl or hrd copies of ll or prt of this work for personl or clssroom use is grnted without fee provided tht copies re not mde or distriuted for profit or commercil dvntge nd tht copies er this notice nd the full cittion on the first pge. To copy otherwise, to repulish, to post on servers or to redistriute to lists, requires prior specific permission nd/or fee. KDD 4, August 22 25, 24, Settle, Wshington, USA. Copyright 24 ACM /4/8...$5.. turl ptterns from chemicl compound dtses, pln dtses, XML documents, we logs, cittion networks, nd so forth. Severl efficient lgorithms hve een proposed in the previous studies [2, 5, 6, 8, 11, 9], rnging from mining grph ptterns, with nd without constrints, to mining closed grph ptterns. Most of the eisting methods ssume implicitly or eplicitly tht the dtses re not very lrge, nd the grphs in the dtse re reltively simple. Tht is, either the dtses or the mjor prt of them cn fit into min memory, nd the numer of possile lels in the grphs [6] is smll. For emple, [11] reports the performnce of gspn, n efficient frequent grph pttern mining lgorithm, on dt sets of size up to 32 KB, using computer with 448 MB min memory. Clerly, the grph dtse nd the projected dtses cn e esily ccommodted into min memory. Under the lrge min memory ssumption, the computtion is CPUounded insted of I/Oounded. Then, the lgorithms focus on effective heuristics to prune the serch spce. Few of them ddress the concern of hndling lrge grph dtses tht cnnot e held in min memory. While the previous studies hve mde ecellent progress in mining grph dtses of moderte size, mining lrge, disksed grph dtses remins chllenging prolem. When mining grph dtse tht cnnot fit into min memory, the lgorithms hve to scn the dtse nd nvigte the grphs repetedly. The computtion ecomes I/Oounded. For emple, we otin the eecutle of gspn from the uthors nd test its sclility. In one of our eperiments 1, we increse the numer of grphs in the dtse to test the sclility of gspn on the dtse size. gspn cn only hndle up to 3 thousnd grphs. In nother eperiment, we increse the numer of possile lels in grphs. We oserve tht the runtime of gspn increses eponentilly. It finishes dt set of 3 thousnd grphs with 636 seconds when there re only 1 possile lels, ut needs 15 hours for dt set with the sme size ut the numer of possile lels is 45! This result is consistent with the results reported in [11]. Are there ny rellife pplictions tht need to mine lrge grph dtses? The nswer is yes. For emple, in dt integrtion of XML documents or mining semntic we, it is often required to find the common sustructures from huge collection of XML documents. It is esy to see pplictions with collections of millions of XML documents. There re 1 Detils will e provided in Section 6
2 hundreds of even thousnds of different lels. As nother emple, chemicl structures cn e modeled s grphs. A chemicl dtse for drug development cn contin millions of different chemicl structures, nd the numer of different lels in the grphs cn esily go to up to 1. These lrge dtses re disksed nd often cnnot e held into min memory. Why is mining lrge disksed grph dtses so chllenging? In most of the previous studies, the mjor dt structures re designed for eing held in min memory. For emple, the djcencylist or djcencymtri representtions re often used to represent grphs. Moreover, most of the previous methods re sed on efficient rndom ccesses to elements (e.g., edges nd their djcent edges) in grphs. However, if the djcencylist or djcencymtri representtions cnnot e held in min memory, the rndom ccesses to them ecome very epensive. For disksed dt, without ny inde, rndom ccesses cn e etremely costly. Cn we mke mining lrge, disksed grph dtses fesile nd sclle? This is the motivtion of our study. Since the ottleneck is the rndom ccesses to the lrge disksed grph dtses, nturl ide is to inde the grph dtses properly. Designing effective nd efficient inde structures is one of the most invlule eercises in dtse reserch. A good inde structure cn support generl ctegory of dt ccess opertions. Prticulrly, good inde should e efficient nd sclle in construction nd mintennce, nd fst for dt ccess. Insted of inventing new lgorithms to mine lrge, disksed grph ptterns, cn we devise n efficient inde structure for grph dtses so tht mining vrious grph ptterns cn e conducted sclly? Moreover, the inde structure should e esy to e dopted in vrious eisting methods with minor dpttions. Stimulted y the ove thinking, in this pper, we study the prolem of efficient inde for sclle mining of lrge, disksed grph dtses, nd mke the following contriutions. By nlyzing the frequent grph pttern mining prolem nd the typicl grph pttern mining lgorithms (tking gspn s n emple), we identify severl ottleneck dt ccess opertions in mining lrge, disksed grph dtses. We propose ADI (for djcency inde), n effective inde structure for grphs. We show tht the mjor opertions in grph mining cn e fcilitted efficiently y n ADI structure. The construction lgorithm of ADI structure is presented. We dpt the gspn lgorithm y using the ADI structure on mining lrge, disksed grph dtses, nd chieve lgorithm. We show tht outperforms gspn in mining comple grph dtses nd cn mine much lrger dtses thn gspn. A systemtic performnce study is reported to verify our design. The results show tht our new inde structure nd lgorithm re sclle on lrge dt sets. The reminder of the pper is orgnized s follows. We define the prolem of frequent grph pttern mining in Section 2. The ide of minimum DFS code nd lgorithm gspn z y z z v2 v v1 y v3 z v3 v2 v1 y v () Grph () Sugrph (c) DFStree (d) DFStree G G T 1 T 2 Figure 1: Sugrph nd DFS codes re reviewed in Section 3, nd the mjor dt ccess opertions in grph mining re lso identified. The ADI structure is developed in Section 4. The efficient lgorithm for mining lrge, disksed grph dtses using ADI is presented in Section 5. The eperimentl results re reported in Section 6. The relted work is discussed in Section 7. Section 8 concludes the pper. 2. PROBLEM DEFINITION In this pper, we focus on undirected leled simple grphs. A leled grph is 4tuple G = (V, E, L, l), where V is set of vertices, E V V is set of edges, L is set of lels, nd l : V E L is leling function tht ssigns lel to n edge or verte. We denote the verte set nd the edge set of grph G y V (G) nd E(G), respectively. A grph G is clled connected if for ny vertices u, v V (G), there eist vertices w 1,..., w n V (G) such tht {(u, w 1), (w 1, w 2),..., (w n 1, w n), (w n, v)} E(G). Frequent ptterns in grphs re defined sed on sugrph isomorphism. Definition 1 (Sugrph isomorphism). Given grphs G = (V, E, L, l) nd G = (V, E, L, l ). An injective function f : V V is clled sugrph isomorphism from G to G if (1) for ny verte u V, f(u) V nd l (u) = l(f(u)); nd (2) for ny edge (u, v) E, (f(u), f(v)) E nd l (u, v) = l(f(u), f(v)). If there eists sugrph isomorphism from G to G, then G is clled sugrph of G nd G is clled supergrph of G, denoted s G G. For emple, the grph G in Figure 1() is sugrph of G in Figure 1(). A grph dtse is set of tuples (gid, G), where gid is grph identity nd G is grph. Given grph dtse GDB, the support of grph G in GDB, denoted s sup(g ) for short, is the numer of grphs in the dtse tht re supergrphs of G, i.e., {(gid, G) GDB G G}. For support threshold min sup ( min sup GDB ), grph G is clled frequent grph pttern if sup(g ) min sup. In mny pplictions, users re only interested in the frequent recurring components of grphs. Thus, we put constrint on the grph ptterns: we only find the frequent grph ptterns tht re connected. Prolem definition. Given grph dtse GDB nd support threshold min sup. The prolem of mining frequent connected grph ptterns is to find the complete set of connected grphs tht re frequent in GDB.
3 3. MINIMUM DFS CODE AND GSPAN In [11], Yn nd Hn developed the leicogrphic ordering technique to fcilitte the grph pttern mining. They lso propose n efficient lgorithm, gspn, one of the most efficient grph pttern mining lgorithms so fr. In this section, we review the essentil ides of gspn, nd point out the ottlenecks in the grph pttern mining from lrge disksed dtses. 3.1 Minimum DFS Code In order to enumerte ll frequent grph ptterns efficiently, we wnt to identify liner order on representtion of ll grph ptterns such tht if two grphs re in identicl representtion, then they re isomorphic. Moreover, ll the (possile) grph ptterns cn e enumerted in the order without ny redundncy. The depthfirst serch tree (DFStree for short) [3] is populrly used for nvigting connected grphs. Thus, it is nturl to encode the edges nd vertices in grph sed on its DFStree. All the vertices in G cn e encoded in the preorder of T. However, the DFStree is generlly not unique for grph. Tht is, there cn e multiple DFStrees corresponding to given grph. For emple, Figures 1(c) nd 1(d) show two DFStrees of the grph G in Figure 1(). The thick edges in Figures 1(c) nd 1(d) re those in the DFStrees, nd re clled forwrd edges, while the thin edges re those not in the DFStrees, nd re clled ckwrd edges. The vertices in the grph re encoded v to v 3 ccording to the preorder of the corresponding DFStrees. To solve the uniqueness prolem, minimum DFS code nottion is proposed in [11]. For ny connected grph G, let T e DFStree of G. Then, n edge is lwys listed s (v i, v j ) such tht i < j. A liner order on the edges in G cn e defined s follows. Given edges e = (v i, v j) nd e = (v i, v j ). e e if (1) when oth e nd e re forwrd edges (i.e., in DFStree T ), j < j or (i > i j = j ); (2) when oth e nd e re ckwrd edges (i.e., edges not in DFStree T ), i < i or (i = i j < j ); (3) when e is forwrd edge nd e is ckwrd edge, j i ; or (4) when e is ckwrd edge nd e is forwrd edge, i < j. For grph G nd DFStree T, list of ll edges in E(G) in order is clled the DFS code of G with respect to T, denoted s code(g, T ). For emple, the DFS code with respect to the DFStree T 1 in Figure 1(c) is code(g, T 1) = (v, v 1,,, )(v 1, v 2,,, z)(v 2, v, z,, )(v 1, v 3,,, y), where n edge (v i, v j ) is written s (v i, v j, l(v i ), l(v i, v j ), l(v j )), i.e., the lels re included. Similrly, the DFS code with respect to the DFStree T 2 in Figure 1(d) is code(g, T 2) = (v, v 1, y,, )(v 1, v 2,,, )(v 2, v 3,,, z) (v 3, v 1, z,, ). Suppose there is liner order over the lel set L. Then, for DFStrees T 1 nd T 2 on the sme grph G, their DFS codes cn e compred leiclly ccording to the lels of the edges. For emple, we hve code(g, T 1) < code(g, T 2) in Figures 1(c) nd 1(d). The leiclly minimum DFS code is selected s the representtion of the grph, denoted s min(g). In our emple in Figure 1, min(g) = code(g, T 1). Minimum DFS code hs nice property: two grphs G nd G re isomorphic if nd only if min(g) = min(g ). Moreover, with the minimum DFS code of grphs, the pro Input: DFS code s, grph dtse GDB nd min sup Output: the frequent grph ptterns Method: if s is not minimum DFS code then return; output s s pttern if s is frequent in GDB; let C = ; scn GDB once, find every edge e such tht e cn e conctented to s to form DFS code s e nd s e is frequent; C = C {s e}; sort the DFS codes in C in leicogrphic order; for ech s e C in leicogrphic order do cll gspn(s e, GDB, min sup); return; Figure 2: Algorithm gspn. lem of mining frequent grph ptterns is reduced to mining frequent minimum DFS codes, which re sequences, with some constrints tht preserve the connectivity of the grph ptterns. 3.2 Algorithm gspn Bsed on the minimum DFS codes of grphs, depthfirst serch, ptterngrowth lgorithm, gspn, is developed in [11], s shown in Figure 2. The centrl ide is to conduct depthfirst serch of minimum DFS codes of possile grph ptterns, nd otin longer DFS codes of lrger grph ptterns y ttching new edges to the end of the minimum DFS code of the eisting grph pttern. The ntimonotonicity of frequent grph ptterns, i.e., ny super pttern of n infrequent grph pttern cnnot e frequent, is used to prune. Compring to the previous methods on grph pttern mining, gspn is efficient, since gspn employs the smrt ide of minimum DFS codes of grph ptterns tht fcilittes the isomorphism test nd pttern enumertion. Moreover, gspn inherits the depthfirst serch, ptterngrowth methodology to void ny cndidtegenertionndtest. As reported in [11], the dvntges of gspn re verified y the eperimentl results on oth rel dt sets nd synthetic dt sets. 3.3 Bottlenecks in Mining Disksed Grph Dtses Algorithm gspn is efficient when the dtse cn e held into min memory. For emple, in [11], gspn is sclle for dtses of size up to 32 KB using computer with 448 MB min memory. However, it my encounter difficulties when mining lrge dtses. The mjor overhed is tht gspn hs to rndomly ccess elements (e.g., edges nd vertices) in the grph dtse s well s the projections of the grph dtse mny times. For dtses tht cnnot e held into min memory, the mining ecomes I/O ounded nd thus is costly. Rndom ccesses to elements in grph dtses nd checking the isomorphism re not unique to gspn. Insted, such opertions re etensive in mny grph pttern mining lgorithms, such s FSG [6] (nother efficient frequent grph pttern mining lgorithm) nd CloseGrph [9] (n efficient lgorithm for mining frequent closed grph ptterns). In mining frequent grph ptterns, the mjor dt ccess opertions re s follows.
4 OP1: Edge support checking. Find the support of n edge (l u, l e, l v ), where l u nd l v re the lels of vertices nd l e is the lel of the edge, respectively; OP2: Edgehost grph checking. For n edge e = (l u, l e, l v ), find the grphs in the dtse where e ppers; OP3: Adjcent edge checking. For n edge e = (l u, l e, l v), find the djcent edges of e in the grphs where e ppers, so tht the djcent edges cn e used to epnd the current grph pttern to lrger ones. Ech of the ove opertions my hppen mny times during the mining of frequent grph ptterns. Without n pproprite inde, ech of the ove opertions my hve to scn the grph dtse or its projections. If the dtse nd its projections cnnot fit into min memory, the scnning nd checking cn e very costly. Cn we devise n inde structure so tht the relted informtion cn e kept nd ll the ove opertions cn e chieved using the inde only, nd thus without scnning the grph dtse nd checking the grphs? This motivtes the design of the ADI structure. 4. THE ADI STRUCTURE In this section we will devise n effective dt structure, ADI (for djcency inde), to fcilitte the sclle mining of frequent grph ptterns from disksed grph dtses. 4.1 Dt Structure The ADI inde structure is threelevel inde for edges, grphids nd djcency informtion. An emple is shown in Figure 3, where two grphs, G 1 nd G 2, re indeed Edge Tle There cn e mny edges in grph dtse. The edges re often retrieved y the lels during the grph pttern mining, such s in the opertions identified in Section 3.3. Therefore, the edges re indeed y their lels in the ADI structure. In ADI, n edge e = (u, v) is recorded s tuple (l(u), l(u, v), l(v)) in the edge tle, nd is indeed y the lels of the vertices, i.e., l(u) nd l(v), nd the lel of the edge itself, i.e., l(u, v). Ech edge ppers only once in the edge tle, no mtter how mny times it ppers in the grphs. For emple, in Figure 3, edge (A, d, C) ppers once in grph G 1 nd twice in grph G 2. However, there is only one entry for the edge in the edge tle in the ADI structure. All edges in the edge tle in the ADI structure re sorted. When the edge tle is stored on disk, B+tree is uilt on the edges. When prt of the edge tle is loded into min memory, it is orgnized s sorted list. Thus, inry serch cn e conducted Linked Lists of Grphids For ech edge e, the identities of the grphs tht contin e form linked list of grphids. Grphid G i is in the list of edge e if nd only if there eists t lest one instnce of e in G i. For emple, in Figure 3, oth G 1 nd G 2 pper in the list of edge (A, d, C), since the edge ppers in G 1 once nd in G 2 twice. Plese note tht the identity of grph G i B 2 Edges (A,, B) (A, d, C) (B,, D) (B, c, C) (B, d, D) (C, d, D) 1 A 3 D d 4 C d G1 G2 Grph ids (on disk) Adjcency (on disk) G1 G2 G1 G2 G1 G2 G2 G1 d 4 C c 5 B A 1 d C B 2 6 d 3 D Block Block Figure 3: An ADI structure. ppers in the linked list of edge e only once if e ppers in G i, no mtter how mny times edge e ppers in G i. A list of grphids of n edge re stored together. Therefore, given n edge, it is efficient to retrieve ll the identities of grphs tht contin the edge. Every entry in the edge tle is linked to its grphid linked list. By this linkge, the opertion OP2: edgehost grph checking cn e conducted efficiently. Moreover, to fcilitte opertion OP1: edge support checking, the length of the grphid linked list, i.e., the support of n edge, is registered in the edge tle Adjcency Informtion The edges in grph re stored s list of the edges encoded. Adjcent edges re linked together y the common vertices, s shown in Figure 3. For emple, in lock 1, ll the vertices hving the sme lel (e.g., 1) re linked together s list. Since ech edge hs two vertices, only two pointers re needed for ech edge. Moreover, ll the edges in grph re physiclly stored in one lock on disk (or on consecutive locks if more spce is needed), so tht the informtion out grph cn e retrieved y reding one or severl consecutive locks from disk. Often, when the grph is not lrge, diskpge (e.g., of size 4k) cn hold more thn one grph. Encoded edges recording the djcency informtion re linked to the grphids tht re further ssocited with the edges in the edge tle. 4.2 Spce Requirement The storge of n ADI structure is fleile. If the grph dtse is smll, then the whole inde cn e held into min memory. On the other hnd, if the grph dtse is lrge nd thus the ADI structure cnnot fit into min
5 memory, some levels cn e stored on disk. The level of djcency informtion is the most detiled nd cn e put on disk. If the min memory is too smll to hold the grphid linked lists, they cn lso e ccommodted on disk. In the etreme cse, even the edge tle cn e held on disk nd B+tree or hsh inde cn e uilt on the edge tle. Theorem 1 (Spce compleity). For grph dtse GDB = {G 1,..., G n }, the spce compleity is O( n i=1 E(Gi) ). Proof. The spce compleity is determined y the following fcts. (1) The numer of tuples in the edge tle is equl to the numer of distinct edges in the grph dtse, which is ounded y n i=1 E(G i) ; (2) The numer of entries in the grphid linked lists in the worst cse is the numer of edges in the grph dtse, i.e., n i=1 E(Gi) gin; nd (3) The djcency informtion prt records every edge ectly once. Plese note tht, in mny ppliction, it is resonle to ssume tht the edge tle cn e held into min memory. For emple, suppose we hve 1, distinct verte lels nd 1, distinct edge lels. There cn e up to = different edges, i.e., ll possile comintions of verte nd edge lels. Suppose up to 1% edges re frequent, there re only less thn 5 million different edges, nd thus the edge tle cn e esily held into min memory. In rel pplictions, the grphs re often sprse, tht is, not ll possile comintions of verte nd edge lels pper in the grphs s n edge. Moreover, users re often interested in only those frequent edges. Tht shrinks the edge tle sustntilly. 4.3 Serch Using ADI Now, let us emine how the ADI structure cn fcilitte the mjor dt ccess opertions in grph pttern mining tht re identified in Section 3.3. OP1: Edge support checking Once n ADI structure is constructed, this informtion is registered on the edge tle for every edge. We only need to serch the edge tle, which is either indeed (when the tle is on disk) or cn e serched using inry serch (when the tle is in min memory). In some cses, we my need to count the support of n edge in suset of grphs G G. Then, the linked list of the grphids of the edge is serched. There is no need to touch ny record in the djcency informtion prt. Tht is, we do not need to serch ny detil out the edges. Moreover, for counting supports of edges in projected dtses, we cn mintin the support of ech edge in the current projected dtse nd thus we do not even serch the grphid linked lists. OP2: Edgehost grph checking We only need to serch the edge tle for the specific edge nd follow the link from the edge to the list of grphids. There is no need to serch ny detil from the prt of djcency informtion. OP3: Adjcent edge checking Agin, we strt from n entry in the edge tle nd follow the links to find the list of grphs where the edge ppers. Then, only Input: grph dtse GDB nd min sup Output: the ADI structure Method: scn GDB once, find the frequent edges; initilize the edge tle for frequent edges; for ech grph do remove infrequent edges; compute the mininmum DFS code [11]; use the DFStree to encode the vertices; store the edges in the grph onto disk nd form the djcency informtion; for ech edge do insert the grphid to the grphid list ssocited with the edge; link the grphid to the relted djcency informtion; end for end for Figure 4: Algorithm of ADI construction. the locks contining the detils of the instnces of the edge re visited, nd there is no need to scn the whole dtse. The verge I/O compleity is O(log n + m + l), where n is the numer of distinct edges in the grph, m is the verge numer of grphids in the linked lists of edges, nd l is the verge numer of locks occupied y grph. In mny pplictions, m is orders of mgnitudes smller thn the n, nd l is very smll numer (e.g., 1 or 2). The lgorithms for the ove opertions re simple. Limited y spce, we omit the detils here. As cn e seen, once the ADI structure is constructed, there is no need to scn the dtse for ny of the ove opertions. Tht is, the ADI structure cn support the rndom ccesses nd the mining efficiently. 4.4 Construction of ADI Given grph dtse, the corresponding ADI structure is esy to construct y scnning the dtse only twice. In the first scn, the frequent edges re identified. According to the priori property of frequent grph ptterns, only those frequent edges cn pper in frequent grph ptterns nd thus should e indeed in the ADI structure. After the first scn, the edge tle of frequent edges is initilized. In the second scn, grphs in the dtse re red nd processed one y one. For ech grph, the vertices re encoded ccording to the DFStree in the minimum DFS code, s descried in [11] nd Section 3. Only the vertices involved in some frequent edges should e encoded. Then, for ech frequent edge, the grphid is inserted into the corresponding linked list, nd the djcency informtion is stored. The sketch of the lgorithm is shown in Figure 4. Cost Anlysis There re two mjor costs in the ADI construction: writing the djcency informtion nd updting the linked lists of grphids. Since ll edges in grph will reside on disk pge or severl consecutive disk pges, the writing of djcency informtion is sequentil. Thus, the cost of writing djcency informtion is comprle to tht of mking
6 2 B 1 A 3 D d 4 C d 1 A 2 B 3 D 4 C d 3 d 4 d 4 d 1 () The grph nd the djcencylists 1 A 2 B 3 D 4 C 1 A d 2 B 3 D d 4 C d d () The djcencymtri Figure 5: The djcencylist nd djcencymtri representtions of grphs. copy of the originl dtse plus some ookkeeping. Updting the linked lists of grphids requires rndom ccesses to the edge tle nd the linked lists. In mny cses, the edge tle cn e held into min memory, ut not the linked list. Therefore, it is importnt to cche the linked lists of grphids in uffer. The linked lists cn e cched ccording to the frequency of the corresponding edges. Constructing ADI for lrge, disksed grph dtse my not e chep. However, the ADI structure cn e uilt once nd used y the mining mny times. Tht is, we cn uild n ADI structure using very low support threshold, or even set min sup = 1. 2 The inde is stored on disk. Then, the mining in the future cn use the inde directly, s long s the support threshold is no less thn the one tht is used in the ADI structure construction. 4.5 Projected Dtses Using ADI Mny depthfirst serch, ptterngrowth lgorithms utilize proper projected dtses. During the depthfirst serch in grph pttern mining, the grphs contining the current grph pttern P should e collected nd form the P  projected dtse. Then, the further serch of lrger grph ptterns hving P s the prefi of their minimum DFS codes cn e chieved y serching only the P projected dtse. Interestingly, the projected dtses cn e constructed using ADI structures. A projected dtse cn e stored in the form of n ADI structure. In fct, only the edge tle nd the list of grphids should e constructed for new projected dtse nd the djcency informtion residing on disk cn e shred y ll projected dtses. Tht cn sve lot of time nd spce when mining lrge grph dtses tht contin mny grph ptterns, where mny projected dtses my hve to e constructed. 4.6 Why Is ADI Good for Lrge Dtses? In most of the previous methods for grph pttern mining, the djcencylist or djcencymtri representtions re used to represent grphs. Ech grph is represented y n djcencymtri or set of djcencylists. An emple is shown in Figure 5. 2 If min sup = 1, then the ADI structure cn e constructed y scnning the grph dtse only once. We do not need to find frequent edges, since every edge ppering in the grph dtse is frequent. In Figure 5(), the djcencylists hve 8 nodes nd 8 pointers. It stores the sme informtion s Block 1 in Figure 3, where the lock hs 4 nodes nd 12 pointers. The spce requirements of djcencylists nd ADI structure re comprle. From the figure, we cn see tht ech edge in grph hs to e stored twice: one instnce for ech verte. (If we wnt to remove this redundncy, the trdeoff is the sustntil increse of cost in finding djcency informtion). In generl, for grph of n edges, the djcencylist representtion needs 2n nodes nd 2n pointers. An ADI structure stores ech edge once, nd use the linkge mong the edges from the sme verte to record the djcency informtion. In generl, for grph of n edges, it needs n nodes nd 3n pointers. Then, wht is the dvntge of ADI structure ginst djcencylist representtion? The key dvntge is tht the ADI structure etrcts the informtion out continments of edges in grphs in the first two levels (i.e., the edge tle nd the linked list of grphids). Therefore, in mny opertions, such s the edge support checking nd edgehost grph checking, there is no need to visit the djcency informtion t ll. To the contrst, if the djcencylist representtion is used, every opertion hs to check the linked lists. When the dtse is lrge so tht either the djcencylists of ll grphs or the djcency informtion in the ADI structure cnnot e ccommodted into min memory, using the first two levels of the ADI structure cn sve mny clls to the djcency informtion, while the djcencylists of vrious grphs hve to e trnsferred etween the min memory nd the disk mny times. Usully, the djcencymtri is sprse. The djcencymtri representtion is inefficient in spce nd thus is not used. 5. ALGORITHM ADIMINE With the help from the ADI structure, how cn we improve the sclility nd efficiency of frequent grph pttern mining? Here, we present ptterngrowth lgorithm ADI Mine, which is n improvement of lgorithm gspn. The lgorithm is shown in Figure 6. If the ADI structure is unville, then the lgorithm scns the grph dtse nd constructs the inde. Otherwise, it just uses the ADI structure on the disk. The frequent edges cn e otined from the edge tle in the ADI structure. Ech frequent edge is one of the smllest frequent grph ptterns nd thus should e output. Then, the frequent edges should e used s the seeds to grow lrger frequent grph ptterns, nd the frequent djcent edges of e should e used in the ptterngrowth. An edge e is frequent djcent edge of e if e is n djcent edge of e in t lest min sup grphs. The set of frequent djcent edges cn e retrieved efficiently from the ADI structure since the identities of the grphs contining e re indeed s linkedlist, nd the djcent edges re lso indeed in the djcency informtion prt in the ADI structure. The pttern growth is implemented s clls to procedure sugrphmine. Procedure sugrphmine tries every frequent djcent edge e (i.e., edges in set F e ) nd checks whether e cn e dded into the current frequent grph pttern G to form lrger pttern G. We use the DFS code to test the redundncy. Only the ptterns G whose DFS code is minimum is output nd further grown. All other ptterns G re either found efore or will e found lter t other
7 Input: grph dtse GDB nd min sup Output: the complete set of frequent grph ptterns Method: construct the ADI structure for the grph dtse if it is not ville; for ech frequent edge e in the edge tle do output e s grph pttern; from the ADI structure, find set F e, the set of frequent djcent edges for e; cll sugrphmine(e, F e ); end for Procedure sugrphmine Prmeters: frequent grph pttern G, nd the set of frequent djcent edges F e // output the frequent grph ptterns whose // minimum DFScodes contin tht of G s prefi Method: for ech edge e in F e do let G e the grph y dding e into G; compute the DFS code of G ; if the DFS code is not minimum, then return; output G s frequent grph pttern; updte the set F e of djcent edges; cll sugrphmine(g, F e ); end for return; Figure 6: Algorithm. rnches. The correctness of this step is gurnteed y the property of DFS code [11]. Once lrger pttern G is found, the set of djcent edges of the current pttern should e updted, since the djcent edges of the newly inserted edge should lso e considered in the future growth from G. This updte opertion cn e implemented efficiently, since the identities of grphs tht contin n edge e re linked together in the ADI structure, nd the djcency informtion is lso indeed nd linked ccording to the grphids. Differences Between nd gspn At high level, the structure s well s the serch strtegies of nd gspn re similr. The criticl difference is on the storge structure for grphs uses ADI structure nd gspn uses djcencylist representtion. In the recursive mining, the criticl opertion is finding the grphs tht contin the current grph pttern (i.e., the test of sugrph isomorphism) nd finding the djcent edges to grow lrger grph ptterns. The current grph pttern is recorded using the lels. Thus, the edges re serched using the lels of the vertices nd tht of the edges. In gspn, the test of sugrph isomorphism is chieved y scnning the current (projected) dtse. Since the grphs re stored in djcencylist representtion, nd one lel my pper more thn once in grph, the serch cn e costly. For emple, in grph G 2 in Figure 3, in order to find n edge (C, d, A), the djcencylist for vertices 4 nd 6 my hve to e serched. If the grph is lrge nd the lels pper multiple times in grph, there my e mny djcencylists for vertices of the sme lel, nd the djcencylists re long. Moreover, for lrge grph dtse tht cnnot e held into min memory, the djcencylist representtion of grph hs to e loded into min memory efore the grph cn e serched. In, the grphs re stored in the ADI structure. The edges re indeed y their lels. Then, the grphs tht contin the edges cn e retrieved immeditely. Moreover, ll edges with the sme lels re linked together y the links etween the grphid nd the instnces. Tht helps the test of sugrph isomorphism sustntilly. Furthermore, using the inde of edges y their lels, only the grphs tht contin the specific edge will e loded into min memory for further sugrph isomorphism test. Irrelevnt grphs cn e filtered out immeditely y the inde. When the dtse is too lrge to fit into min memory, it sves sustntil prt of trnsfers of grphs etween disk nd min memory. 6. EXPERIMENTAL RESULTS In this section, we report systemtic performnce study on the ADI structure nd comprison of gspn nd ADI Mine on mining oth smll, memorysed dtses nd lrge, disksed dtses. We otin the eecutle of gspn from the uthors. The ADI structure nd lgorithm re implemented using C/C Eperiment Setting All the eperiments re conducted on n IBM NetFinity 51 mchine with n Intel PIII 733MHz CPU, 512M RAM nd 18G hrd disk. The speed of the hrd disk is 1, RPM. The operting system is Redht Linu 9.. We implement synthetic dt genertor following the procedure descried in [6]. The dt genertor tkes five prmeters s follows. D: the totl numer of grphs in the dt set T : the verge numer of edges in grphs I: the verge numer of edges in potentilly frequent grph ptterns (i.e., the frequent kernels) L: the numer of potentilly frequent kernels N: the numer of possile lels Plese refer to [6] for the detils of the dt genertor. For emple, dt set D1kN4I1T 2L2 mens tht the dt set contins 1k grphs; there re 4 possile lels; the verge numer of edges in the frequent kernel grphs is 1; the verge numer of edges in the grphs is 2; nd the numer of potentilly frequent kernels is 2. Herefter in this section, when we sy prmeters, it mens the prmeters for the dt genertor to crete the dt sets. In [11], L is fied to 2. In our eperiments, we lso set L = 2 s the defult vlue, ut will test the sclility of our lgorithm on L s well. Plese note tht, in ll eperiments, the runtime of ADI Mine includes oth the ADI construction time nd the mining time. 6.2 Mining Min Memorysed Dtses In this set of eperiments, oth gspn nd run in min memory.
8 6.2.1 Sclility on Minimum Support Threshold We test the sclility of gspn nd on the minimum support threshold. Dt set D1kN3I5T 2L2 is used. The minimum support threshold vries from 4% to 1%. The results re shown in Figure 7(). As cn e seen, oth gspn nd re sclle, ut is out 1 times fster. We discussed the result with Mr. X. Yn, the uthor of gspn. He confirms tht counting frequent edges in gspn is time consuming. On the other hnd, the construction of ADI structure is reltively efficient. When the minimum support threshold is set to 1, i.e., ll edges re indeed, the ADI structure uses pproimtely 57M min memory nd costs 86 seconds in construction Sclility on Dtse Size We test the sclility of gspn nd on the size of dtses. We fi the prmeters N = 3, I = 5, T = 2 nd L = 2, nd vry the numer of grphs in dtse from 5 thousnd to 1 thousnd. The minimum support threshold is set to 1% of the numer of grphs in the dtse. The results re shown in Figure 7(). The construction time of ADI structure is lso plotted in the figure. Both the lgorithms nd the construction of ADI structure re linerly sclle on the size of dtses. is fster. We oserve tht the size of ADI structure is lso sclle. For emple, it uses 28M when the dtse hs 5 thousnd grphs, nd 57M when the dtse hs 1 thousnd grphs. This oservtion concurs with Theorem Effects of Dt Set Prmeters We test the sclility of the two lgorithms on prmeter N the numer of possile lels. We use dt set D1kN25I5T 2L2, tht is, the N vlue vries from 2 to 5. The minimum support threshold is fied t 1%. The results re shown in Figure 7(c). Plese note tht the Y is is in logrithmic scle. We cn oserve tht the runtime of gspn increses eponentilly s N increses. This result is consistent with the result reported in [11]. 3 When there re mny possile lels in the dtse, the serch without inde ecomes drmticlly more costly. Interestingly, oth nd the construction of ADI structure re linerly sclle on N. As discussed efore, the edge tle in ADI structure only indees the unique edges in grph dtse. Serching using the indeed edge tle is efficient. The time compleity of serching n edge y lels is O(log n), where n is the numer of distinct edges in the dtse. This is not ffected y the increse of the possile lels. As epected, the size of the ADI structure is stle, out 57M in this eperiment. We use dt set D1kN3I5T 13L2 to test the sclility of the two lgorithms on prmeter T the verge numer of edges in grph. The minimum support threshold is set to 1%. The results re shown in Figure 7(d). As the numer of edges increses, the grph ecomes more comple. The cost of storing nd serching the grph lso increses ccordingly. As shown in the figure, oth lgorithms nd the construction of ADI re linerly sclle. We lso test the effects of other prmeters. The eperimentl results show tht oth gspn nd re not 3 Plese refer to Figures 5() nd 5(c) in the UIUC technicl report version of [11]. sensitive to I the verge numer of edges in potentilly frequent grph ptterns nd L the numer of potentilly frequent kernels. The construction time nd spce cost of ADI structures re lso stle. The reson is tht the effects of those two prmeters on the distriution in the dt sets re minor. Similr oservtions hve een reported y previous studies on mining frequent itemsets nd sequentil ptterns. Limited y spce, we omit the detils here. 6.3 Mining Disksed Dtses Now, we report the eperimentl results on mining lrge, disksed dtses. In this set of eperiments, we reserve lock of min memory of fied size for ADI structure. When the size is too smll for the ADIstructure, some levels of the ADI structure re ccommodted on disk. On the other hnd, we do not confine the memory usge for gspn Sclility on Dtse Size We test the sclility of oth gspn nd on the size of dtses. We use dt set D1k1mN3I5T 2L2. The numer of grphs in the dtse is vried from 1 thousnd to 1 million. The min memory lock for ADI structure is limited to 25M. The results re shown in Figure 8(). The construction time of ADI structure is lso plotted. Plese note tht the Y is is in logrithmic scle. The construction runtime of ADI structure is pproimtely liner on the dtse size. Tht is, the construction of the ADI inde is highly sclle. We lso mesure the size of ADI structure. The results re shown in Figure 8(). We cn oserve tht the size of the ADI structure is liner to the dtse size. In this eperiment, the rtio size of ADI structure in megytes is out.6. When the numer of grphs in thousnds dtse size is 1 million, the size of ADI structure is 61M, which eceeds the min memory size of our mchine. Even in such cse, the construction runtime is still liner. As eplined efore, the construction of ADI structure mkes sequentil scns of the dtse nd conducts sequentil write of the djcency informtion. The overhed of construction of edge tle nd the linked lists of grphids is reltively smll nd thus hs minor effect on the construction time. While gspn cn hndle dtses of only up to 3 thousnd grphs in this eperiment, cn hndle dtses of 1 million grphs. The curve of the runtime of cn e divided into three stges. First, when the dtse hs up to 3 thousnd grphs, the ADI structure cn e fully ccommodted in min memory. is fster thn gspn. Second, when the dtse hs 3 to 6 thousnd grphs, gspn cnnot finish. The ADI structure cnnot e fully held in min memory. Some prt of the djcency informtion is put on disk. We see significnt jump in the runtime curve of etween the dtses of 3 thousnd grphs nd 4 thousnd grphs. Lst, when the dtse hs 8 thousnd or more grphs, even the linked lists of grphids cnnot e fully put into min memory. Thus, nother significnt jump in the runtime curve cn e oserved Trdeoff Between Efficiency nd Min Memory Consumption It is interesting to emine the trdeoff etween efficiency nd size of ville min memory. We use dt set
9 Runtime (second) gspn Runtime (second) gspn ADIconstruction Runtime (second) gspn ADIconstruction Runtime (second) gspn ADIconstruction min_sup (%) Numer of grphs (thousnd) N T () sclility on min sup () Sclility on size (c) Sclility on N (d) Sclility on T D1kN3I5T 2L2 D51kN3I5T 2L2 D1kN25I5T 2L2 D1kN3I5T 13L2 min sup = 1% min sup = 1% min sup = 1% Figure 7: The eperimentl results of mining min memorysed dtses. Runtime (second) gspn ADIconstruction Size (M) ADI structure Runtime (s) Numer of grphs (thousnd) Numer of grphs (thousnd) Size of ville min memory (M) () sclility on size () Size of ADI structure (c) Runtime vs. min memory D1k1mN3I5T 2L2 D1k1mN3I5T 2L2 D1kN3I5T 2L2 min sup = 1% min sup = 1% min sup = 1% Figure 8: The eperimentl results of mining lrge disksed dtses. D1kN3I5T 2L2, set the minimum support threshold to 1%, vry the min memory limit from 1M to 15M for ADI structure, nd mesure the runtime of. The results re shown in Figure 8(c). In this eperiment, the size of ADI structure is 57M. The construction time is 86 seconds. The highest wtermrk of min memory usge for gspn in mining this dt set is 87M. gspn uses 1161 seconds in the mining if it hs sufficient min memory. When the ADI structure cn e completely loded into min memory (57M or lrger), runs fst. Further increse of the ville min memory cnnot reduce the runtime. When the ADI structure cnnot e fully put into min memory, the runtime increses. The more min memory, the fster runs. When the ville min memory is too smll to even hold the linked lists of grphids, the runtime of increses sustntilly. However, it still cn finish the mining with 1M min memory limit in 2 hours Numer of Disk Block Reds In ddition to runtime, the efficiency of mining lrge disksed dtses cn lso e mesured y the numer of disk lock red opertions. Figure 9() shows the numer of disk lock reds versus the minimum support threshold. When the support threshold is high (e.g., 9% or up), the numer of frequent edges is smll. The ADI structure cn e held into min memory nd thus the I/O cost is very low. As the support threshold goes down, lrger nd lrger prt of the ADI structure is stored on disk, nd the I/O cost increses. This curve is consistent with the trend in Figure 7(). Figure 9() shows the numer of disk lock reds versus the numer of grphs in the dtse. As the dtse size goes up, the I/O cost increses eponentilly. This eplins the curve of in Figure 8(). We lso test the I/O cost on ville min memory. The result is shown in Figure 9(c), which is consistent with the trend of runtime curve in Figure 8(c) Effects of Other Prmeters We lso test the effects of the other prmeters on the efficiency. We oserve similr trends s in mining memorysed dtses. Limited y spce, we omit the detils here. 6.4 Summry of Eperimentl Results The etensive performnce study clerly shows the following. First, oth gspn nd re sclle when dtse cn e held into min memory. is fster thn gspn. Second, cn mine very lrge grph dtses y ccommodting the ADI structure on disk. The performnce of on mining lrge disksed dtses is highly sclle. Third, the size of ADI structure is linerly sclle with respect to the size of dtses. Fourth, we cn control the trdeoff etween the mining efficiency nd the min memory consumption. Lst, is more sclle thn gspn in mining comple grphs the grphs tht hve mny different kinds of lels. 7. RELATED WORK The prolem of finding frequent common structures hs een studied since erly 199s. For emple, [1, 7] study the the prolem of finding common sustructures from chemicl compounds. SUBDUE [4] proposes n pproimte lgorithm to identify some, insted of the complete set of,
10 Numer of locks red min_sup (%) Numer of locks red 2.5e+8 2e+8 1.5e+8 1e+8 5e Numer of grphs (thousnd) Numer of locks red 4e+7 3.5e+7 3e+7 2.5e+7 2e+7 1.5e+7 1e+7 5e Size of ville min memory (M) () # locks vs. support threshold () # locks vs. dtse size (c) # locks vs. min memory size D1kN3I5T 2L2 D1k1mN3I5T 2L2 D1kN3I5T 2L2 min sup = 1% min sup = 1% Figure 9: The numer of disk locks red in the mining. frequent sustructures. However, these methods do not im t sclle lgorithms for mining lrge grph dtses. The prolem of mining the complete set of frequent grph ptterns is firstly eplored y Inokuchi et l. [5]. An Apriorilike lgorithm AGM is proposed. Kurmochi nd Krypis [6] develop n efficient lgorithm, FSG, for grph pttern mining. The mjor ide is to utilize n effective grph representtion, nd conduct the edgegrowth mining insted of vertegrowth mining. Both AGM nd FSG dopt redthfirst serch. Recently, Yn nd Hn propose the depthfirst serch pproch, gspn [11] for grph mining. They lso investigte the prolem of mining frequent closed grphs [9], which is nonredundnt representtion of frequent grph ptterns. As ltest result, Yn et l. [1] uses frequent grph ptterns to inde grphs. As specil cse of grph mining, tree mining lso receives intensive reserch recently. Zki [12] proposes the first lgorithm for mining frequent tree ptterns. Although there re quite few studies on the efficient mining of frequent grph ptterns, none of them ddresses the prolem of effective inde structure for mining lrge disksed grph dtses. When the dtse is too lrge to fit into min memory, the mining ecomes I/O ounded, nd the pproprite inde structure ecomes very criticl for the sclility. 8. CONCLUSIONS In this pper, we study the prolem of sclle mining of lrge disksed grph dtse. The ADI structure, n effective inde structure, is developed. Tking gspn s concrete emple, we propose, n efficient lgorithm dopting the ADI structure, to improve the sclility of the frequent grph mining sustntilly. The structure is generl inde for grph mining. As future work, it is interesting to emine the effect of the inde structure on improving other grph pttern mining methods, such s mining frequent closed grphs nd mining grphs with constrints. Furthermore, devising inde structures to support sclle dt mining on lrge disksed dtses is n importnt nd interesting reserch prolem with etensive pplictions nd industril vlues. Acknowledgements We re very grteful to Mr. Xifeng Yn nd Dr. Jiwei Hn for kindly providing us the eecutle of gspn nd nswering our questions promptly. We would like to thnk the nonymous reviewers for their insightful comments, which help to improve the qulity of the pper. 9. REFERENCES [1] D.M. Byd, R. W. Simpson, nd A. P. Johnson. An lgorithm for the multiple common sugrph prolem. J. of Chemicl Informtion & Computer Sci., 32:68 685, [2] C. Borgelt nd M.R. Berthold. Mining moleculr frgments: Finding relevnt sustructures of molecules. In Proc. 22 Int. Conf. Dt Mining (ICDM 2), Meshi TERRSA, Meshi City, Jpn, Dec. 22. [3] Thoms H. Cormen, Chrles E. Leiserson, Ronld L. Rivest, nd Clifford Stein. Introduction to Algorithms, Second Edition. MIT Press nd McGrwHill, 22. [4] L. B. Holder, D. J. Cook, nd S. Djoko. Sustructure discovery in the sudue system. In Proc. AAAI 94 Workshop Knowledge Discovery in Dtses (KDD 94), pges , Settle, WA, July [5] A. Inokuchi, T. Wshio, nd H. Motod. An priorised lgorithm for mining frequent sustructures from grph dt. In Proc. 2 Europen Symp. Principle of Dt Mining nd Knowledge Discovery (PKDD ), pges 13 23, Lyon, Frnce, Sept. 2. [6] M. Kurmochi nd G. Krypis. Frequent sugrph discovery. In Proc. 21 Int. Conf. Dt Mining (ICDM 1), pges , Sn Jose, CA, Nov. 21. [7] Y. Tkhshi, Y. Stoh, nd S. Sski. Recognition of lrgest common frgment mong vriety of chemicl structures. Anlyticl Sciences, 3:23 38, [8] N. Vnetik, E. Gudes, nd S.E. Shimony. Computing frequent grph ptterns from semistructured dt. In Proc. 22 Int. Conf. Dt Mining (ICDM 2), Meshi TERRSA, Meshi City, Jpn, Dec. 22. [9] X. Yn nd J. Hn. Closegrph: Mining closed frequent grph ptterns. In Proceedings of the 9th ACM SIGKDD Interntionl Conference on Knowledge Discovery nd Dt Mining (KDD 3), Wshington, D.C, 23. [1] X. Yn, P.S. Yu, nd J. Hn. Grph indeing: A frequent structuresed pproch. In Proc. 24 ACM SIGMOD Int. Conf. on Mngement of Dt (SIGMOD 4), Pris, Frnce, June 24. [11] Y. Yn nd J. Hn. gspn: Grphsed sustructure pttern mining. In Proc. 22 Int. Conf. on Dt Mining (ICDM 2), Meshi, Jpn, Decemer 22. [12] M.J. Zki. Efficiently mining frequent trees in forest. In Proc. 22 Int. Conf. on Knowledge Discovery nd Dt Mining (KDD 2), Edmonton, Alert, Cnd, July 22.
Reasoning to Solve Equations and Inequalities
Lesson4 Resoning to Solve Equtions nd Inequlities In erlier work in this unit, you modeled situtions with severl vriles nd equtions. For exmple, suppose you were given usiness plns for concert showing
More informationOr more simply put, when adding or subtracting quantities, their uncertainties add.
Propgtion of Uncertint through Mthemticl Opertions Since the untit of interest in n eperiment is rrel otined mesuring tht untit directl, we must understnd how error propgtes when mthemticl opertions re
More informationVariable Dry Run (for Python)
Vrile Dr Run (for Pthon) Age group: Ailities ssumed: Time: Size of group: Focus Vriles Assignment Sequencing Progrmming 7 dult Ver simple progrmming, sic understnding of ssignment nd vriles 2050 minutes
More informationPolynomial Functions. Polynomial functions in one variable can be written in expanded form as ( )
Polynomil Functions Polynomil functions in one vrible cn be written in expnded form s n n 1 n 2 2 f x = x + x + x + + x + x+ n n 1 n 2 2 1 0 Exmples of polynomils in expnded form re nd 3 8 7 4 = 5 4 +
More informationSmall Business Networking
Why network is n essentil productivity tool for ny smll business Effective technology is essentil for smll businesses looking to increse the productivity of their people nd business. Introducing technology
More informationSmall Businesses Decisions to Offer Health Insurance to Employees
Smll Businesses Decisions to Offer Helth Insurnce to Employees Ctherine McLughlin nd Adm Swinurn, June 2014 Employersponsored helth insurnce (ESI) is the dominnt source of coverge for nonelderly dults
More informationSmall Business Networking
Why network is n essentil productivity tool for ny smll business Effective technology is essentil for smll businesses looking to increse the productivity of their people nd processes. Introducing technology
More informationSmall Business Networking
Why network is n essentil productivity tool for ny smll business Effective technology is essentil for smll businesses looking to increse the productivity of their people nd business. Introducing technology
More informationSmall Business Networking
Why network is n essentil productivity tool for ny smll business Effective technology is essentil for smll businesses looking to increse the productivity of their people nd processes. Introducing technology
More informationP.3 Polynomials and Factoring. P.3 an 1. Polynomial STUDY TIP. Example 1 Writing Polynomials in Standard Form. What you should learn
33337_0P03.qp 2/27/06 24 9:3 AM Chpter P Pge 24 Prerequisites P.3 Polynomils nd Fctoring Wht you should lern Polynomils An lgeric epression is collection of vriles nd rel numers. The most common type of
More informationEQUATIONS OF LINES AND PLANES
EQUATIONS OF LINES AND PLANES MATH 195, SECTION 59 (VIPUL NAIK) Corresponding mteril in the ook: Section 12.5. Wht students should definitely get: Prmetric eqution of line given in pointdirection nd twopoint
More informationQuadratic Equations  1
Alger Module A60 Qudrtic Equtions  1 Copyright This puliction The Northern Alert Institute of Technology 00. All Rights Reserved. LAST REVISED Novemer, 008 Qudrtic Equtions  1 Sttement of Prerequisite
More informationSmall Business Networking
Why Network is n Essentil Productivity Tool for Any Smll Business TechAdvisory.org SME Reports sponsored by Effective technology is essentil for smll businesses looking to increse their productivity. Computer
More informationIn the following there are presented four different kinds of simulation games for a given Büchi automaton A = :
Simultion Gmes Motivtion There re t lest two distinct purposes for which it is useful to compute simultion reltionships etween the sttes of utomt. Firstly, with the use of simultion reltions it is possile
More informationBayesian Updating with Continuous Priors Class 13, 18.05, Spring 2014 Jeremy Orloff and Jonathan Bloom
Byesin Updting with Continuous Priors Clss 3, 8.05, Spring 04 Jeremy Orloff nd Jonthn Bloom Lerning Gols. Understnd prmeterized fmily of distriutions s representing continuous rnge of hypotheses for the
More informationAppendix D: Completing the Square and the Quadratic Formula. In Appendix A, two special cases of expanding brackets were considered:
Appendi D: Completing the Squre nd the Qudrtic Formul Fctoring qudrtic epressions such s: + 6 + 8 ws one of the topics introduced in Appendi C. Fctoring qudrtic epressions is useful skill tht cn help you
More informationSmall Business Networking
Why network is n essentil productivity tool for ny smll business Effective technology is essentil for smll businesses looking to increse the productivity of their people nd processes. Introducing technology
More informationSection 4.3. By the Mean Value Theorem, for every i = 1, 2, 3,..., n, there exists a point c i in the interval [x i 1, x i ] such that
Difference Equtions to Differentil Equtions Section 4.3 The Fundmentl Theorem of Clculus We re now redy to mke the longpromised connection between differentition nd integrtion, between res nd tngent lines.
More informationClearPeaks Customer Care Guide. Business as Usual (BaU) Services Peace of mind for your BI Investment
ClerPeks Customer Cre Guide Business s Usul (BU) Services Pece of mind for your BI Investment ClerPeks Customer Cre Business s Usul Services Tble of Contents 1. Overview...3 Benefits of Choosing ClerPeks
More informationMath Review 1. , where α (alpha) is a constant between 0 and 1, is one specific functional form for the general production function.
Mth Review Vribles, Constnts nd Functions A vrible is mthemticl bbrevition for concept For emple in economics, the vrible Y usully represents the level of output of firm or the GDP of n economy, while
More informationSuffix Trees CMSC 423
Suffix Trees CMSC 423 Preprocessing Strings Over the next few lectures, we ll see severl methods for preprocessing string dt into dt structures tht mke mny questions (like serching) esy to nswer: Suffix
More information2 DIODE CLIPPING and CLAMPING CIRCUITS
2 DIODE CLIPPING nd CLAMPING CIRCUITS 2.1 Ojectives Understnding the operting principle of diode clipping circuit Understnding the operting principle of clmping circuit Understnding the wveform chnge of
More informationhas the desired form. On the other hand, its product with z is 1. So the inverse x
First homework ssignment p. 5 Exercise. Verify tht the set of complex numers of the form x + y 2, where x nd y re rtionl, is sufield of the field of complex numers. Solution: Evidently, this set contins
More informationAssuming all values are initially zero, what are the values of A and B after executing this Verilog code inside an always block? C=1; A <= C; B = C;
B26 Appendix B The Bsics of Logic Design Check Yourself ALU n [Arthritic Logic Unit or (rre) Arithmetic Logic Unit] A rndomnumer genertor supplied s stndrd with ll computer systems Stn KellyBootle,
More informationSection 54 Trigonometric Functions
5 Trigonometric Functions Section 5 Trigonometric Functions Definition of the Trigonometric Functions Clcultor Evlution of Trigonometric Functions Definition of the Trigonometric Functions Alternte Form
More informationFAULT TREES AND RELIABILITY BLOCK DIAGRAMS. Harry G. Kwatny. Department of Mechanical Engineering & Mechanics Drexel University
SYSTEM FAULT AND Hrry G. Kwtny Deprtment of Mechnicl Engineering & Mechnics Drexel University OUTLINE SYSTEM RBD Definition RBDs nd Fult Trees System Structure Structure Functions Pths nd Cutsets Reliility
More informationRegular Sets and Expressions
Regulr Sets nd Expressions Finite utomt re importnt in science, mthemtics, nd engineering. Engineers like them ecuse they re super models for circuits (And, since the dvent of VLSI systems sometimes finite
More informationBinary Golomb Codes. CSE 589 Applied Algorithms Autumn Constructing a Binary Golomb Code. Example. Example. Comparison of GC with Entropy
CSE 589 pplied lgorithms utumn Golom Coding rithmetic Coding LZW Sequitur Binry Golom Codes Binry source with s much more frequent thn s. Vriletovrile length code Prefix code. Golom code of order 4 input
More informationSmall Business Cloud Services
Smll Business Cloud Services Summry. We re thick in the midst of historic sechnge in computing. Like the emergence of personl computers, grphicl user interfces, nd mobile devices, the cloud is lredy profoundly
More informationMorgan Stanley Ad Hoc Reporting Guide
spphire user guide Ferury 2015 Morgn Stnley Ad Hoc Reporting Guide An Overview For Spphire Users 1 Introduction The Ad Hoc Reporting tool is ville for your reporting needs outside of the Spphire stndrd
More information9 CONTINUOUS DISTRIBUTIONS
9 CONTINUOUS DISTIBUTIONS A rndom vrible whose vlue my fll nywhere in rnge of vlues is continuous rndom vrible nd will be ssocited with some continuous distribution. Continuous distributions re to discrete
More informationAntiSpyware Enterprise Module 8.5
AntiSpywre Enterprise Module 8.5 Product Guide Aout the AntiSpywre Enterprise Module The McAfee AntiSpywre Enterprise Module 8.5 is n ddon to the VirusScn Enterprise 8.5i product tht extends its ility
More informationEconomics Letters 65 (1999) 9 15. macroeconomists. a b, Ruth A. Judson, Ann L. Owen. Received 11 December 1998; accepted 12 May 1999
Economics Letters 65 (1999) 9 15 Estimting dynmic pnel dt models: guide for q mcroeconomists b, * Ruth A. Judson, Ann L. Owen Federl Reserve Bord of Governors, 0th & C Sts., N.W. Wshington, D.C. 0551,
More informationSirindhorn International Institute of Technology Thammasat University at Rangsit
Sirindhorn Interntionl Institute of Technology Thmmst University t Rngsit School of Informtion, Computer nd Communiction Technology COURSE : ECS 204 Bsic Electricl Engineering L INSTRUCTOR : Asst. Prof.
More informationUse Geometry Expressions to create a more complex locus of points. Find evidence for equivalence using Geometry Expressions.
Lerning Objectives Loci nd Conics Lesson 3: The Ellipse Level: Preclculus Time required: 120 minutes In this lesson, students will generlize their knowledge of the circle to the ellipse. The prmetric nd
More informationSolutions to Section 1
Solutions to Section Exercise. Show tht nd. This follows from the fct tht mx{, } nd mx{, } Exercise. Show tht = { if 0 if < 0 Tht is, the bsolute vlue function is piecewise defined function. Grph this
More information10.5 Graphing Quadratic Functions
0.5 Grphing Qudrtic Functions Now tht we cn solve qudrtic equtions, we wnt to lern how to grph the function ssocited with the qudrtic eqution. We cll this the qudrtic function. Grphs of Qudrtic Functions
More informationUnleashing the Power of Cloud
Unleshing the Power of Cloud A Joint White Pper by FusionLyer nd NetIQ Copyright 2015 FusionLyer, Inc. All rights reserved. No prt of this publiction my be reproduced, stored in retrievl system, or trnsmitted,
More informationCombined Liability Insurance. Information and Communication Technology Proposal form
Comined Liility Insurnce Informtion nd Communiction Technology Proposl form Comined Liility Insurnce Informtion nd Communiction Technology  Proposl form This proposl form must e completed nd signed y
More informationDlNBVRGH + Sickness Absence Monitoring Report. Executive of the Council. Purpose of report
DlNBVRGH + + THE CITY OF EDINBURGH COUNCIL Sickness Absence Monitoring Report Executive of the Council 8fh My 4 I.I...3 Purpose of report This report quntifies the mount of working time lost s result of
More informationProtocol Analysis. 17654/17764 Analysis of Software Artifacts Kevin Bierhoff
Protocol Anlysis 17654/17764 Anlysis of Softwre Artifcts Kevin Bierhoff TkeAwys Protocols define temporl ordering of events Cn often be cptured with stte mchines Protocol nlysis needs to py ttention
More informationConcept Formation Using Graph Grammars
Concept Formtion Using Grph Grmmrs Istvn Jonyer, Lwrence B. Holder nd Dine J. Cook Deprtment of Computer Science nd Engineering University of Texs t Arlington Box 19015 (416 Ytes St.), Arlington, TX 760190015
More informationMeasuring Similarity between Graphs Based on the Levenshtein Distance
Appl. Mth. Inf. Sci. 7, No. 1L, 169175 (01) 169 Applied Mthemtics & Informtion Sciences An Interntionl Journl Mesuring Similrity etween Grphs Bsed on the Levenshtein Distnce Bin Co, ing Li nd Jinwei in
More informationSection 5.2, Commands for Configuring ISDN Protocols. Section 5.3, Configuring ISDN Signaling. Section 5.4, Configuring ISDN LAPD and Call Control
Chpter 5 Configurtion of ISDN Protocols This chpter provides instructions for configuring the ISDN protocols in the SP201 for signling conversion. Use the sections tht reflect the softwre you re configuring.
More informationA Note on Complement of Trapezoidal Fuzzy Numbers Using the αcut Method
Interntionl Journl of Applictions of Fuzzy Sets nd Artificil Intelligence ISSN  Vol.  A Note on Complement of Trpezoidl Fuzzy Numers Using the αcut Method D. Stephen Dingr K. Jivgn PG nd Reserch Deprtment
More informationCONIC SECTIONS. Chapter 11
CONIC SECTIONS Chpter 11 11.1 Overview 11.1.1 Sections of cone Let l e fied verticl line nd m e nother line intersecting it t fied point V nd inclined to it t n ngle α (Fig. 11.1). Fig. 11.1 Suppose we
More informationOperations with Polynomials
38 Chpter P Prerequisites P.4 Opertions with Polynomils Wht you should lern: Write polynomils in stndrd form nd identify the leding coefficients nd degrees of polynomils Add nd subtrct polynomils Multiply
More informationIntegration. 148 Chapter 7 Integration
48 Chpter 7 Integrtion 7 Integrtion t ech, by supposing tht during ech tenth of second the object is going t constnt speed Since the object initilly hs speed, we gin suppose it mintins this speed, but
More informationLINEAR TRANSFORMATIONS AND THEIR REPRESENTING MATRICES
LINEAR TRANSFORMATIONS AND THEIR REPRESENTING MATRICES DAVID WEBB CONTENTS Liner trnsformtions 2 The representing mtrix of liner trnsformtion 3 3 An ppliction: reflections in the plne 6 4 The lgebr of
More informationSquare Roots Teacher Notes
Henri Picciotto Squre Roots Techer Notes This unit is intended to help students develop n understnding of squre roots from visul / geometric point of view, nd lso to develop their numer sense round this
More informationBasic Research in Computer Science BRICS RS0213 Brodal et al.: Solving the String Statistics Problem in Time O(n log n)
BRICS Bsic Reserch in Computer Science BRICS RS0213 Brodl et l.: Solving the String Sttistics Prolem in Time O(n log n) Solving the String Sttistics Prolem in Time O(n log n) Gerth Stølting Brodl Rune
More informationCS99S Laboratory 2 Preparation Copyright W. J. Dally 2001 October 1, 2001
CS99S Lortory 2 Preprtion Copyright W. J. Dlly 2 Octoer, 2 Ojectives:. Understnd the principle of sttic CMOS gte circuits 2. Build simple logic gtes from MOS trnsistors 3. Evlute these gtes to oserve logic
More informationVoIP for the Small Business
Reducing your telecommunictions costs VoIP (Voice over Internet Protocol) offers low cost lterntive to expensive trditionl phone services nd is rpidly becoming the communictions system of choice for smll
More informationEcon 4721 Money and Banking Problem Set 2 Answer Key
Econ 472 Money nd Bnking Problem Set 2 Answer Key Problem (35 points) Consider n overlpping genertions model in which consumers live for two periods. The number of people born in ech genertion grows in
More informationTuring Machine Extensions
Red K & S 4.3.1, 4.4. Do Homework 19. Turing Mchine Extensions Turing Mchine Definitions An lterntive definition of Turing mchine: (K, Σ, Γ, δ, s, H): Γ is finite set of llowble tpe symbols. One of these
More informationSolving the String Statistics Problem in Time O(n log n)
Solving the String Sttistics Prolem in Time O(n log n) Gerth Stølting Brodl 1,,, Rune B. Lyngsø 3, Ann Östlin1,, nd Christin N. S. Pedersen 1,2, 1 BRICS, Deprtment of Computer Science, University of Arhus,
More informationHillsborough Township Public Schools Mathematics Department Computer Programming 1
Essentil Unit 1 Introduction to Progrmming Pcing: 15 dys Common Unit Test Wht re the ethicl implictions for ming in tody s world? There re ethicl responsibilities to consider when writing computer s. Citizenship,
More informationA.7.1 Trigonometric interpretation of dot product... 324. A.7.2 Geometric interpretation of dot product... 324
A P P E N D I X A Vectors CONTENTS A.1 Scling vector................................................ 321 A.2 Unit or Direction vectors...................................... 321 A.3 Vector ddition.................................................
More informationSCHOOL OF ENGINEERING & BUILT ENVIRONMENT. Mathematics. Basic Algebra
SCHOOL OF ENGINEERING & BUILT ENVIRONMENT Mthemtics Bsic Alger. Opertions nd Epressions. Common Mistkes. Division of Algeric Epressions. Eponentil Functions nd Logrithms. Opertions nd their Inverses. Mnipulting
More informationHelicopter Theme and Variations
Helicopter Theme nd Vritions Or, Some Experimentl Designs Employing Pper Helicopters Some possible explntory vribles re: Who drops the helicopter The length of the rotor bldes The height from which the
More informationAlgorithms Chapter 4 Recurrences
Algorithms Chpter 4 Recurrences Outline The substitution method The recursion tree method The mster method Instructor: Ching Chi Lin 林清池助理教授 chingchilin@gmilcom Deprtment of Computer Science nd Engineering
More informationBasics of Logic Design: Boolean Algebra, Logic Gates. Administrative
Bsics of Logic Design: Boolen Alger, Logic Gtes Computer Science 104 Administrtive Homework #3 Due Sundy Midterm I Mondy in clss, closed ook, closed notes Ø Will provide IA32 instruction set hndout Ø Lst
More informationLecture 15  Curve Fitting Techniques
Lecture 15  Curve Fitting Techniques Topics curve fitting motivtion liner regression Curve fitting  motivtion For root finding, we used given function to identify where it crossed zero where does fx
More informationRTL Power Optimization with Gatelevel Accuracy
RTL Power Optimiztion with Gtelevel Accurcy Qi Wng Cdence Design Systems, Inc Sumit Roy Clypto Design Systems, Inc 555 River Oks Prkwy, Sn Jose 95125 2903 Bunker Hill Lne, Suite 208, SntClr 95054 qwng@cdence.com
More informationHomework 3 Solutions
CS 341: Foundtions of Computer Science II Prof. Mrvin Nkym Homework 3 Solutions 1. Give NFAs with the specified numer of sttes recognizing ech of the following lnguges. In ll cses, the lphet is Σ = {,1}.
More informationAddition and subtraction of rational expressions
Lecture 5. Addition nd subtrction of rtionl expressions Two rtionl expressions in generl hve different denomintors, therefore if you wnt to dd or subtrct them you need to equte the denomintors first. The
More informationIntegration by Substitution
Integrtion by Substitution Dr. Philippe B. Lvl Kennesw Stte University August, 8 Abstrct This hndout contins mteril on very importnt integrtion method clled integrtion by substitution. Substitution is
More informationThe Math Learning Center PO Box 12929, Salem, Oregon 97309 0929 Math Learning Center
Resource Overview Quntile Mesure: Skill or Concept: 1010Q Determine perimeter using concrete models, nonstndrd units, nd stndrd units. (QT M 146) Use models to develop formuls for finding res of tringles,
More informationAll pay auctions with certain and uncertain prizes a comment
CENTER FOR RESEARC IN ECONOMICS AND MANAGEMENT CREAM Publiction No. 12015 All py uctions with certin nd uncertin prizes comment Christin Riis All py uctions with certin nd uncertin prizes comment Christin
More informationModular Generic Verification of LTL Properties for Aspects
Modulr Generic Verifiction of LTL Properties for Aspects Mx Goldmn Shmuel Ktz Computer Science Deprtment Technion Isrel Institute of Technology {mgoldmn, ktz}@cs.technion.c.il ABSTRACT Aspects re seprte
More informationHow fast can we sort? Sorting. Decisiontree model. Decisiontree for insertion sort Sort a 1, a 2, a 3. CS 3343  Spring 2009
CS 4  Spring 2009 Sorting Crol Wenk Slides courtesy of Chrles Leiserson with smll chnges by Crol Wenk CS 4 Anlysis of Algorithms 1 How fst cn we sort? All the sorting lgorithms we hve seen so fr re comprison
More informationEziLink. 1 About EziLink PC SOFTWARE GUIDE. Import. Export. Configuration settings. Software updates
PC SOFTWARE GUIDE EziLink 1 Aout EziLink EziLink PC softwre llows you to mnge TruTest devices using your computer. It cn e instlled on computers with Microsoft Windows XP or lter. EziLink will offer different
More informationPlotting and Graphing
Plotting nd Grphing Much of the dt nd informtion used by engineers is presented in the form of grphs. The vlues to be plotted cn come from theoreticl or empiricl (observed) reltionships, or from mesured
More informationUnion, Intersection and Complement. Formal Foundations Computer Theory
Union, Intersection nd Complement FAs Union, Intersection nd Complement FAs Forml Foundtions Computer Theory Ferury 21, 2013 This hndout shows (y exmples) how to construct FAs for the union, intersection
More informationDistributions. (corresponding to the cumulative distribution function for the discrete case).
Distributions Recll tht n integrble function f : R [,] such tht R f()d = is clled probbility density function (pdf). The distribution function for the pdf is given by F() = (corresponding to the cumultive
More information5 a LAN 6 a gateway 7 a modem
STARTER With the help of this digrm, try to descrie the function of these components of typicl network system: 1 file server 2 ridge 3 router 4 ckone 5 LAN 6 gtewy 7 modem Another Novell LAN Router Internet
More informationTests for One Poisson Mean
Chpter 412 Tests for One Poisson Men Introduction The Poisson probbility lw gives the probbility distribution of the number of events occurring in specified intervl of time or spce. The Poisson distribution
More informationCalculus of variations with fractional derivatives and fractional integrals
Anis do CNMAC v.2 ISSN 1984820X Clculus of vritions with frctionl derivtives nd frctionl integrls Ricrdo Almeid, Delfim F. M. Torres Deprtment of Mthemtics, University of Aveiro 3810193 Aveiro, Portugl
More information1 Numerical Solution to Quadratic Equations
cs42: introduction to numericl nlysis 09/4/0 Lecture 2: Introduction Prt II nd Solving Equtions Instructor: Professor Amos Ron Scribes: Yunpeng Li, Mrk Cowlishw Numericl Solution to Qudrtic Equtions Recll
More informationGeneralized Inverses: How to Invert a NonInvertible Matrix
Generlized Inverses: How to Invert NonInvertible Mtrix S. Swyer September 7, 2006 rev August 6, 2008. Introduction nd Definition. Let A be generl m n mtrix. Then nturl question is when we cn solve Ax
More informationVoIP for the Small Business
Reducing your telecommunictions costs Reserch firm IDC 1 hs estimted tht VoIP system cn reduce telephonyrelted expenses by 30%. Voice over Internet Protocol (VoIP) hs become vible solution for even the
More informationVoIP for the Small Business
VoIP for the Smll Business Reducing your telecommunictions costs Reserch firm IDC 1 hs estimted tht VoIP system cn reduce telephonyrelted expenses by 30%. Voice over Internet Protocol (VoIP) hs become
More informationHomework #6: Answers. a. If both goods are produced, what must be their prices?
Text questions, hpter 7, problems 12. Homework #6: Answers 1. Suppose there is only one technique tht cn be used in clothing production. To produce one unit of clothing requires four lborhours nd one
More informationEncoded Bitmp Indexing for Dt Wrehouses MingChun Wu DVS, Computer Science Deprtment Technische Universitt Drmstdt, GERMANY wu@dvsinformtiktudrmstdtde Alejndro P Buchmnn DVS, Computer Science Deprtment
More informationChapter 9: Quadratic Equations
Chpter 9: Qudrtic Equtions QUADRATIC EQUATIONS DEFINITION + + c = 0,, c re constnts (generlly integers) ROOTS Synonyms: Solutions or Zeros Cn hve 0, 1, or rel roots Consider the grph of qudrtic equtions.
More information5.2. LINE INTEGRALS 265. Let us quickly review the kind of integrals we have studied so far before we introduce a new one.
5.2. LINE INTEGRALS 265 5.2 Line Integrls 5.2.1 Introduction Let us quickly review the kind of integrls we hve studied so fr before we introduce new one. 1. Definite integrl. Given continuous relvlued
More informationExperiment 6: Friction
Experiment 6: Friction In previous lbs we studied Newton s lws in n idel setting, tht is, one where friction nd ir resistnce were ignored. However, from our everydy experience with motion, we know tht
More informationGraphs on Logarithmic and Semilogarithmic Paper
0CH_PHClter_TMSETE_ 3//00 :3 PM Pge Grphs on Logrithmic nd Semilogrithmic Pper OBJECTIVES When ou hve completed this chpter, ou should be ble to: Mke grphs on logrithmic nd semilogrithmic pper. Grph empiricl
More informationRethinking Virtual Network Embedding: Substrate Support for Path Splitting and Migration
Rethinking Virtul Network Emedding: Sustrte Support for Pth Splitting nd Migrtion Minln Yu, Yung Yi, Jennifer Rexford, Mung Ching Princeton University Princeton, NJ {minlnyu,yyi,jrex,chingm}@princeton.edu
More informationLiveEngage Agent Guide. Version 1.0
Agent Guide Version 1.0 Tle of Contents Contents LOGGING IN... 3 NAVIGATING LIVEENGAGE... 4 Visitor List, Connection Are, Enggement Br & Going Online... 4 ENGAGEMENT OPTIONS... 5 Tking n Enggment... 5
More informationOne Minute To Learn Programming: Finite Automata
Gret Theoreticl Ides In Computer Science Steven Rudich CS 15251 Spring 2005 Lecture 9 Fe 8 2005 Crnegie Mellon University One Minute To Lern Progrmming: Finite Automt Let me tech you progrmming lnguge
More informationVoIP for the Small Business
Reducing your telecommunictions costs Reserch firm IDC 1 hs estimted tht VoIP system cn reduce telephonyrelted expenses by 30%. Voice over Internet Protocol (VoIP) hs become vible solution for even the
More information1. Find the zeros Find roots. Set function = 0, factor or use quadratic equation if quadratic, graph to find zeros on calculator
AP Clculus Finl Review Sheet When you see the words. This is wht you think of doing. Find the zeros Find roots. Set function =, fctor or use qudrtic eqution if qudrtic, grph to find zeros on clcultor.
More informationRegular Language Membership Constraint
Regulr Lnguge Memership Constrint Niko Pltzer Universität des Srlndes, Deutschlnd, nikopp@ps.unis.de, Advisor: Lutz Strßurger Astrct. This pper minly dels with the work of Gilles Pesnt on the stretch
More informationAn Undergraduate Curriculum Evaluation with the Analytic Hierarchy Process
An Undergrdute Curriculum Evlution with the Anlytic Hierrchy Process Les Frir Jessic O. Mtson Jck E. Mtson Deprtment of Industril Engineering P.O. Box 870288 University of Albm Tuscloos, AL. 35487 Abstrct
More informationVoIP for the Small Business
VoIP for the Smll Business Reducing your telecommunictions costs Reserch firm IDC 1 hs estimted tht VoIP system cn reduce telephonyrelted expenses by 30%. Voice over Internet Protocol (VoIP) hs become
More informationMathematics. Vectors. hsn.uk.net. Higher. Contents. Vectors 128 HSN23100
hsn.uk.net Higher Mthemtics UNIT 3 OUTCOME 1 Vectors Contents Vectors 18 1 Vectors nd Sclrs 18 Components 18 3 Mgnitude 130 4 Equl Vectors 131 5 Addition nd Subtrction of Vectors 13 6 Multipliction by
More informationFactoring Polynomials
Fctoring Polynomils Some definitions (not necessrily ll for secondry school mthemtics): A polynomil is the sum of one or more terms, in which ech term consists of product of constnt nd one or more vribles
More information1. The leves re either lbeled with sentences in ;, or with sentences of the form All X re X. 2. The interior leves hve two children drwn bove them) if
Q520 Notes on Nturl Logic Lrry Moss We hve seen exmples of wht re trditionlly clled syllogisms lredy: All men re mortl. Socrtes is mn. Socrtes is mortl. The ide gin is tht the sentences bove the line should
More information