Self-Organizing Maps – User Manual
Contents
1. button find path of SOM dIl and choose SOMmain from class list Document Self organizing maps Page 12 17 A Manage modules id Description Type _ Runnable class DILL Path 110 Neural network normal Neural etStart DM BPN etwork dll 107 Fill Null Values simple FormFillM ull alues Simple Transform dll 1102 Normalization simple Formhormalization Simple T ransforms dll 1103 Split table simple FormS plit T able Simple Transform dll 50 MBA normal MEM ain amp skola sothware projekt yd Add ok Cancel 3 4 Run SOM module Choose SOM module from the list in Methods menu 3 5 Load data to SOM module Load data for the method You can use File gt Load Data or icon in tool bar 3 6 Specify columns for training and set SOM parameters In menu SOM gt Set Columns choose columns which will be used for network training At least one column must be selected You can choose a descriptive column too This is optional not compulsory SetColumnsDialog l l x Data Columns Available columna Selected columns MNazow varchar double double double double lt lt Remove double double double b C d double double e double Add All f double f double q double g double Clear h double Descriptive Column Select column which will Used for visualisation Nazov varchar Cancel ocument Self organizing2. maps age Ea Ot KNOCKER Set parameters for training in menu SOM gt Parameters Default parameters are sufficient but you can try to set different values to get better results Se lt Kohonen Network Size 5 ili E Parameters Number of iteration Neighbourhood Shape of neighbourhood hexagonal Alpha 0 5 0 2 Cancel 3 7 Run algorithm Now you can run algorithm SOM gt Train If you choose Show row name and you have many entries in the table it can cause that algorithm run will slow down Complexity for painting descriptive string is O N M where N is number of entries in table and M is number of neurons The better way is to run algorithm without showing any row name and after the algorithm finished set this option on 3 8 Classification After SOM is trained you can run cluster algorithm on it to execute the classification Clustering settings are accessible in the Classification gt Parameters dialog There are two basic types of clustering Hierarchical K mean Hierarchical clustering is slower than K mean clustering but it returns better results Document Self organizing maps Page 14 17 O Fn Settings for Clustering Settings Clusters 4 7 Hierarchical Clustering Maximum Distance C Minimum Distance C Mean Distance C K Mean Clustering OF Cancel After the classification algorithm ends you can see lines which separate netwo
3. FA Self Organizing Maps User Manual Author Miroslav Pich Library SOM dll Runnable class SOMmain Document Self organizing maps Page 1 17 IRA SA KNOCKER Content 1 INTRODUCTION TO SELF ORGANIZING MAPS ccccccceccccccccccoccccooeocooesocooseoooesocosesccesecessscouesecesesoouessses 3 2 SELF ORGANIZING MAPS USER INTERFACE cccccssssssccssssscccsssccccccsssccccsssccccessscccessscoccsscccceesccccsesssces 4 2 1 MAINA IN DOW Garren nee RAT E A Pe E EA PO OE NO MA TN RON 4 2 2 Sa 12 51 2 eta ces cts tee ee olsen csc aa tesa vt be oneal ts ed league a a 0110229090552201 20 A E E 4 DEEN gt ANON YAA testers AIEA C0272009 0279909900007072 Y22101204 0240224 296 VEN A N 4 DD Dacascos cet opr E VI teeta ened ee dee een ete eee cence eee NER D 5 22 39 CSS A E a AN sees eters eae ends ene snare daca 8044 049 AA E ace RA PERAS 1 007 7 PAN COURRE lt gt gt lt eo EP A ROE Oe AAE EAA E E 20 909 ny mae rer A 9 22o TIO y E cass tose MAER E augue cee eee ROAD ean eaves eens eee cence eee 10 2 3 SELF ORGANIZING MAPS VISUALIZATION cccccccccccccccssccceessssseeeeeccccccsseeseeeceeeesssseeeeecccecssseeessueeeuussseeeeeeeeseess 11 5 IN TA TR Aca sce wns eaters dc eeneaceasnc oc ecue A E 12 3 1 PRETE A OURA T E nniyarenastaasascoueaseeconansuacoudasunaducanatsnessaueosieonsnsosasasasiaacas 12 3 2 ADD YOUR DATA AS VERSION IN MAIN APPLICATION ccc ccccccccccccecccccccccesocccccssosoooocecsssssooocoassososoceaassaus 12 3 3 LOAD
4. SOM MODULE IF NOT LOADED INTO APPLICATION ccsssccessceesscesssceesaceesaeeesaceesaeeesaeeesaeeesaeeesaeenses 12 3 4 UM SOMMODULE stra sc cose parse sca E E E E T E 13 3 5 LOADDATATOSOM MODULE csc ia ca ste ctesoareaea enencea O E 13 3 6 SPECIFY COLUMNS FOR TRAINING AND SET SOM PARAMETERS cccccccccscesessssseeeeececcccseeseessuuueeenseeeeeeeseess 13 3 7 KONAL OR IN A E sasonasuoasanueassaueasseacuatoagaeaiasieaces 14 3 8 CLASSIFICATION T E E A E A E E O 14 4 RKEOUIREMENT orir T S EET EEE 16 5 DAMPLES aa EA E A 17 Document Self organizing maps Page 2 17 FA St KNOCKER 1 Introduction to Self Organizing Maps Self organizing maps also called Kohonen feature maps are special kinds of neural networks that can be used for clustering tasks They are an extension of so called learning vector quantization Every self organizing map consists of two layers of neurons an input layer and a so called competition layer Weights of the connections from the input neurons to a single neuron in the competition layer are interpreted as a reference vector in the input space Such self organizing map basically represents a set of vectors in the input space one vector for each neuron in the competition layer A self organizing map is trained with a method called competition learning When an input pattern is presented to the network the neuron in the competition layer which reference vector is the closest to the input pattern is dete
5. There will be one single winning neuron It is the neuron whose weight vector lies closest to the input vector This can be simply determined by calculating the Euclidean distance between input vectors and weight vector Detailed information about SOM can be found at the following links http en wikipedia org wiki Self organizing ma http www ai junkie com ann som som1 html Document Self organizing maps Page 3 17 EA SR KNOCKER 2 Self Organizing Maps User Interface 2 1 Main Window You can see menu toolbar and view for maps in the main window The most important commands from the main menu are situated on the tool bar gt Self Organizing Maps oj x File SOM Classification View Help Pejal alhaja Hja vele 2 2 2 Menu 2 2 1 File e Load Data Li This command opens a dialog in which user can choose a version of data for SOM Document Self organizing maps Page 4 17 Ea Of KNOCKER x Choose version Select Cancel e Load SOM This command opens a dialog for loading SOM from file It shows only files with som extension by default e Save SOM i This command saves SOM to the file on disk Default extension of such file is som e Save Picture s This command saves the view area to one of the following picture file types BMP bitmap image format EMF Enhanced Windows metafile image format EXIF Exchangeable Image File format GIF Get
6. changed The main advantages of this algorithm are its simplicity and speed which allows it to run on large datasets Yet it does not systematically return the same result with each run of the algorithm Rather the resulting clusters depend on the initial assignments The k means algorithm maximizes inter cluster or minimizes intra cluster variance but it is not sure that the given solution is not a local minimum of variance e Save Classification Data Ss This command saves classified data to the database as a new version with one more column This column contains cluster ID of the nearest neurons to the vector from appropriate row 2 24 View e Zoom in EN This command zooms in the SOM view If user right clicks on the view and moves mouse to left the view will be zooming in e Zoom out EN This command zooms out the SOM view If user right clicks on the view and moves mouse to right the view will be zooming out e Fitto Size j This command fits the graphics objects on the view to the current size of the client area e Show Row Name Row name is a value in the descriptive column User can set descriptive columns in SOM gt Set Columns gt Descriptive Column Row name will be shown next to the nearest neuron If more then three rows vector are mapped to one neuron than only first two will be painted and text is placed in the third row This signalizes that more than three vectors are mapped Docume
7. ill be affected Decrease whit time o Cooling K determines how fast the fill size of the neighborhood and the alpha parameter decrease e Set Columns User can set columns which will be used for SOM training If there are no specified columns all columns which type is Double or Int will be used Document Self organizing maps Page 6 17 fn SetColumnsDialog l X Data Columns Available columna Selected columns Nazov varchar double double double double lt lt Remove double double double double double double Add sll f double f double q double g double Clear h double Descriptive Column Select column which will Used for visualisation Data Columns are columns which are used for SOM training Descriptive column determined string which will be painted near the best matching neuron e Reset Networks This command creates a new network and sets random weights on the neuron networks 2 2 3 Classificiation e Classify E This command divides neurons to clusters e Parameters C This command opens a dialog to set parameters for classification Settings for Clustering o E Settings Clusters H 4 Hierarchical Clustering Maximum Distance C Minimum Distance C Mean Distance C K Mean Clustering OF Cancel User can choose one of the two types of clustering algorithm Document Self organizing map
8. nt Self organizing maps Page 9 17 FA e Show Cluster ID This command shows the cluster ID above the neuron e Show Cluster Borders This command shows cluster borders between the neurons If the neighbor neurons are in different clusters line between them will be painted as shown bellow 2 2 5 Help e About info dialog about application Document Self organizing maps Page 10 17 FA at KNOCKER 2 3 Self Organizing Maps Visualization U matrix is used for the SOM visualization U matrix unified distance matrix visualizes the distances between the neurons red dots The distance between the adjacent neurons is calculated and presented with different colorings between the adjacent nodes A dark coloring between the neurons corresponds to the large distance and thus a gap between the codebook values in the input space A light coloring between the neurons signifies that the codebook vectors are close to each other in the input space Light areas can be thought as clusters and dark areas as cluster separators This can be a helpful presentation when one tries to find clusters in the input data without having any a priori information about the clusters Figure 7 U matrix representation of the Self Organizing Map We can see the neurons of the network marked as red dots in the Figure 7 The representation reveals that there is a separate cluster in the upper left corner of this representation The clusters are separated b
9. rks to clusters Now you can save classified data to a new version in database Classification gt Save Classified data A new version of data with one more column will be created from the source data table This column contains cluster ID of the nearest neurons from the vector of the appropriate row Final algorithm product can look like the following picture Classified data with K Mean algorithm cument Self organizing maps age FA 4 Requirements Files needed to run SOM module e all common components of main application Knocker e SOM dil e DMTransformStruct dll e Visualization dll e Guikxt dll e Gui dll Document Self organizing maps Page 16 17 FA Ot KNOCKER 5 Samples You can find sample data for SOM in file SOM_WDl csv It s a table of countries with some economical data Document Self organizing maps Page 17 17
10. rmined This neuron is called the winner neuron and it is the focal point of the weight changes Neighborhood relation in self organizing maps is defined on the competition layer The neighborhood indicates which weights of other neurons should also be changed This neighborhood relation is usually represented as a in most cases two dimensional grid The vertices of this grid are the neurons This grid is most often rectangular or hexagonal The weights of all neurons in the competition layer which are situated within a certain radius around the winner neuron are also adapted during the learning process But strength of the adaptation of such close neurons may depend on their distance from the winner neuron Main effect of this method is that the grid is spread out over the region of the input space Input space is covered by the training patterns Like most artificial neural networks the SOM has two modes of operation 1 At the beginning a map is built during the training process Then the neural network organizes itself using the competitive process A large number of input vectors must be given to the network Preferably as much vectors representing the kind of vectors expected during the second phase as possible Otherwise all input vectors ought to be administered several times 2 Each new input vector should be quickly given a location on the map during the mapping process Then the vector is automatically classified or categorized
11. s Page 7 17 LE KNOCKER 2 2 3 1 Hierarchical Clustering Hierarchical clustering builds agglomerative or breaks up divisive hierarchy of clusters A traditional representation of this hierarchy is a tree This tree consists of individual elements on one side and a single cluster with all elements on the other side Agglomerative algorithms begin at the top of the tree whereas divisive algorithms begin at the bottom The arrows in the figure bellow indicate an agglomerative clustering Cutting the tree at a given height will return a clustering at the selected precision Cutting bellow the second row in the following example will return clusters a b c d e f Cutting bellow the third row will return clusters a b c d e f The second cutting is coarser clustering with less number of larger clusters Agglomerative hierarchical clustering This method builds the hierarchy from the individual elements by the progressively merging clusters Again we have six elements a b c d e and f The first step is to determine which elements should be merged into one cluster We usually prefer to take two closest elements therefore we must define a distance d element1 element2 between elements Suppose we have merged two closest elements b and c Now we have the following clusters a b c d e and f and want to merge them further But to do that we need to take the distance between a and b c and therefore define
12. s the Graphics Interchange Format JPEG Joint Photographic Experts Group image format PNG W3C Portable Network Graphics image format TIFF Tag Image File Format WMF Windows metafile image format O O O O O O O 0 2 2 2 SOM e Train b This command runs main algorithm for Kohonen networks training e Parameters User can set some training parameters by the dialog bellow Document Self organizing maps Page 5 17 Fn Kohonen Network Size Parameters Number of iteration Neighbourhood Shape of neighbourhood hexagonal Alpha 0 5 0 2 Cancel o Kohonen Network Size number of neurons in X and Y axis o Number of Iteration this parameter determines how many iterations the algorithm will execute The term iteration means one reading of vector from table and network adaptation to this vector So if the table contains 20 rows and Number of iteration is set to 200 then the network will adapt 10 times for each row o Neighborhood it is initial size of the neighborhood of the Best Matching Unit BMU This will decrease with the running algorithm time If this parameter is set to 0 then the neighborhood will contain only the winner neuron If it is set to 1 then the network will contain neurons which are directly connected with the winner neuron o Shape of neighborhood there are two options square and hexagonal o Alpha determines how much the neighborhood of the BMU w
13. the distance between two clusters Usually the distance between two clusters A and B is defined as following e the maximum distance between elements of each cluster also called complete linkage clustering max d r wy x E A y E Bh e the minimum distance between elements of each cluster also called single linkage clustering minfd z y re A ye Bh e the mean distance between elements of each cluster also called average linkage clustering 1 ma card A card B gt d az y rE yes ocument Self organizing maps age FA Mt KNOCKER 2 2 3 2 K means clustering The k means algorithm assigns each point to the cluster whose center also called centroid is the nearest The center is the average of all the points included in the cluster i e its coordinates are the arithmetic means for all the points in the cluster and for each dimension separately Example The data set has three dimensions and the cluster has two points X x1 x2 x3 and Y y1 y2 y3 Then the centroid Z becomes Z z1 z2 z3 where z1 x1 y1 2 and z2 x2 y2 2 and z3 x3 y3 2 This is the basic structure of the algorithm e Randomly generates k clusters and determines the cluster centers or directly generates k seed points as cluster centers e Assigns each point to the nearest cluster center e Recomputes the new cluster centers e Repeat until some convergence criterion is met usually that the assignment hasn t
14. y the dark gap This result was achieved by unsupervised learning that is without human intervention Teaching a SOM and representing it with the U matrix offers a fast way to get insight of the data distribution Document Self organizing maps Page 11 17 FA 3 SOM Tutorial How to create Self Organizing map 3 1 Prepare your data First you need some data You must prepare data that will be used for the map creation SOM algorithm needs set of vectors so you could save vectors to a table All elements of vectors must be defined If table contains columns with non numerical data or data that isn t needed for training never mind You can choose columns used for training in Set Columns dialog 3 2 Add your data as version in main application In the main application you can create a new version of data which will be used for training You can create a new version from a table in the database or from a file Some sample data are placed in the file SOM_WDI cvs E New version name SOM data Data from database Table name SOM_WDI_1999 urtaxdb C file File name Erawse Create Close 3 3 Load SOM module if not loaded into application You can find SOM in menu Methods of the main application window Methods Neural network ME If there is no SOM method loaded add it to the list of methods by choosing a command Methods gt Methods as shown bellow Click Add
Download Pdf Manuals
Related Search