Search Results

Search found 59554 results on 2383 pages for 'distributed data mining'.

Page 1/2383 | 1 2 3 4 5 6 7 8 9 10 11 12  | Next Page >

• SQLAuthority News – Links to Book On Line – Data Mining Algorithms (Analysis Services – Data Mining)

  - by pinaldave

  I have quite often received request for the Data Mining Algorithms details. Book Online has wonderful resources for the same. I suggest to read them here. Data Mining Algorithms (Analysis Services – Data Mining) The data mining algorithm is the mechanism that creates a data mining model. To create a model, an algorithm first analyzes a set [...]

  Read the article

• Using a "white list" for extracting terms for Text Mining

  - by [email protected]

  In Part 1 of my post on "Generating cluster names from a document clustering model" (part 1, part 2, part 3), I showed how to build a clustering model from text documents using Oracle Data Miner, which automates preparing data for text mining. In this process we specified a custom stoplist and lexer and relied on Oracle Text to identify important terms.  However, there is an alternative approach, the white list, which uses a thesaurus object with the Oracle Text CTXRULE index to allow you to specify the important terms. INTRODUCTIONA stoplist is used to exclude, i.e., black list, specific words in your documents from being indexed. For example, words like a, if, and, or, and but normally add no value when text mining. Other words can also be excluded if they do not help to differentiate documents, e.g., the word Oracle is ubiquitous in the Oracle product literature. One problem with stoplists is determining which words to specify. This usually requires inspecting the terms that are extracted, manually identifying which ones you don't want, and then re-indexing the documents to determine if you missed any. Since a corpus of documents could contain thousands of words, this could be a tedious exercise. Moreover, since every word is considered as an individual token, a term excluded in one context may be needed to help identify a term in another context. For example, in our Oracle product literature example, the words "Oracle Data Mining" taken individually are not particular helpful. The term "Oracle" may be found in nearly all documents, as with the term "Data." The term "Mining" is more unique, but could also refer to the Mining industry. If we exclude "Oracle" and "Data" by specifying them in the stoplist, we lose valuable information. But it we include them, they may introduce too much noise. Still, when you have a broad vocabulary or don't have a list of specific terms of interest, you rely on the text engine to identify important terms, often by computing the term frequency - inverse document frequency metric. (This is effectively a weight associated with each term indicating its relative importance in a document within a collection of documents. We'll revisit this later.) The results using this technique is often quite valuable. As noted above, an alternative to the subtractive nature of the stoplist is to specify a white list, or a list of terms--perhaps multi-word--that we want to extract and use for data mining. The obvious downside to this approach is the need to specify the set of terms of interest. However, this may not be as daunting a task as it seems. For example, in a given domain (Oracle product literature), there is often a recognized glossary, or a list of keywords and phrases (Oracle product names, industry names, product categories, etc.). Being able to identify multi-word terms, e.g., "Oracle Data Mining" or "Customer Relationship Management" as a single token can greatly increase the quality of the data mining results. The remainder of this post and subsequent posts will focus on how to produce a dataset that contains white list terms, suitable for mining. CREATING A WHITE LIST We'll leverage the thesaurus capability of Oracle Text. Using a thesaurus, we create a set of rules that are in effect our mapping from single and multi-word terms to the tokens used to represent those terms. For example, "Oracle Data Mining" becomes "ORACLEDATAMINING." First, we'll create and populate a mapping table called my_term_token_map. All text has been converted to upper case and values in the TERM column are intended to be mapped to the token in the TOKEN column. TERM                                TOKEN DATA MINING                         DATAMINING ORACLE DATA MINING                  ORACLEDATAMINING 11G                                 ORACLE11G JAVA                                JAVA CRM                                 CRM CUSTOMER RELATIONSHIP MANAGEMENT    CRM ... Next, we'll create a thesaurus object my_thesaurus and a rules table my_thesaurus_rules: CTX_THES.CREATE_THESAURUS('my_thesaurus', FALSE); CREATE TABLE my_thesaurus_rules (main_term     VARCHAR2(100),                                  query_string  VARCHAR2(400)); We next populate the thesaurus object and rules table using the term token map. A cursor is defined over my_term_token_map. As we iterate over  the rows, we insert a synonym relationship 'SYN' into the thesaurus. We also insert into the table my_thesaurus_rules the main term, and the corresponding query string, which specifies synonyms for the token in the thesaurus. DECLARE   cursor c2 is     select token, term     from my_term_token_map; BEGIN   for r_c2 in c2 loop     CTX_THES.CREATE_RELATION('my_thesaurus',r_c2.token,'SYN',r_c2.term);     EXECUTE IMMEDIATE 'insert into my_thesaurus_rules values                        (:1,''SYN(' || r_c2.token || ', my_thesaurus)'')'     using r_c2.token;   end loop; END; We are effectively inserting the token to return and the corresponding query that will look up synonyms in our thesaurus into the my_thesaurus_rules table, for example:     'ORACLEDATAMINING'        SYN ('ORACLEDATAMINING', my_thesaurus)At this point, we create a CTXRULE index on the my_thesaurus_rules table: create index my_thesaurus_rules_idx on        my_thesaurus_rules(query_string)        indextype is ctxsys.ctxrule; In my next post, this index will be used to extract the tokens that match each of the rules specified. We'll then compute the tf-idf weights for each of the terms and create a nested table suitable for mining.

  Read the article

• Data Quality and Master Data Management Resources

  - by Dejan Sarka

  Many companies or organizations do regular data cleansing. When you cleanse the data, the data quality goes up to some higher level. The data quality level is determined by the amount of work invested in the cleansing. As time passes, the data quality deteriorates, and you need to repeat the cleansing process. If you spend an equal amount of effort as you did with the previous cleansing, you can expect the same level of data quality as you had after the previous cleansing. And then the data quality deteriorates over time again, and the cleansing process starts over and over again. The idea of Data Quality Services is to mitigate the cleansing process. While the amount of time you need to spend on cleansing decreases, you will achieve higher and higher levels of data quality. While cleansing, you learn what types of errors to expect, discover error patterns, find domains of correct values, etc. You don’t throw away this knowledge. You store it and use it to find and correct the same issues automatically during your next cleansing process. The following figure shows this graphically. The idea of master data management, which you can perform with Master Data Services (MDS), is to prevent data quality from deteriorating. Once you reach a particular quality level, the MDS application—together with the defined policies, people, and master data management processes—allow you to maintain this level permanently. This idea is shown in the following picture. OK, now you know what DQS and MDS are about. You can imagine the importance on maintaining the data quality. Here are some resources that help you preparing and executing the data quality (DQ) and master data management (MDM) activities. Books Dejan Sarka and Davide Mauri: Data Quality and Master Data Management with Microsoft SQL Server 2008 R2 – a general introduction to MDM, MDS, and data profiling. Matching explained in depth. Dejan Sarka, Matija Lah and Grega Jerkic: MCTS Self-Paced Training Kit (Exam 70-463): Building Data Warehouses with Microsoft SQL Server 2012 – I wrote quite a few chapters about DQ and MDM, and introduced also SQL Server 2012 DQS. Thomas Redman: Data Quality: The Field Guide – you should start with this book. Thomas Redman is the father of DQ and MDM. Tyler Graham: Microsoft SQL Server 2012 Master Data Services – MDS in depth from a product team mate. Arkady Maydanchik: Data Quality Assessment – data profiling in depth. Tamraparni Dasu, Theodore Johnson: Exploratory Data Mining and Data Cleaning – advanced data profiling with data mining. Forthcoming presentations I am presenting a DQS and MDM seminar at PASS SQL Rally Amsterdam 2013: Wednesday, November 6th, 2013: Enterprise Information Management with SQL Server 2012 – a good kick start to your first DQ and / or MDM project. Courses Data Quality and Master Data Management with SQL Server 2012 – I wrote a 2-day course for SolidQ. If you are interested in this course, which I could also deliver in a shorter seminar way, you can contact your closes SolidQ subsidiary, or, of course, me directly on addresses [email protected] or [email protected]. This course could also complement the existing courseware portfolio of training providers, which are welcome to contact me as well. Start improving the quality of your data now!

  Read the article

• Using a "white list" for extracting terms for Text Mining, Part 2

  - by [email protected]

  In my last post, we set the groundwork for extracting specific tokens from a white list using a CTXRULE index. In this post, we will populate a table with the extracted tokens and produce a case table suitable for clustering with Oracle Data Mining. Our corpus of documents will be stored in a database table that is defined as create table documents(id NUMBER, text VARCHAR2(4000)); However, any suitable Oracle Text-accepted data type can be used for the text. We then create a table to contain the extracted tokens. The id column contains the unique identifier (or case id) of the document. The token column contains the extracted token. Note that a given document many have many tokens, so there will be one row per token for a given document. create table extracted_tokens (id NUMBER, token VARCHAR2(4000)); The next step is to iterate over the documents and extract the matching tokens using the index and insert them into our token table. We use the MATCHES function for matching the query_string from my_thesaurus_rules with the text. DECLARE     cursor c2 is       select id, text       from documents; BEGIN     for r_c2 in c2 loop        insert into extracted_tokens          select r_c2.id id, main_term token          from my_thesaurus_rules          where matches(query_string,                        r_c2.text)>0;     end loop; END; Now that we have the tokens, we can compute the term frequency - inverse document frequency (TF-IDF) for each token of each document. create table extracted_tokens_tfidf as   with num_docs as (select count(distinct id) doc_cnt                     from extracted_tokens),        tf       as (select a.id, a.token,                            a.token_cnt/b.num_tokens token_freq                     from                        (select id, token, count(*) token_cnt                        from extracted_tokens                        group by id, token) a,                       (select id, count(*) num_tokens                        from extracted_tokens                        group by id) b                     where a.id=b.id),        doc_freq as (select token, count(*) overall_token_cnt                     from extracted_tokens                     group by token)   select tf.id, tf.token,          token_freq * ln(doc_cnt/df.overall_token_cnt) tf_idf   from num_docs,        tf,        doc_freq df   where df.token=tf.token; From the WITH clause, the num_docs query simply counts the number of documents in the corpus. The tf query computes the term (token) frequency by computing the number of times each token appears in a document and divides that by the number of tokens found in the document. The doc_req query counts the number of times the token appears overall in the corpus. In the SELECT clause, we compute the tf_idf. Next, we create the nested table required to produce one record per case, where a case corresponds to an individual document. Here, we COLLECT all the tokens for a given document into the nested column extracted_tokens_tfidf_1. CREATE TABLE extracted_tokens_tfidf_nt              NESTED TABLE extracted_tokens_tfidf_1                  STORE AS extracted_tokens_tfidf_tab AS              select id,                     cast(collect(DM_NESTED_NUMERICAL(token,tf_idf)) as DM_NESTED_NUMERICALS) extracted_tokens_tfidf_1              from extracted_tokens_tfidf              group by id;   To build the clustering model, we create a settings table and then insert the various settings. Most notable are the number of clusters (20), using cosine distance which is better for text, turning off auto data preparation since the values are ready for mining, the number of iterations (20) to get a better model, and the split criterion of size for clusters that are roughly balanced in number of cases assigned. CREATE TABLE km_settings (setting_name  VARCHAR2(30), setting_value VARCHAR2(30)); BEGIN  INSERT INTO km_settings (setting_name, setting_value) VALUES     VALUES (dbms_data_mining.clus_num_clusters, 20);  INSERT INTO km_settings (setting_name, setting_value)     VALUES (dbms_data_mining.kmns_distance, dbms_data_mining.kmns_cosine);   INSERT INTO km_settings (setting_name, setting_value) VALUES     VALUES (dbms_data_mining.prep_auto,dbms_data_mining.prep_auto_off);   INSERT INTO km_settings (setting_name, setting_value) VALUES     VALUES (dbms_data_mining.kmns_iterations,20);   INSERT INTO km_settings (setting_name, setting_value) VALUES     VALUES (dbms_data_mining.kmns_split_criterion,dbms_data_mining.kmns_size);   COMMIT; END; With this in place, we can now build the clustering model. BEGIN     DBMS_DATA_MINING.CREATE_MODEL(     model_name          => 'TEXT_CLUSTERING_MODEL',     mining_function     => dbms_data_mining.clustering,     data_table_name     => 'extracted_tokens_tfidf_nt',     case_id_column_name => 'id',     settings_table_name => 'km_settings'); END;To generate cluster names from this model, check out my earlier post on that topic.

  Read the article

• Data Mining Resources

  - by Dejan Sarka

  There are many different types of analyses, each one with its own pros and cons. Relational reports have a predefined structure, and end users cannot change it. They are simple to use for end users. Reports can use real-time data and snapshots of data to show the state of a report at specific points in time. One of the drawbacks is that report authoring is limited to IT pros and advanced users. Any kind of dynamic restructuring is very limited. If real-time data is used for a report, the report has a negative impact on the performance of the source system. Processing of the reports might be slow because the data comes from relational database management systems, which are not optimized for reporting only. If you create a semantic model of your data, your end users can create ad-hoc report structures. However, the development is more complex because a developer is needed to create these semantic models. For OLAP, you typically use specialized database management systems. You get lightning speed of analyses. End users can use rich and thin clients to interactively change the structure of the report. Typically, they do it graphically. However, the development of an OLAP system is many times quite complex. It involves the preparation and maintenance of an enterprise data warehouse and OLAP cubes. In order to exploit the possibility of real-time restructuring of reports, the users must be both active and educated. The data is usually stale, as it is loaded into data warehouses and OLAP cubes with a scheduled process. With data mining, a structure is not selected in advance; it searches for the structure. As a result, data mining can give you the most valuable results because you can discover patterns you did not expect. A data mining model structure is limited only by the attributes that you use to train the model. One of the drawbacks is that a lot of knowledge is needed for a successful data mining project. End users have to understand the results. Subject matter experts and IT professionals need to understand business problem thoroughly. The development might be sometimes even more complex than the development of OLAP cubes. Each type of analysis has its own place in an enterprise system. SQL Server has tools for all kinds of analyses. However, data mining is the most advanced way of analyzing the data; this is the “I” in BI. In order to get the most out of it, you need to learn quite a lot. In this blog post, I am gathering together resources for learning, including forthcoming events. Books Multiple authors: SQL Server MVP Deep Dives – I wrote an introductory data mining chapter there. Erik Veerman, Teo Lachev and Dejan Sarka: MCTS Self-Paced Training Kit (Exam 70-448): Microsoft SQL Server 2008 - Business Intelligence Development and Maintenance – you can find a good overview of a complete BI solution, including data mining, in this book. Jamie MacLennan, ZhaoHui Tang, and Bogdan Crivat: Data Mining with Microsoft SQL Server 2008 – can’t miss this book if you want to mine your data with SQL Server tools. Michael Berry, Gordon Linoff: Mastering Data Mining: The Art and Science of Customer Relationship Management – data mining from both, business and technical perspective. Dorian Pyle: Data Preparation for Data Mining – an in-depth book about data preparation. Thomas and Ronald Wonnacott: Introductory Statistics – if you thought that you could get away without statistics, then you are not serious about data mining. Jiawei Han and Micheline Kamber: Data Mining Concepts and Techniques – in-depth explanation of the most popular data mining algorithms. Michael Berry and Gordon Linoff: Data Mining Techniques – another book that explains data mining algorithms, more fro a business perspective. Paolo Guidici: Applied Data Mining – very mathematical book, only if you enjoy statistics and mathematics in general. Forthcoming presentations I am presenting two data mining related sessions during the PASS Summit in Charlotte, NC: Wednesday, October 16th, 2013 - Fraud Detection: Notes from the Field – I am showing how to use data mining for a specific business problem. The presentation is based on real-life projects. Friday, October 18th: Excel 2013 Advanced Analytics – I am focusing on Excel Data Mining Add-ins, and how to use them together with Power Pivot and other add-ins. This is the most you can get out of Excel. Sinergija 2013, Belgrade, Serbia Tuesday, October 22nd: Excel 2013 Analytics to the Max – another presentation focusing on the most advanced analytics you can get in Excel. SQL Rally Amsterdam, Netherlands Thursday, November 7th: Advanced Analytics in Excel 2013 – and again I am presenting about data mining in Excel. Why three different titles for the same presentation? I don’t know, I guess I forgot the name I proposed every time right after I sent the proposal. Courses Data Mining with SQL Server 2012 – I wrote a 3-day course for SolidQ. If you are interested in this course, which I could also deliver in a shorter seminar way, you can contact your closes SolidQ subsidiary, or, of course, me directly on addresses [email protected] or [email protected]. This course could also complement the existing courseware portfolio of training providers, which are welcome to contact me as well. OK, now you know: no more excuses, start learning data mining, get the most out of your data

  Read the article

• Deploying Data Mining Models using Model Export and Import

  - by [email protected]

  In this post, we'll take a look at how Oracle Data Mining facilitates model deployment. After building and testing models, a next step is often putting your data mining model into a production system -- referred to as model deployment. The ability to move data mining model(s) easily into a production system can greatly speed model deployment, and reduce the overall cost. Since Oracle Data Mining provides models as first class database objects, models can be manipulated using familiar database techniques and technology. For example, one or more models can be exported to a flat file, similar to a database table dump file (.dmp). This file can be moved to a different instance of Oracle Database EE, and then imported. All methods for exporting and importing models are based on Oracle Data Pump technology and found in the DBMS_DATA_MINING package. Before performing the actual export or import, a directory object must be created. A directory object is a logical name in the database for a physical directory on the host computer. Read/write access to a directory object is necessary to access the host computer file system from within Oracle Database. For our example, we'll work in the DMUSER schema. First, DMUSER requires the privilege to create any directory. This is often granted through the sysdba account. grant create any directory to dmuser; Now, DMUSER can create the directory object specifying the path where the exported model file (.dmp) should be placed. In this case, on a linux machine, we have the directory /scratch/oracle. CREATE OR REPLACE DIRECTORY dmdir AS '/scratch/oracle'; If you aren't sure of the exact name of the model or models to export, you can find the list of models using the following query: select model_name from user_mining_models; There are several options when exporting models. We can export a single model, multiple models, or all models in a schema using the following procedure calls: BEGIN   DBMS_DATA_MINING.EXPORT_MODEL ('MY_MODEL.dmp','dmdir','name =''MY_DT_MODEL'''); END; BEGIN   DBMS_DATA_MINING.EXPORT_MODEL ('MY_MODELS.dmp','dmdir',              'name IN (''MY_DT_MODEL'',''MY_KM_MODEL'')'); END; BEGIN   DBMS_DATA_MINING.EXPORT_MODEL ('ALL_DMUSER_MODELS.dmp','dmdir'); END; A .dmp file can be imported into another schema or database using the following procedure call, for example: BEGIN   DBMS_DATA_MINING.IMPORT_MODEL('MY_MODELS.dmp', 'dmdir'); END; As with models from any data mining tool, when moving a model from one environment to another, care needs to be taken to ensure the transformations that prepare the data for model building are matched (with appropriate parameters and statistics) in the system where the model is deployed. Oracle Data Mining provides automatic data preparation (ADP) and embedded data preparation (EDP) to reduce, or possibly eliminate, the need to explicitly transport transformations with the model. In the case of ADP, ODM automatically prepares the data and includes the necessary transformations in the model itself. In the case of EDP, users can associate their own transformations with attributes of a model. These transformations are automatically applied when applying the model to data, i.e., scoring. Exporting and importing a model with ADP or EDP results in these transformations being immediately available with the model in the production system.

  Read the article

• Oracle Data Mining a Star Schema: Telco Churn Case Study

  - by charlie.berger

  There is a complete and detailed Telco Churn case study "How to" Blog Series just posted by Ari Mozes, ODM Dev. Manager.  In it, Ari provides detailed guidance in how to leverage various strengths of Oracle Data Mining including the ability to: mine Star Schemas and join tables and views together to obtain a complete 360 degree view of a customer combine transactional data e.g. call record detail (CDR) data, etc. define complex data transformation, model build and model deploy analytical methodologies inside the Database  His blog is posted in a multi-part series.  Below are some opening excerpts for the first 3 blog entries.  This is an excellent resource for any novice to skilled data miner who wants to gain competitive advantage by mining their data inside the Oracle Database.  Many thanks Ari! Mining a Star Schema: Telco Churn Case Study (1 of 3) One of the strengths of Oracle Data Mining is the ability to mine star schemas with minimal effort.  Star schemas are commonly used in relational databases, and they often contain rich data with interesting patterns.  While dimension tables may contain interesting demographics, fact tables will often contain user behavior, such as phone usage or purchase patterns.  Both of these aspects - demographics and usage patterns - can provide insight into behavior.Churn is a critical problem in the telecommunications industry, and companies go to great lengths to reduce the churn of their customer base.  One case study1 describes a telecommunications scenario involving understanding, and identification of, churn, where the underlying data is present in a star schema.  That case study is a good example for demonstrating just how natural it is for Oracle Data Mining to analyze a star schema, so it will be used as the basis for this series of posts...... Mining a Star Schema: Telco Churn Case Study (2 of 3) This post will follow the transformation steps as described in the case study, but will use Oracle SQL as the means for preparing data.  Please see the previous post for background material, including links to the case study and to scripts that can be used to replicate the stages in these posts.1) Handling missing values for call data recordsThe CDR_T table records the number of phone minutes used by a customer per month and per call type (tariff).  For example, the table may contain one record corresponding to the number of peak (call type) minutes in January for a specific customer, and another record associated with international calls in March for the same customer.  This table is likely to be fairly dense (most type-month combinations for a given customer will be present) due to the coarse level of aggregation, but there may be some missing values.  Missing entries may occur for a number of reasons: the customer made no calls of a particular type in a particular month, the customer switched providers during the timeframe, or perhaps there is a data entry problem.  In the first situation, the correct interpretation of a missing entry would be to assume that the number of minutes for the type-month combination is zero.  In the other situations, it is not appropriate to assume zero, but rather derive some representative value to replace the missing entries.  The referenced case study takes the latter approach.  The data is segmented by customer and call type, and within a given customer-call type combination, an average number of minutes is computed and used as a replacement value.In SQL, we need to generate additional rows for the missing entries and populate those rows with appropriate values.  To generate the missing rows, Oracle's partition outer join feature is a perfect fit.  select cust_id, cdre.tariff, cdre.month, minsfrom cdr_t cdr partition by (cust_id) right outer join     (select distinct tariff, month from cdr_t) cdre     on (cdr.month = cdre.month and cdr.tariff = cdre.tariff);   ....... Mining a Star Schema: Telco Churn Case Study (3 of 3) Now that the "difficult" work is complete - preparing the data - we can move to building a predictive model to help identify and understand churn.The case study suggests that separate models be built for different customer segments (high, medium, low, and very low value customer groups).  To reduce the data to a single segment, a filter can be applied: create or replace view churn_data_high asselect * from churn_prep where value_band = 'HIGH'; It is simple to take a quick look at the predictive aspects of the data on a univariate basis.  While this does not capture the more complex multi-variate effects as would occur with the full-blown data mining algorithms, it can give a quick feel as to the predictive aspects of the data as well as validate the data preparation steps.  Oracle Data Mining includes a predictive analytics package which enables quick analysis. begin  dbms_predictive_analytics.explain(   'churn_data_high','churn_m6','expl_churn_tab'); end; /select * from expl_churn_tab where rank <= 5 order by rank; ATTRIBUTE_NAME       ATTRIBUTE_SUBNAME EXPLANATORY_VALUE RANK-------------------- ----------------- ----------------- ----------LOS_BAND                                      .069167052          1MINS_PER_TARIFF_MON  PEAK-5                   .034881648          2REV_PER_MON          REV-5                    .034527798          3DROPPED_CALLS                                 .028110322          4MINS_PER_TARIFF_MON  PEAK-4                   .024698149          5From the above results, it is clear that some predictors do contain information to help identify churn (explanatory value > 0).  The strongest uni-variate predictor of churn appears to be the customer's (binned) length of service.  The second strongest churn indicator appears to be the number of peak minutes used in the most recent month.  The subname column contains the interior piece of the DM_NESTED_NUMERICALS column described in the previous post.  By using the object relational approach, many related predictors are included within a single top-level column. .....   NOTE:  These are just EXCERPTS.  Click here to start reading the Oracle Data Mining a Star Schema: Telco Churn Case Study from the beginning.    

  Read the article

• Deploying Data Mining Models using Model Export and Import, Part 2

  - by [email protected]

  In my last post, Deploying Data Mining Models using Model Export and Import, we explored using DBMS_DATA_MINING.EXPORT_MODEL and DBMS_DATA_MINING.IMPORT_MODEL to enable moving a model from one system to another. In this post, we'll look at two distributed scenarios that make use of this capability and a tip for easily moving models from one machine to another using only Oracle Database, not an external file transport mechanism, such as FTP. The first scenario, consider a company with geographically distributed business units, each collecting and managing their data locally for the products they sell. Each business unit has in-house data analysts that build models to predict which products to recommend to customers in their space. A central telemarketing business unit also uses these models to score new customers locally using data collected over the phone. Since the models recommend different products, each customer is scored using each model. This is depicted in Figure 1.Figure 1: Target instance importing multiple remote models for local scoring In the second scenario, consider multiple hospitals that collect data on patients with certain types of cancer. The data collection is standardized, so each hospital collects the same patient demographic and other health / tumor data, along with the clinical diagnosis. Instead of each hospital building it's own models, the data is pooled at a central data analysis lab where a predictive model is built. Once completed, the model is distributed to hospitals, clinics, and doctor offices who can score patient data locally.Figure 2: Multiple target instances importing the same model from a source instance for local scoring Since this blog focuses on model export and import, we'll only discuss what is necessary to move a model from one database to another. Here, we use the package DBMS_FILE_TRANSFER, which can move files between Oracle databases. The script is fairly straightforward, but requires setting up a database link and directory objects. We saw how to create directory objects in the previous post. To create a database link to the source database from the target, we can use, for example: create database link SOURCE1_LINK connect to <schema> identified by <password> using 'SOURCE1'; Note that 'SOURCE1' refers to the service name of the remote database entry in your tnsnames.ora file. From SQL*Plus, first connect to the remote database and export the model. Note that the model_file_name does not include the .dmp extension. This is because export_model appends "01" to this name.  Next, connect to the local database and invoke DBMS_FILE_TRANSFER.GET_FILE and import the model. Note that "01" is eliminated in the target system file name.  connect <source_schema>/<password>@SOURCE1_LINK; BEGIN  DBMS_DATA_MINING.EXPORT_MODEL ('EXPORT_FILE_NAME' || '.dmp',                                 'MY_SOURCE_DIR_OBJECT',                                 'name =''MY_MINING_MODEL'''); END; connect <target_schema>/<password>; BEGIN  DBMS_FILE_TRANSFER.GET_FILE ('MY_SOURCE_DIR_OBJECT',                               'EXPORT_FILE_NAME' || '01.dmp',                               'SOURCE1_LINK',                               'MY_TARGET_DIR_OBJECT',                               'EXPORT_FILE_NAME' || '.dmp' );  DBMS_DATA_MINING.IMPORT_MODEL ('EXPORT_FILE_NAME' || '.dmp',                                 'MY_TARGET_DIR_OBJECT'); END; To clean up afterward, you may want to drop the exported .dmp file at the source and the transferred file at the target. For example, utl_file.fremove('&directory_name', '&model_file_name' || '.dmp');

  Read the article

• Recommendations for distributed processing/distributed storage systems

  - by Eddie

  At my organization we have a processing and storage system spread across two dozen linux machines that handles over a petabyte of data. The system right now is very ad-hoc; processing automation and data management is handled by a collection of large perl programs on independent machines. I am looking at distributed processing and storage systems to make it easier to maintain, evenly distribute load and data with replication, and grow in disk space and compute power. The system needs to be able to handle millions of files, varying in size between 50 megabytes to 50 gigabytes. Once created, the files will not be appended to, only replaced completely if need be. The files need to be accessible via HTTP for customer download. Right now, processing is automated by perl scripts (that I have complete control over) which call a series of other programs (that I don't have control over because they are closed source) that essentially transforms one data set into another. No data mining happening here. Here is a quick list of things I am looking for: Reliability: These data must be accessible over HTTP about 99% of the time so I need something that does data replication across the cluster. Scalability: I want to be able to add more processing power and storage easily and rebalance the data on across the cluster. Distributed processing: Easy and automatic job scheduling and load balancing that fits with processing workflow I briefly described above. Data location awareness: Not strictly required but desirable. Since data and processing will be on the same set of nodes I would like the job scheduler to schedule jobs on or close to the node that the data is actually on to cut down on network traffic. Here is what I've looked at so far: Storage Management: GlusterFS: Looks really nice and easy to use but doesn't seem to have a way to figure out what node(s) a file actually resides on to supply as a hint to the job scheduler. GPFS: Seems like the gold standard of clustered filesystems. Meets most of my requirements except, like glusterfs, data location awareness. Ceph: Seems way to immature right now. Distributed processing: Sun Grid Engine: I have a lot of experience with this and it's relatively easy to use (once it is configured properly that is). But Oracle got its icy grip around it and it no longer seems very desirable. Both: Hadoop/HDFS: At first glance it looked like hadoop was perfect for my situation. Distributed storage and job scheduling and it was the only thing I found that would give me the data location awareness that I wanted. But I don't like the namename being a single point of failure. Also, I'm not really sure if the MapReduce paradigm fits the type of processing workflow that I have. It seems like you need to write all your software specifically for MapReduce instead of just using Hadoop as a generic job scheduler. OpenStack: I've done some reading on this but I'm having trouble deciding if it fits well with my problem or not. Does anyone have opinions or recommendations for technologies that would fit my problem well? Any suggestions or advise would be greatly appreciated. Thanks!

  Read the article

• Text mining on large database (data mining)

  - by yox

  Hello, I have a large database of resumes (CV), and a certain table skills grouping all users skills. inside that table there's a field skill_text that describes the skill in full text. I'm looking for an algorithm/software/method to extract significant terms/phrases from that table in order to build a new table with standarized skills.. Here are some examples skills extracted from the DB : Sectoral and competitive analysis Business Development (incl. in international settings) Specific structure and road design software - Microstation, Macao, AutoCAD (basic knowledge) Creative work (Photoshop, In-Design, Illustrator) checking and reporting back on campaign progress organising and attending events and exhibitions Development : Aptana Studio, PHP, HTML, CSS, JavaScript, SQL, AJAX Discipline: One to one marketing, E-marketing (SEO & SEA, display, emailing, affiliate program) Mix marketing, Viral Marketing, Social network marketing. The output shoud be something like : Sectoral and competitive analysis Business Development Specific structure and road design software - Macao AutoCAD Photoshop In-Design Illustrator organising events Development Aptana Studio PHP HTML CSS JavaScript SQL AJAX Mix marketing Viral Marketing Social network marketing emailing SEO One to one marketing As you see only skills remains no other representation text. I know this is possible using text mining technics but how to do it ? the database is realy large.. it's a good thing because we can calculate text frequency and decide if it's a real skill or just meaningless text... The big problem is .. how to determin that "blablabla" is a skill ? thanks

  Read the article

• what is the difference between Association rule mining & frequent itemset mining

  - by zeedotcom

  i am new to data mining and confuse about Association rules and frequent item mining. for me i think both are same but i need views from experts on this forum My question is what is the difference between Association rule mining & frequent itemset mining? Thanks

  Read the article

• Building a Redundant / Distributed Application

  - by MattW

  This is more of a "point me in the right direction" question. My team of three and I have built a hosted web app that queues and routes customer chat requests to available customer service agents (It does other things as well, but this is enough background to illustrate the issue). The basic dev architecture today is: a single page ajax web UI (ASP.NET MVC) with floating chat windows (think Gmail) a backend Windows service to queue and route the chat requests this service also logs the chats, calculates service levels, etc a Comet server product that routes data between the web frontend and the backend Windows service this also helps us detect which Agents are still connected (online) And our hardware architecture today is: 2 servers to host the web UI portion of the application a load balancer to route requests to the 2 different web app servers a third server to host the SQL Server DB and the backend Windows service responsible for queuing / delivering chats So as it stands today, one of the web app servers could go down and we would be ok. However, if something would happen to the SQL Server / Windows Service server we would be boned. My question - how can I make this backend Windows service logic be able to be spread across multiple machines (distributed)? The Windows service is written to accept requests from the Comet server, check for available Agents, and route the chat to those agents. How can I make this more distributed? How can I make it so that I can distribute the work of the backend Windows service can be spread across multiple machines for redundancy and uptime purposes? Will I need to re-write it with distributed computing in mind? I should also note that I am hosting all of this on Rackspace Cloud instances - so maybe it is something I should be less concerned about? Thanks in advance for any help!

  Read the article

• To sample or not to sample...

  - by [email protected]

  Ideally, we would know the exact answer to every question. How many people support presidential candidate A vs. B? How many people suffer from H1N1 in a given state? Does this batch of manufactured widgets have any defective parts? Knowing exact answers is expensive in terms of time and money and, in most cases, is impractical if not impossible. Consider asking every person in a region for their candidate preference, testing every person with flu symptoms for H1N1 (assuming every person reported when they had flu symptoms), or destructively testing widgets to determine if they are "good" (leaving no product to sell). Knowing exact answers, fortunately, isn't necessary or even useful in many situations. Understanding the direction of a trend or statistically significant results may be sufficient to answer the underlying question: who is likely to win the election, have we likely reached a critical threshold for flu, or is this batch of widgets good enough to ship? Statistics help us to answer these questions with a certain degree of confidence. This focuses on how we collect data. In data mining, we focus on the use of data, that is data that has already been collected. In some cases, we may have all the data (all purchases made by all customers), in others the data may have been collected using sampling (voters, their demographics and candidate choice). Building data mining models on all of your data can be expensive in terms of time and hardware resources. Consider a company with 40 million customers. Do we need to mine all 40 million customers to get useful data mining models? The quality of models built on all data may be no better than models built on a relatively small sample. Determining how much is a reasonable amount of data involves experimentation. When starting the model building process on large datasets, it is often more efficient to begin with a small sample, perhaps 1000 - 10,000 cases (records) depending on the algorithm, source data, and hardware. This allows you to see quickly what issues might arise with choice of algorithm, algorithm settings, data quality, and need for further data preparation. Instead of waiting for a model on a large dataset to build only to find that the results don't meet expectations, once you are satisfied with the results on the initial sample, you can  take a larger sample to see if model quality improves, and to get a sense of how the algorithm scales to the particular dataset. If model accuracy or quality continues to improve, consider increasing the sample size. Sampling in data mining is also used to produce a held-aside or test dataset for assessing classification and regression model accuracy. Here, we reserve some of the build data (data that includes known target values) to be used for an honest estimate of model error using data the model has not seen before. This sampling transformation is often called a split because the build data is split into two randomly selected sets, often with 60% of the records being used for model building and 40% for testing. Sampling must be performed with care, as it can adversely affect model quality and usability. Even a truly random sample doesn't guarantee that all values are represented in a given attribute. This is particularly troublesome when the attribute with omitted values is the target. A predictive model that has not seen any examples for a particular target value can never predict that target value! For other attributes, values may consist of a single value (a constant attribute) or all unique values (an identifier attribute), each of which may be excluded during mining. Values from categorical predictor attributes that didn't appear in the training data are not used when testing or scoring datasets. In subsequent posts, we'll talk about three sampling techniques using Oracle Database: simple random sampling without replacement, stratified sampling, and simple random sampling with replacement.

  Read the article

• Big Data – Learning Basics of Big Data in 21 Days – Bookmark

  - by Pinal Dave

  Earlier this month I had a great time to write Bascis of Big Data series. This series received great response and lots of good comments I have received, I am going to follow up this basics series with further in-depth series in near future. Here is the consolidated blog post where you can find all the 21 days blog posts together. Bookmark this page for future reference. Big Data – Beginning Big Data – Day 1 of 21 Big Data – What is Big Data – 3 Vs of Big Data – Volume, Velocity and Variety – Day 2 of 21 Big Data – Evolution of Big Data – Day 3 of 21 Big Data – Basics of Big Data Architecture – Day 4 of 21 Big Data – Buzz Words: What is NoSQL – Day 5 of 21 Big Data – Buzz Words: What is Hadoop – Day 6 of 21 Big Data – Buzz Words: What is MapReduce – Day 7 of 21 Big Data – Buzz Words: What is HDFS – Day 8 of 21 Big Data – Buzz Words: Importance of Relational Database in Big Data World – Day 9 of 21 Big Data – Buzz Words: What is NewSQL – Day 10 of 21 Big Data – Role of Cloud Computing in Big Data – Day 11 of 21 Big Data – Operational Databases Supporting Big Data – RDBMS and NoSQL – Day 12 of 21 Big Data – Operational Databases Supporting Big Data – Key-Value Pair Databases and Document Databases – Day 13 of 21 Big Data – Operational Databases Supporting Big Data – Columnar, Graph and Spatial Database – Day 14 of 21 Big Data – Data Mining with Hive – What is Hive? – What is HiveQL (HQL)? – Day 15 of 21 Big Data – Interacting with Hadoop – What is PIG? – What is PIG Latin? – Day 16 of 21 Big Data – Interacting with Hadoop – What is Sqoop? – What is Zookeeper? – Day 17 of 21 Big Data – Basics of Big Data Analytics – Day 18 of 21 Big Data – How to become a Data Scientist and Learn Data Science? – Day 19 of 21 Big Data – Various Learning Resources – How to Start with Big Data? – Day 20 of 21 Big Data – Final Wrap and What Next – Day 21 of 21 Reference: Pinal Dave (http://blog.sqlauthority.com) Filed under: Big Data, PostADay, SQL, SQL Authority, SQL Query, SQL Server, SQL Tips and Tricks, T SQL

  Read the article

• New Communications Industry Data Model with "Factory Installed" Predictive Analytics using Oracle Da

  - by charlie.berger

  Oracle Introduces Oracle Communications Data Model to Provide Actionable Insight for Communications Service Providers   We've integrated pre-installed analytical methodologies with the new Oracle Communications Data Model to deliver automated, simple, yet powerful predictive analytics solutions for customers.  Churn, sentiment analysis, identifying customer segments - all things that can be anticipated and hence, preconcieved and implemented inside an applications.  Read on for more information! TM Forum Management World, Nice, France - 18 May 2010 News Facts To help communications service providers (CSPs) manage and analyze rapidly growing data volumes cost effectively, Oracle today introduced the Oracle Communications Data Model. With the Oracle Communications Data Model, CSPs can achieve rapid time to value by quickly implementing a standards-based enterprise data warehouse that features communications industry-specific reporting, analytics and data mining. The combination of the Oracle Communications Data Model, Oracle Exadata and the Oracle Business Intelligence (BI) Foundation represents the most comprehensive data warehouse and BI solution for the communications industry. Also announced today, Hong Kong Broadband Network enhanced their data warehouse system, going live on Oracle Communications Data Model in three months. The leading provider increased its subscriber base by 37 percent in six months and reduced customer churn to less than one percent. Product Details Oracle Communications Data Model provides industry-specific schema and embedded analytics that address key areas such as customer management, marketing segmentation, product development and network health. CSPs can efficiently capture and monitor critical data and transform it into actionable information to support development and delivery of next-generation services using: More than 1,300 industry-specific measurements and key performance indicators (KPIs) such as network reliability statistics, provisioning metrics and customer churn propensity. Embedded OLAP cubes for extremely fast dimensional analysis of business information. Embedded data mining models for sophisticated trending and predictive analysis. Support for multiple lines of business, such as cable, mobile, wireline and Internet, which can be easily extended to support future requirements. With Oracle Communications Data Model, CSPs can jump start the implementation of a communications data warehouse in line with communications-industry standards including the TM Forum Information Framework (SID), formerly known as the Shared Information Model. Oracle Communications Data Model is optimized for any Oracle Database 11g platform, including Oracle Exadata, which can improve call data record query performance by 10x or more. Supporting Quotes "Oracle Communications Data Model covers a wide range of business areas that are relevant to modern communications service providers and is a comprehensive solution - with its data model and pre-packaged templates including BI dashboards, KPIs, OLAP cubes and mining models. It helps us save a great deal of time in building and implementing a customized data warehouse and enables us to leverage the advanced analytics quickly and more effectively," said Yasuki Hayashi, executive manager, NTT Comware Corporation. "Data volumes will only continue to grow as communications service providers expand next-generation networks, deploy new services and adopt new business models. They will increasingly need efficient, reliable data warehouses to capture key insights on data such as customer value, network value and churn probability. With the Oracle Communications Data Model, Oracle has demonstrated its commitment to meeting these needs by delivering data warehouse tools designed to fill communications industry-specific needs," said Elisabeth Rainge, program director, Network Software, IDC. "The TM Forum Conformance Mark provides reassurance to customers seeking standards-based, and therefore, cost-effective and flexible solutions. TM Forum is extremely pleased to work with Oracle to certify its Oracle Communications Data Model solution. Upon successful completion, this certification will represent the broadest and most complete implementation of the TM Forum Information Framework to date, with more than 130 aggregate business entities," said Keith Willetts, chairman and chief executive officer, TM Forum. Supporting Resources Oracle Communications Oracle Communications Data Model Data Sheet Oracle Communications Data Model Podcast Oracle Data Warehousing Oracle Communications on YouTube Oracle Communications on Delicious Oracle Communications on Facebook Oracle Communications on Twitter Oracle Communications on LinkedIn Oracle Database on Twitter The Data Warehouse Insider Blog

  Read the article

• Kipróbálható az ingyenes új Oracle Data Miner 11gR2 grafikus workflow-val

  - by Fekete Zoltán

  Oracle Data Mining technológiai információs oldal. Oracle Data Miner 11g Release 2 - Early Adopter oldal. Megjelent, letöltheto és kipróbálható az Oracle Data Mining, az Oracle adatbányászat új grafikus felülete, az Oracle Data Miner 11gR2. Az Oracle Data Minerhez egyszeruen az SQL Developer-t kell letöltenünk, mivel az adatbányászati felület abból indítható. Az Oracle Data Mining az Oracle adatbáziskezelobe ágyazott adatbányászati motor, ami az Oracle Database Enterprise Edition opciója. Az adatbányászat az adattárházak elemzésének kifinomult eszköze és folyamata. Az Oracle Data Mining in-database-mining elonyeit felvonultatja: - nincs felesleges adatmozgatás, a teljes adatbányászati folyamatban az adatbázisban maradnak az adatok - az adatbányászati modellek is az Oracle adatbázisban vannak - az adatbányászati eredmények, cluster adatok, döntések, valószínuségek, stb. szintén az adatbázisban keletkeznek, és ott közvetlenül elemezhetoek Az új ingyenes Data Miner felület "hatalmas gazdagodáson" ment keresztül az elozo verzióhoz képest. - grafikus adatbányászati workflow szerkesztés és futtatás jelent meg! - továbbra is ingyenes - kibovült a felület - új elemzési lehetoségekkel bovült - az SQL Developer 3.0 felületrol indítható, ez megkönnyíti az adatbányászati funkciók meghívását az adatbázisból, ha épp nem a grafikus felületetet szeretnénk erre használni Az ingyenes Data Miner felület az Oracle SQL Developer kiterjesztéseként érheto el, így az elemzok közvetlenül dolgozhatnak az adatokkal az adatbázisban és a Data Miner grafikus felülettel is, építhetnek és kiértékelhetnek, futtathatnak modelleket, predikciókat tehetnek és elemezhetnek, támogatást kapva az adatbányászati módszertan megvalósítására. A korábbi Oracle Data Miner felület a Data Miner Classic néven fut és továbbra is letöltheto az OTN-rol. Az új Data Miner GUI-ból egy képernyokép: Milyen feladatokra ad megoldási lehetoséget az Oracle Data Mining: - ügyfél viselkedés megjövendölése, prediktálása - a "legjobb" ügyfelek eredményes megcélzása - ügyfél megtartás, elvándorlás kezelés (churn) - ügyfél szegmensek, klaszterek, profilok keresése és vizsgálata - anomáliák, visszaélések felderítése - stb.

  Read the article

• Integrating Data Mining into your BI Solution (Presentation)

  I recently gave a live meeting presentation to the UK User Group on Integrating Data Mining into your BI Solution.  In it I talk about and demo ways of using your data mining models inside Integration Services, Analysis Services and Reporting Services.  This is the first in a series of presentations I will be doing for the UG as I try to get the word out that Data Mining can be for the masses. You can download my deck and my line meeting recording from here.

  Read the article

• Nagy dobás készül az Oracle adatányászati felületen, Oracle Data Mining

  - by Fekete Zoltán

  Ahogyan már a tavaly oszi Oracle OpenWorld hírekben és eloadásokban is láthattuk a beharangozót, az Oracle nagy dobásra készül az adatbányászati fronton (Oracle Data Mining), mégpedig a remekül használható adatbányászati motor grafikus felületének a kiterjesztésével. Ha jól megfigyeljük ezt az utóbbi linket, az eddigi grafikus felület már Oracle Data Miner Classic néven fut. Hogyan is lehet használni az Oracle Data Mining-ot? - Oracle Data Miner (ingyenesen letöltheto GUI az OTN-rol) - Java-ból és PL/SQL-bol, Oracle Data Mining JDeveloper and SQL Developer Extensions - Excel felületrol, Oracle Spreadsheet Add-In for Predictive Analytics - ODM Connector for mySAP BW Oracle Data Mining technikai információ.

  Read the article

• Integrating Data Mining into your BI Solution (Presentation)

  I recently gave a live meeting presentation to the UK User Group on Integrating Data Mining into your BI Solution.  In it I talk about and demo ways of using your data mining models inside Integration Services, Analysis Services and Reporting Services.  This is the first in a series of presentations I will be doing for the UG as I try to get the word out that Data Mining can be for the masses. You can download my deck and my line meeting recording from here.

  Read the article

• Java - System design with distributed Queues and Locks

  - by sunny

  Looking for inputs to evaluate a design for a system (java) which would have a distributed queue serving several (but not too many) nodes. These nodes would process objects present in the distributed queue and on occasion require a distributed lock across the cluster on an arbitrary (distributed) data structures. These (distributed) data structures could potentially lie in a distributed cache. Eliminating Terracotta (DSO),Hazelcast and Akka what could be alternative choices. Currently considering zookeeper as a distributed locking mechanism. Since the recommendation of a znode is not to exceed the 1M size , the understanding is that zookeeper should not be used a distributed queue. And also from Netflix curator tech note 4. So should a distributed cache, say like memcached, or redis be used to emulate a distributed queue ? i.e. The distributed queue will be stored in the caches and will be locked cluster-wide via zookeeper. Are there potential pitfalls with this high-level approach. The objects don't need to be taken off the queue. The object will pass through a lifecycle which will determine its removal from the queue. There would be about 10k+ objects in a queue at a given time changing states and any node could service one stage of the object's lifecycle. (Although not strictly necessary .. i.e. one node could serve the entire lifecycle if that is more efficient.) Any suggestions/alternatives ? sidenote: new to zookeeper ; redis etc.

  Read the article

• Big Data&rsquo;s Killer App&hellip;

  - by jean-pierre.dijcks

  Recently Keith spent  some time talking about the cloud on this blog and I will spare you my thoughts on the whole thing. What I do want to write down is something about the Big Data movement and what I think is the killer app for Big Data... Where is this coming from, ok, I confess... I spent 3 days in cloud land at the Cloud Connect conference in Santa Clara and it was quite a lot of fun. One of the nice things at Cloud Connect was that there was a track dedicated to Big Data, which prompted me to some extend to write this post. What is Big Data anyways? The most valuable point made in the Big Data track was that Big Data in itself is not very cool. Doing something with Big Data is what makes all of this cool and interesting to a business user! The other good insight I got was that a lot of people think Big Data means a single gigantic monolithic system holding gazillions of bytes or documents or log files. Well turns out that most people in the Big Data track are talking about a lot of collections of smaller data sets. So rather than thinking "big = monolithic" you should be thinking "big = many data sets". This is more than just theoretical, it is actually relevant when thinking about big data and how to process it. It is important because it means that the platform that stores data will most likely consist out of multiple solutions. You may be storing logs on something like HDFS, you may store your customer information in Oracle and you may store distilled clickstream information in some distilled form in MySQL. The big question you will need to solve is not what lives where, but how to get it all together and get some value out of all that data. NoSQL and MapReduce Nope, sorry, this is not the killer app... and no I'm not saying this because my business card says Oracle and I'm therefore biased. I think language is important, but as with storage I think pragmatic is better. In other words, some questions can be answered with SQL very efficiently, others can be answered with PERL or TCL others with MR. History should teach us that anyone trying to solve a problem will use any and all tools around. For example, most data warehouses (Big Data 1.0?) get a lot of data in flat files. Everyone then runs a bunch of shell scripts to massage or verify those files and then shoves those files into the database. We've even built shell script support into external tables to allow for this. I think the Big Data projects will do the same. Some people will use MapReduce, although I would argue that things like Cascading are more interesting, some people will use Java. Some data is stored on HDFS making Cascading the way to go, some data is stored in Oracle and SQL does do a good job there. As with storage and with history, be pragmatic and use what fits and neither NoSQL nor MR will be the one and only. Also, a language, while important, does in itself not deliver business value. So while cool it is not a killer app... Vertical Behavioral Analytics This is the killer app! And you are now thinking: "what does that mean?" Let's decompose that heading. First of all, analytics. I would think you had guessed by now that this is really what I'm after, and of course you are right. But not just analytics, which has a very large scope and means many things to many people. I'm not just after Business Intelligence (analytics 1.0?) or data mining (analytics 2.0?) but I'm after something more interesting that you can only do after collecting large volumes of specific data. That all important data is about behavior. What do my customers do? More importantly why do they behave like that? If you can figure that out, you can tailor web sites, stores, products etc. to that behavior and figure out how to be successful. Today's behavior that is somewhat easily tracked is web site clicks, search patterns and all of those things that a web site or web server tracks. that is where the Big Data lives and where these patters are now emerging. Other examples however are emerging, and one of the examples used at the conference was about prediction churn for a telco based on the social network its members are a part of. That social network is not about LinkedIn or Facebook, but about who calls whom. I call you a lot, you switch provider, and I might/will switch too. And that just naturally brings me to the next word, vertical. Vertical in this context means per industry, e.g. communications or retail or government or any other vertical. The reason for being more specific than just behavioral analytics is that each industry has its own data sources, has its own quirky logic and has its own demands and priorities. Of course, the methods and some of the software will be common and some will have both retail and service industry analytics in place (your corner coffee store for example). But the gist of it all is that analytics that can predict customer behavior for a specific focused group of people in a specific industry is what makes Big Data interesting. Building a Vertical Behavioral Analysis System Well, that is going to be interesting. I have not seen much going on in that space and if I had to have some criticism on the cloud connect conference it would be the lack of concrete user cases on big data. The telco example, while a step into the vertical behavioral part is not really on big data. It used a sample of data from the customers' data warehouse. One thing I do think, and this is where I think parts of the NoSQL stuff come from, is that we will be doing this analysis where the data is. Over the past 10 years we at Oracle have called this in-database analytics. I guess we were (too) early? Now the entire market is going there including companies like SAS. In-place btw does not mean "no data movement at all", what it means that you will do this on data's permanent home. For SAS that is kind of the current problem. Most of the inputs live in a data warehouse. So why move it into SAS and back? That all worked with 1 TB data warehouses, but when we are looking at 100TB to 500 TB of distilled data... Comments? As it is still early days with these systems, I'm very interested in seeing reactions and thoughts to some of these thoughts...

  Read the article

• Fraud Detection with the SQL Server Suite Part 2

  - by Dejan Sarka

  This is the second part of the fraud detection whitepaper. You can find the first part in my previous blog post about this topic. My Approach to Data Mining Projects It is impossible to evaluate the time and money needed for a complete fraud detection infrastructure in advance. Personally, I do not know the customer’s data in advance. I don’t know whether there is already an existing infrastructure, like a data warehouse, in place, or whether we would need to build one from scratch. Therefore, I always suggest to start with a proof-of-concept (POC) project. A POC takes something between 5 and 10 working days, and involves personnel from the customer’s site – either employees or outsourced consultants. The team should include a subject matter expert (SME) and at least one information technology (IT) expert. The SME must be familiar with both the domain in question as well as the meaning of data at hand, while the IT expert should be familiar with the structure of data, how to access it, and have some programming (preferably Transact-SQL) knowledge. With more than one IT expert the most time consuming work, namely data preparation and overview, can be completed sooner. I assume that the relevant data is already extracted and available at the very beginning of the POC project. If a customer wants to have their people involved in the project directly and requests the transfer of knowledge, the project begins with training. I strongly advise this approach as it offers the establishment of a common background for all people involved, the understanding of how the algorithms work and the understanding of how the results should be interpreted, a way of becoming familiar with the SQL Server suite, and more. Once the data has been extracted, the customer’s SME (i.e. the analyst), and the IT expert assigned to the project will learn how to prepare the data in an efficient manner. Together with me, knowledge and expertise allow us to focus immediately on the most interesting attributes and identify any additional, calculated, ones soon after. By employing our programming knowledge, we can, for example, prepare tens of derived variables, detect outliers, identify the relationships between pairs of input variables, and more, in only two or three days, depending on the quantity and the quality of input data. I favor the customer’s decision of assigning additional personnel to the project. For example, I actually prefer to work with two teams simultaneously. I demonstrate and explain the subject matter by applying techniques directly on the data managed by each team, and then both teams continue to work on the data overview and data preparation under our supervision. I explain to the teams what kind of results we expect, the reasons why they are needed, and how to achieve them. Afterwards we review and explain the results, and continue with new instructions, until we resolve all known problems. Simultaneously with the data preparation the data overview is performed. The logic behind this task is the same – again I show to the teams involved the expected results, how to achieve them and what they mean. This is also done in multiple cycles as is the case with data preparation, because, quite frankly, both tasks are completely interleaved. A specific objective of the data overview is of principal importance – it is represented by a simple star schema and a simple OLAP cube that will first of all simplify data discovery and interpretation of the results, and will also prove useful in the following tasks. The presence of the customer’s SME is the key to resolving possible issues with the actual meaning of the data. We can always replace the IT part of the team with another database developer; however, we cannot conduct this kind of a project without the customer’s SME. After the data preparation and when the data overview is available, we begin the scientific part of the project. I assist the team in developing a variety of models, and in interpreting the results. The results are presented graphically, in an intuitive way. While it is possible to interpret the results on the fly, a much more appropriate alternative is possible if the initial training was also performed, because it allows the customer’s personnel to interpret the results by themselves, with only some guidance from me. The models are evaluated immediately by using several different techniques. One of the techniques includes evaluation over time, where we use an OLAP cube. After evaluating the models, we select the most appropriate model to be deployed for a production test; this allows the team to understand the deployment process. There are many possibilities of deploying data mining models into production; at the POC stage, we select the one that can be completed quickly. Typically, this means that we add the mining model as an additional dimension to an existing DW or OLAP cube, or to the OLAP cube developed during the data overview phase. Finally, we spend some time presenting the results of the POC project to the stakeholders and managers. Even from a POC, the customer will receive lots of benefits, all at the sole risk of spending money and time for a single 5 to 10 day project: The customer learns the basic patterns of frauds and fraud detection The customer learns how to do the entire cycle with their own people, only relying on me for the most complex problems The customer’s analysts learn how to perform much more in-depth analyses than they ever thought possible The customer’s IT experts learn how to perform data extraction and preparation much more efficiently than they did before All of the attendees of this training learn how to use their own creativity to implement further improvements of the process and procedures, even after the solution has been deployed to production The POC output for a smaller company or for a subsidiary of a larger company can actually be considered a finished, production-ready solution It is possible to utilize the results of the POC project at subsidiary level, as a finished POC project for the entire enterprise Typically, the project results in several important “side effects” Improved data quality Improved employee job satisfaction, as they are able to proactively contribute to the central knowledge about fraud patterns in the organization Because eventually more minds get to be involved in the enterprise, the company should expect more and better fraud detection patterns After the POC project is completed as described above, the actual project would not need months of engagement from my side. This is possible due to our preference to transfer the knowledge onto the customer’s employees: typically, the customer will use the results of the POC project for some time, and only engage me again to complete the project, or to ask for additional expertise if the complexity of the problem increases significantly. I usually expect to perform the following tasks: Establish the final infrastructure to measure the efficiency of the deployed models Deploy the models in additional scenarios Through reports By including Data Mining Extensions (DMX) queries in OLTP applications to support real-time early warnings Include data mining models as dimensions in OLAP cubes, if this was not done already during the POC project Create smart ETL applications that divert suspicious data for immediate or later inspection I would also offer to investigate how the outcome could be transferred automatically to the central system; for instance, if the POC project was performed in a subsidiary whereas a central system is available as well Of course, for the actual project, I would repeat the data and model preparation as needed It is virtually impossible to tell in advance how much time the deployment would take, before we decide together with customer what exactly the deployment process should cover. Without considering the deployment part, and with the POC project conducted as suggested above (including the transfer of knowledge), the actual project should still only take additional 5 to 10 days. The approximate timeline for the POC project is, as follows: 1-2 days of training 2-3 days for data preparation and data overview 2 days for creating and evaluating the models 1 day for initial preparation of the continuous learning infrastructure 1 day for presentation of the results and discussion of further actions Quite frequently I receive the following question: are we going to find the best possible model during the POC project, or during the actual project? My answer is always quite simple: I do not know. Maybe, if we would spend just one hour more for data preparation, or create just one more model, we could get better patterns and predictions. However, we simply must stop somewhere, and the best possible way to do this, according to my experience, is to restrict the time spent on the project in advance, after an agreement with the customer. You must also never forget that, because we build the complete learning infrastructure and transfer the knowledge, the customer will be capable of doing further investigations independently and improve the models and predictions over time without the need for a constant engagement with me.

  Read the article

• Data Quality Through Data Governance

  Data Quality Governance Data quality is very important to every organization, bad data cost an organization time, money, and resources that could be prevented if the proper governance was put in to place.  Data Governance Program Criteria: Support from Executive Management and all Business Units Data Stewardship Program  Cross Functional Team of Data Stewards Data Governance Committee Quality Structured Data It should go without saying but any successful project in today’s business world must get buy in from executive management and all stakeholders involved with the project. If management does not fully support a project because they see it is in there and the company’s best interest then they will remove/eliminate funding, resources and allocated time to work on the project. In essence they can render a project dead until it is official killed by the business. In addition, buy in from stake holders is also very important because they can cause delays increased spending in time, money and resources because they do not support a project. Data Stewardship programs are administered by a data steward manager who primary focus is to support, train and manage a cross functional data stewards team. A cross functional team of data stewards are pulled from various departments act to ensure that all systems work to ensure that an organization’s goals are achieved. Typically, data stewards are subject matter experts that act as mediators between their respective departments and IT. Data Quality Procedures Data Governance Committees are composed of data stewards, Upper management, IT Leadership and various subject matter experts depending on a company. The primary goal of this committee is to define strategic goals, coordinate activities, set data standards and offer data guidelines for the business. Data Quality Policies In 1997, Claudia Imhoff defined a Data Stewardship’s responsibility as to approve business naming standards, develop consistent data definitions, determine data aliases, develop standard calculations and derivations, document the business rules of the corporation, monitor the quality of the data in the data warehouse, define security requirements, and so forth. She further explains data stewards responsible for creating and enforcing polices on the following but not limited to issues. Resolving Data Integration Issues Determining Data Security Documenting Data Definitions, Calculations, Summarizations, etc. Maintaining/Updating Business Rules Analyzing and Improving Data Quality

  Read the article

• SQL SERVER – Advanced Data Quality Services with Melissa Data – Azure Data Market

  - by pinaldave

  There has been much fanfare over the new SQL Server 2012, and especially around its new companion product Data Quality Services (DQS). Among the many new features is the addition of this integrated knowledge-driven product that enables data stewards everywhere to profile, match, and cleanse data. In addition to the homegrown rules that data stewards can design and implement, there are also connectors to third party providers that are hosted in the Azure Datamarket marketplace.  In this review, I leverage SQL Server 2012 Data Quality Services, and proceed to subscribe to a third party data cleansing product through the Datamarket to showcase this unique capability. Crucial Questions For the purposes of the review, I used a database I had in an Excel spreadsheet with name and address information. Upon a cursory inspection, there are miscellaneous problems with these records; some addresses are missing ZIP codes, others missing a city, and some records are slightly misspelled or have unparsed suites. With DQS, I can easily add a knowledge base to help standardize my values, such as for state abbreviations. But how do I know that my address is correct? And if my address is not correct, what should it be corrected to? The answer lies in a third party knowledge base by the acknowledged USPS certified address accuracy experts at Melissa Data. Reference Data Services Within DQS there is a handy feature to actually add reference data from many different third-party Reference Data Services (RDS) vendors. DQS simplifies the processes of cleansing, standardizing, and enriching data through custom rules and through service providers from the Azure Datamarket. A quick jump over to the Datamarket site shows me that there are a handful of providers that offer data directly through Data Quality Services. Upon subscribing to these services, one can attach a DQS domain or composite domain (fields in a record) to a reference data service provider, and begin using it to cleanse, standardize, and enrich that data. Besides what I am looking for (address correction and enrichment), it is possible to subscribe to a host of other services including geocoding, IP address reference, phone checking and enrichment, as well as name parsing, standardization, and genderization.  These capabilities extend the data quality that DQS has natively by quite a bit. For my current address correction review, I needed to first sign up to a reference data provider on the Azure Data Market site. For this example, I used Melissa Data’s Address Check Service. They offer free one-month trials, so if you wish to follow along, or need to add address quality to your own data, I encourage you to sign up with them. Once I subscribed to the desired Reference Data Provider, I navigated my browser to the Account Keys within My Account to view the generated account key, which I then inserted into the DQS Client – Configuration under the Administration area. Step by Step to Guide That was all it took to hook in the subscribed provider -Melissa Data- directly to my DQS Client. The next step was for me to attach and map in my Reference Data from the newly acquired reference data provider, to a domain in my knowledge base. On the DQS Client home screen, I selected “New Knowledge Base” under Knowledge Base Management on the left-hand side of the home screen. Under New Knowledge Base, I typed a Name and description of my new knowledge base, then proceeded to the Domain Management screen. Here I established a series of domains (fields) and then linked them all together as a composite domain (record set). Using the Create Domain button, I created the following domains according to the fields in my incoming data: Name Address Suite City State Zip I added a Suite column in my domain because Melissa Data has the ability to return missing Suites based on last name or company. And that’s a great benefit of using these third party providers, as they have data that the data steward would not normally have access to. The bottom line is, with these third party data providers, I can actually improve my data. Next, I created a composite domain (fulladdress) and added the (field) domains into the composite domain. This essentially groups our address fields together in a record to facilitate the full address cleansing they perform. I then selected my newly created composite domain and under the Reference Data tab, added my third party reference data provider –Melissa Data’s Address Check- and mapped in each domain that I had to the provider’s Schema. Now that my composite domain has been married to the Reference Data service, I can take the newly published knowledge base and create a project to cleanse and enrich my data. My next task was to create a new Data Quality project, mapping in my data source and matching it to the appropriate domain column, and then kick off the verification process. It took just a few minutes with some progress indicators indicating that it was working. When the process concluded, there was a helpful set of tabs that place the response records into categories: suggested; new; invalid; corrected (automatically); and correct. Accepting the suggestions provided by  Melissa Data allowed me to clean up all the records and flag the invalid ones. It is very apparent that DQS makes address data quality simplistic for any IT professional. Final Note As I have shown, DQS makes data quality very easy. Within minutes I was able to set up a data cleansing and enrichment routine within my data quality project, and ensure that my address data was clean, verified, and standardized against real reference data. As reviewed here, it’s easy to see how both SQL Server 2012 and DQS work to take what used to require a highly skilled developer, and empower an average business or database person to consume external services and clean data. Reference: Pinal Dave (http://blog.sqlauthority.com) Filed under: PostADay, SQL, SQL Authority, SQL Query, SQL Server, SQL Tips and Tricks, SQL Utility, T SQL, Technology Tagged: DQS

  Read the article

• More Value From Data Using Data Mining Presentation

  Here is a presentation I gave at the SQLBits conference in September which was recorded by Microsoft.  Usually I speak about SSIS but on this particular event I thought people would like to hear something different from me. Microsoft are making a big play for making Data Mining more accessible to everyone and not just boffins.  In this presentation I give an overview of data mining and then do some demonstrations using the excellent Excel Add-Ins available from Microsoft SQL Server 2008 SQL Server 2005 I hope you enjoy this presentation http://go.microsoft.com/?linkid=9633764

  Read the article

1 2 3 4 5 6 7 8 9 10 11 12  | Next Page >