Dmitri Kondratiev

Questions and Answers (9) View all

• Answer added in Database Mining
  2 Data Mining Toy Problem
  By Dmitri Kondratiev
  Dmitri Kondratiev

  How about searching in hyperspace? Every mission can be presented as a vector. Value of a component in this vector is a count of diamonds in correspon... [more]

  How about searching in hyperspace? Every mission can be presented as a vector. Value of a component in this vector is a count of diamonds in corresponding point of hyperspace. And every vector component is addressed by a tuple(dd,m_provider_id ,obj_id ,obj_man_id ,reg_id ,obj_freq ,obj_illum_cnt ,obj_probe_cnt ,obj_eat_cnt). After building such vectors from training set we cluster them. Next we build vectors from short history test set. Now, for each test vector find cluster it belongs to. After that using cosine similarity for each test vector find the closest one in his cluster. Will this work?
  Following
• Question asked in Database Mining
  2 Data Mining Toy Problem
  On some distant planet a very interesting form of life was found. No one ever had a chance to see these creatures yet it was discovered that they coul... [more]
  On some distant planet a very interesting form of life was found. No one ever had a chance to see these creatures yet it was discovered that they could consume all kinds objects delivered from Earth that for some inexplicable reason they found interesting. Planet researches investigating these unusual life-forms named the creatures omnivores and the planet where they lived - Gluttonia. The most prominent feature of omnivores was that after consuming some object of Earth origin they in return produced some amount of beautiful diamonds on the same place where the object was put on their home planet Gluttonia. Needless to say that discovery of this wonderful planet caused Earth economy revolution. From now on all Earth businesses started to fight each other to get as much diamonds as possible from Glouttonia omnivores. Many research expeditions were conducted to Glouttonia in attempt to find out what object omnivores like and how many diamonds they would produce in return for any given object. It was further discovered that omnivore interest depends not only on the nature of the object, but also on the frequency at which this object was dropped to Glouttonia surface and the way the object was presented to omnivore. Glouttonia is a strange planet staying most of the time in dark. Finally top-notch scientist from Earth came to the conclusion that to consume object omnivores first need to observe it in the rays of electric light. In general the more times object gets illuminated the higher chance that omnivore will be interested in it. It was also found that when omnivore detects illuminated object and before consuming it, omnivore gives the object one or more probes. Nobody knows how exactly these creatures do their probes. Maybe with beak or claws? The answer to this question we will never know, as omnivore probes the objects in the dark only, between flashes of light used for the object illumination that spaceships beam straight from Glouttonia orbit to her surface. Even with all proper illumination done there is no guarantee that omnivore will eventually eat the object. To say more these capricious creatures may get completely disinterested in the object even after they did several probes on it. It was decided to organize a study of omnivore behavior in a series of missions. During each mission many different objects were scattered all around the planet and omnivore consumption statistics were collected rigorously. As a result 10 Gb data-set was collected. This data-set summarizes a history of many missions attempted in different Glouttonia regions with different objects during 2 week period. Data-set consists of approximately 7 million records, where each record has the following fields: dd – date from 2-week survey. m_id - mission integer ID m_provider_id - mission provider integer ID obj_id - object integer ID obj_man_id - object manufacturer integer ID reg_id - ID of a region that Glouttonia surface was divided into, in all 50 different regions obj_freq - object drop frequency - how many times object was dropped to Glouttonia surface obj_illum_cnt - count of times object was illuminated obj_probe_cnt - count of times object was probed by omnivore obj_eat_cnt - count of times object was consumed by omnivore obj_diamond_cnt - count of diamonds omnivore gives for this object. As long as Glouttonia mission is an expensive enterprise, big corporations on Earth started to work on a solution that would allow them optimize mission expenses. Ideally, for every new object corporations would like to start with several trial missions with a short history - dropping objects in just a few Glouttonia regions for a few days only thus saving money on costly object illuminations. For each trial mission object consumption history would then be collected. Next, from histories of these short, trial missions companies would like to be able to predict the behavior of a full-fledged mission for the same objects with the same mission parameters as a trial ones had. Problem statement: Design an algorithm that for a given mission, object, Glouttonia region and day of the week will predict: 1) Mission revenue, measured in diamond count 2) Count of times object will be probed by omnivore 3) Count of times object will be consumed by omnivore To solve this problem existing data-set may be split in training and test sets. From test set several missions with short histories may be created. One possible approach would be for every mission in a test set assign 20% of records to a set of short history missions, so remaining 80% of records could be used for algorithm verification. What algorithms would you use to solve this problem?
  By Dmitri Kondratiev
  Following
• Answer added in Advanced Machine Learning
  51 What is your opinion about which programming language is better or best for Advanced Machine Learning for perfect AI; LISP, or something else?
  By Mahesh Sankpal · Dr. J. J. Magdum College of Engineering
  Dmitri Kondratiev

  Predicate logic, first order calculus - Prolog Symbolic programming - Lisp Numeric methods, ML - Python Neural Networks - Domain Specific Languages A... [more]

  Predicate logic, first order calculus - Prolog Symbolic programming - Lisp Numeric methods, ML - Python Neural Networks - Domain Specific Languages All the above together - Haskell ;))
  Following
• Answer added in Ontology
  2 Ontology of common sense?
  By Dmitri Kondratiev
  Dmitri Kondratiev

  Hector, thanks a lot for your advise! I was looking at CYC already before. Trying to query definition of "dark" at OpenSyc at: http://sw.opencyc.org... [more]

  Hector, thanks a lot for your advise! I was looking at CYC already before. Trying to query definition of "dark" at OpenSyc at: http://sw.opencyc.org/2012/05/10/concept/en/dark Results in the following output: [quote] A predicate corresponding to the function DarkFn. (dark X (DarkFn X)) will always be true for X meeting the relevant arg constraints. Subject Type: color Object Type: color More General: More Specific: Instance of: predicate that is strictly functional in its second argument [/quote] So in CYC 'dark' is a color, not an attribute of a color. How using CYC one can deduce that 'black', 'blue', 'red' and 'green' can be 'dark' and 'white' can not be dark? Thanks!
  Following
• Question asked in Ontology
  2 Ontology of common sense?
  I am working on a semantic search engine and looking for ready to use ontology of commonsense. I need simple relations of everyday objects ready to wo... [more]
  I am working on a semantic search engine and looking for ready to use ontology of commonsense. I need simple relations of everyday objects ready to work with computer tools. /--- For example: 1) Green --> isA --> Color 2) Dark --> isA --> Hue 3) Hue --> attributeOf --> Color /--- end of example ---/. Please advise on existing knowledge bases and tools for commonsense that have enough data and functionality to work with the above example. For example a query "dark car" may not return any results, yet data may contain "black car". Using ontology I will find that: 1) "dark" is an attribute of "color" class; 2) "white is a color instance and dosn't have "dark" attribute. 3) "black" is a color instance and has "dark" attribute. Now I can replace original query with "black car" and get non-empty result set. Any ideas how to do this?
  By Dmitri Kondratiev
  Following