Open Hub
Open Hub

 Settings |  Report Duplicate

0
I Use This!
Very Low Activity

Commits : Listings

Analyzed about 9 hours ago. based on code collected 1 day ago.
Nov 05, 2025 — Nov 05, 2026
Commit Message Contributor Files Modified Lines Added Lines Removed Code Location Date
(THEORY and FEATURE) Compressed Sensing for Texts - Syllable Vector Vowelless Text Compression
Srinivasan Kannan
More... about 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 710. (THEORY and FEATURE) Compressed Sensing for Texts - Syllable Vector Vowelless Text Compression - 18 September 2019 - Related to 2, 708 --------------------------------------------------------------------------------------------------------------------------------------------- 1.Compressed Sensing which is mostly restricted to digital signal and image processing, reconstructs original signal from lossy signal with high probability.(https://en.wikipedia.org/wiki/Compressed_sensing) 2.Vowelless Text Compression which compresses english texts by stripping vowels also creates a lossy compressed signal from which actual text has to be recovered with high probability and thus is a Text Compressed Sensing. 3.Existing Vowelless Text Compression implementation in neuronRain relies on Hidden Markov Model estimation to infer maximum likely missing vowels. 4.Syllables of a string in any language are phonetic divisions in the string. 5.Syllable boundary name clustering for clustering similar Names differing in spellings mentioned previously maps each string to a vector of syllables which is audio/linguistic representation of the string 6.Thus every n-syllable string irrespective of languages can be represented in an n-dimensional syllable space as a syllable vector and distance between two strings is the L2 norm in this syllable space - each syllable is phonetically represented than by a script. This is an alternative word2vec embedding model and is a better distance measure than levenshtein edit distance because of language script independence - 2 strings of different natural languages can be compared by phonetic syllable distance. 7.Syllable space embedding for similarity of names in NeuronRain is different from Phonetic Match Rating algorithm (https://en.wikipedia.org/wiki/Match_rating_approach) which also removes vowels but does not consider syllable boundaries or hyphenations. 8.CompressedSensing Python implementation has been changed to include a new function syllable_boundary_text_compression() which invokes PyHyphen python hyphenator and gets syllable vector of a string and compressed syllable vector by removevowels() in TextCompression implementation.
Srinivasan Kannan
More... about 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... about 7 years ago
FEATURE - Social Network Analysis - People Analytics - Contextual Name Parsing
Srinivasan Kannan
More... about 7 years ago
----------------------------------------------------------------------------------------------------------------------------------------- 709. (FEATURE) Social Network Analysis - People Analytics - Contextual Name Parsing - 14 September 2019 ----------------------------------------------------------------------------------------------------------------------------------------- 1.Name Parsing is an open problem and there are many tools already available (IBM Name Parser, Python Human Name Parser,...) 2.In the context of People Analytics, it is often required to find the first name, second name, middle name, surname, last name etc., from full names in a BigData set and there are cultural barriers e.g some countries have convention of <second name> <first name> and others have <first name> <second name> and <second name> is initial,surname or name of the parent/spouse. 3.Because of cultural barriers across nations parsing first name or second name in the absence of some context (e.g. an ID card) is difficult and prone to error. 4.New function nameparser() has been defined in SocialNetworkAnalysis_PeopleAnalytics.py which accepts full name, a regular expression and an ID contextual text as arguments. 5.nameparser() searches for the regular expression pattern and the tokenized full name in the ID contextual text and tries to infer First Name and Second Name. 6.An example name parsing for the author and failure of Python Human Name Parser has been demonstrated - Some past forum mails of the author from https://marc.info/?a=104547613400001&r=1&w=2 and https://marc.info/?l=apache-modules&m=105610024116012 sent from official mail id: [email protected] have wrong first (Kannan) and last (Srinivasan) names while final salutation in the emails contain the name per Indian convention as "K.Srinivasan" and Python Human Name parser fails to parse the names correctly in the absence of this context. 7.NeuronRain nameparser() defined in this commit is passed the email texts as contexts and the nameparser() searches for places in the emails containing the tokens of the full name by regular expression pattern. 8.Following lines define two named regex patterns which are searched in the email ID contexts mentioned in (6) as per Indian Cultural Convention of Second Name or Initial followed by First Name: hranal.nameparser("Kannan Srinivasan",r"(?P<second_name>\w+).(?P<first_name>\w+)",emailcontext_text) hranal.nameparser("Kannan Srinivasan",r"(?P<second_name>\w+) (?P<first_name>\w+)",emailcontext_text) 9.nameparser() returns the regex groupdicts for the matched named-parentheses 10.nameparser() has been made generic to suit all kinds of regular expressions for names and ID contexts and can be invoked for any social or professional network names. ID Context could be text from any source (e.g emails, texts, ID card texts) and regex pattern argument must comply with the ID context - if Context argument is an ID card text having Name, Address and Parent Names, UniqueID etc., regex pattern must accordingly follow e.g "Parent Name:(?P<parent_name>\w+)" 11.Logs for this commit are at testlogs/SocialNetworkAnalysis_PeopleAnalytics.log.14September2019 and email ID contexts at testlogs/SocialNetworkAnalysis_PeopleAnalytics_NameParsing/
Srinivasan Kannan
More... about 7 years ago
Theory - Word Hyphenation and Syllable boundaries - Reference
Srinivasan Kannan
More... about 7 years ago
THEORY and FEATURE - Vowelless String Compression and Syllable based Proper Noun clustering
Srinivasan Kannan
More... about 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Social Network Analysis - People Analytics - PIPL.com python API integration - Syllable based name clustering
Srinivasan Kannan
More... about 7 years ago
---------------------------------------------------------------------------------------------------------------------------------------------- 708. (FEATURE) Social Network Analysis - People Analytics - PIPL.com python API integration - Syllable based name clustering - 11 September 2019 ---------------------------------------------------------------------------------------------------------------------------------------------- 1.PIPL.com (people dot com) is a people search engine which tries to unqiuely identify a person based on name, email etc., 2.PIPL.com python API have been imported in SocialNetworkAnalysis_PeopleAnalytics.py and invoked in a newly defined function pipldotcom_analytics() which has query parameters: first name, last name and email 3.pipldotcom_analytics() has been invoked on 3 example emails and names and JSON response is printed in logs at testlogs/SocialNetworkAnalysis_PeopleAnalytics.log.11September2019 which contains identities, photos, followers and relationships sourced from multiple social networks (linkedin, twitter, facebook, google, outlook etc.,) 4.Example email queries are: [email protected], [email protected], [email protected] for the name of the author - "Srinivasan Kannan" 5.Previous query is a perfect example of multiple english name spellings and online identities author has for the single Hindi/Tamizh/Sanskrit name of the author "Sri-Ni-Va-San": Shrinivas Kannan, Srinivasan Kannan, Shrinivaasan Kannan (Shrinivaasan Ka) 6.This multiple name spelling problem arises in English because of multiple alphabets per syllable (hyphenated previously) and could easily cause identity confusion while in Tamizh, Hindi and Sanskrit name syllables have single vowel+consonant compound alphabets 7.Finding and clustering similarly named people and proving their uniqueness is a challenge in BigData because of spelling conflicts in English for similarly pronounced names. 8.Syllable based (acoustic) linguistic clustering of names by hyphenation previously as opposed to edit distance based clustering is thus a good model to classify similar names of different spellings. 9.Apart from Syllables, the 3 name spellings in previous example have same meaning which requires a Name ontology (NameNet) for semantically relating names of same meaning. 10. Names are non-dictionary words and creating a NameNet could be exhaustive and daunting.
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Recursive Lambda Function Growth - Machine Translation - simplified
Srinivasan Kannan
More... about 7 years ago
---------------------------------------------------------------------------------------------------------------------------------------------- 707.(FEATURE) Recursive Lambda Function Growth - Machine Translation - simplified - 6 September 2019 ---------------------------------------------------------------------------------------------------------------------------------------------- 1.machine_translation() function in RecursiveLambdaFunctionGrowth.py has been simplified to remove unnecessary arguments and clauses and do the graph traversal and create summary in target language from it by default. This is because relevance_to_text() function which chooses sentences from text based on some distance measure to create summary is not necessary during translation 2.3 clauses for choice of translator - googletrans, goslate, PyDictionary - have been simplified to translate the connectives "and" and "are" to respective target language. 3.3 logs of translation for Hindi, Kannada and Telugu have been committed to testlogs: testlogs/RecursiveLambdaFunctionGrowth.log.Telugu.MachineTranslation.6September2019 testlogs/RecursiveLambdaFunctionGrowth.log.Kannada.MachineTranslation.6September2019 testlogs/RecursiveLambdaFunctionGrowth.log.Hindi.MachineTranslation.6September2019 4.Previous logs have some intermittent errors because of Google translate API ratelimit. 5.ConceptNet is an alternative to Google Translate which is a multilingual ontology. 5.Translated text printed has some weird words not relevant to the text subject because of encoding/decoding problems of the language fonts. 6.Normal english language summary of the text - from create_summary() - is also printed 7.Usual machine translations are rule based and recently transformers (sequence attention neural network) are gaining traction.Main motivation of NeuronRain is to implement an alternative, non-statistical, formal languages theory based framework for analyzing texts exploiting deep linguistic structure in them 8.This formal languages focussed approach to measure merit of texts - NER, translation, extracting lambda expression from a text etc., is superior to inferring a neural network from text because text is approximated to a Turing machine indirectly by recursively grown lambda expressions which has strong cognitive psychology basis - Grounded Cognition - "phrasal structures are embedded recursively" 9.As far as machine translation is concerned, instead of sequence transducers, text graph of one language is mapped to text graph of another language bijectively thus preserving deep structure and meaning irrespective of language - minimizes "lost in translation".
Srinivasan Kannan
More... about 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Recursive Lambda Function Growth - Machine Translation - updated implementation
Srinivasan Kannan
More... about 7 years ago
----------------------------------------------------------------------------------------------------------------------------------------- 706.(FEATURE) Recursive Lambda Function Growth - Machine Translation - updated implementation - 5 September 2019 ----------------------------------------------------------------------------------------------------------------------------------------- 1.machine_translation() function in RecursiveLambdaFunctionGrowth.py has been changed to create a summary text in the destination language similar to create_summary() 2.As done in create_summary() summary translated text in target language is either created by traversing the textgraph or without traversing the textgraph 3.Presently translation creates only simple sentences. Complex sentences could be created in next iteration by compounding two neighbouring sentences. 4.Goslate and googletrans python libraries have been imported for translation because of frequent Invalid Word errors in PyDictionary() 5.Clauses specific to Goslate, googletrans and PyDictionary have been created along with exception handlers 6.Caveat is rate limitations of Google Translate REST API underlying all previous three translate libraries which might or might not work depending on load and requested time point. 7.If there are exceptions, source language edges are populated in translation text graph 8.logs for previous commit are at testlogs/RecursiveLambdaFunctionGrowth.log.MachineTranslation.5September2019 - because of Google translate REST API ratelimit translation textgraph could not be created
Srinivasan Kannan
More... about 7 years ago
(FEATURE) ConceptNet 5.7 Text Graph - additional Graph Complexity Intrinsic Merit measures
Srinivasan Kannan
More... about 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------- 705.(FEATURE) ConceptNet 5.7 Text Graph - additional Graph Complexity Intrinsic Merit measures - 31 August 2019 -------------------------------------------------------------------------------------------------------------------------------------- 1.ConceptNet 5.7 Text Graph implementation has been updated to print a comprehensive dictionary of graph complexity intrinsic merit measures of the ConceptNet text graph similar to WordNet text graph in RecursiveLambdaFunctionGrowth implementation 2.grow_lambda_function3() in RecursiveLambdaFunctionGrowth python implementation has been changed to accept a textgraph as argument in the absence of a text and directly grow lambda functions from argument textgraph without creating it. 3.Text is read from a file ConceptNet5Client.txt and passed on as conceptnet textgraph function argument. 4.logs for this commit are at testlogs/ConceptNet5Client.log.31August2019 5.Special concept vertex '/c/en/\xb0_c' is in all the paths found by least common ancestor algorithm and therefore the graph has a hub-spoke star structure always. This node has been filtered by a new function remove_noise() to avoid star graphs. To increase path quality more vertices might be added whenever deemed necessary in remove_noise() 6.Exception handling for non-existent paths has been included. 7.Resulting filtered textgraph has been drawn in testlogs/ConceptNet5Client5.7.TextGraph.31August2019.png and is more like a sunflower (set of subsets of vertices having pairwise intersection forming petals).
Srinivasan Kannan
More... about 7 years ago
(FEATURE) ConceptNet 5.7 Text Graph - shortest path filter
Srinivasan Kannan
More... about 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------- 704.(FEATURE) ConceptNet 5.7 Text Graph - shortest path filter - 25 August 2019 -------------------------------------------------------------------------------------------------------------------------------------- 1.ConceptNet 5.7 Text Graph implementation has been updated to find shortest paths from least common ancestor graph for each pair of words in the text and merge the shortest paths to create the graph by NetworkX add_path() 2.This filtering improves the path quality to some extent. 3.Centrality measures and Dense Subgraph Classification (Core Numbers) are printed. 4.logs for this commit are at testlogs/ConceptNet5Client.log.25August2019
Srinivasan Kannan
More... about 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Text Graph of a natural languge text from ConceptNet 5.7 - implementation
Srinivasan Kannan
More... about 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 703.(FEATURE) Text Graph of a natural languge text from ConceptNet 5.7 - implementation - 23 August 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1.NeuronRain text graph implementations so far rely only on WordNet for deciphering intrinsic deep structure merit of a text. 2.This commit implements ConceptNet 5.7 text graph alternative which was long pending. 3.New function conceptnet_text_graph() invokes least_common_ancestor graph creating function and aggregates edges from paths between all pairs of words. 4.ConceptNet5.7 Text Graph has been plotted by matplotlib at testlogs/ConceptNet5.7.TextGraph.23August2019 5.Logs for ConceptNet text graph creation are at testlogs/ConceptNet5Client.log.23August2019 6.Quality of Paths in ConceptNet is an open problem e.g https://arxiv.org/abs/1902.07831
Srinivasan Kannan
More... about 7 years ago
(THEORY-FEATURE) Non-statistical Ontology Semantic Paths based Named Entity Recognition
Srinivasan Kannan
More... about 7 years ago
------------------------------------------------------------------------------------------------------------------------------------------- 702.(THEORY-FEATURE) Non-statistical Ontology Semantic Paths based Named Entity Recognition - related to 697 - 22 August 2019 ------------------------------------------------------------------------------------------------------------------------------------------- 1.A non-statistical WordNet Paths based Named Entity Recognition alternative was theoretically described earlier which exploits linguistic deep structure. 2.This alternative has been implemented in NamedEntityRecognition_HMMViterbi_CRF.py which works as below: 2.1 find the semantic path between a PoS annotated word and unannotated adjacent word in an ontology e.g WordNet - path_between() from WordNetPath.py has been invoked (instead the text graph path between the words is also sufficient) 2.2 for each vertex in the path find all possible PoS tags from WordNet Synsets which is a 2 dimensional list - Words * Set-of-PoS-per-Word 2.3 find the cartesian product of the 2 dimensional list from 2.2 which enumerates all possible PoS paths between annotated and unannotated words - itertools product computes cartesian product of lists for all possible PoS paths which is a trellis 2.4 aggregate the penultimate vertices in PoS paths 2.5 find the majority PoS tag Synset of 2.4 which is the most likely PoS of the unannotated word 3.Logs in testlogs/NamedEntityRecognition_HMMViterbi_CRF.log.22August2019 demonstrate previous PoS paths majority voting for NER of an unannotated vertex. 4.Some paths between annotated and unannotated words drew blank because path_between() did not find a path 5.In some respects, previous algorithm resembles a transformer which is an attention model neural network scoring a sequence pipeline in transduction - it maps a sequence of words to sequence of PoS and attention context (important annotated words influencing an unannotated word) is provided by the PoS paths between PoS annotated and unannotated words in the word sequence.
Srinivasan Kannan
More... about 7 years ago
(THEORY) Shell Turing Machines, MRDP Theorem, Turing Degrees, Category Theory, Diophantine Equations, Embedding Formal Languages in Vector Space
Srinivasan Kannan
More... about 7 years ago
(THEORY) Category Theory formulation of EventNet - Objects, Morphisms, Functors
Srinivasan Kannan
More... about 7 years ago
(THEORY) Computational Geometric Factorization, Complement Diophantines, Covering a set by Arithmetic Progressions, Ramsey Coloring
Srinivasan Kannan
More... about 7 years ago
(THEORY-FEATURE) Computational Geometric Factorization - Tile Search Optimization - Spark 2.4.3 and QuadCore benchmarks
Srinivasan Kannan
More... about 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 699.(THEORY-FEATURE) Computational Geometric Factorization - Tile Search Optimization - Spark 2.4.3 and QuadCore benchmarks - 13 August 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1.This is the first genuine benchmark of Computational Geometric Factorization on Spark 2.4.3 and QuadCore. Earlier execution on local[4] was a simulation of 4 spark executor threads on single core 2.SparkContext has been replaced by Spark context of SparkSession 3.logs for factorization are at testlogs/DiscreteHyperbolicFactorizationUpperbound_TileSearch_Optimized.log.13August2019 which have some notable estimates of approximate factors by Hardy-Ramanujan (1 extact factor ray query), Baker-Harman-Pintz and Cramer ray shooting queries
Srinivasan Kannan
More... about 7 years ago
(FEATURE) GIS Remote Sensing Image Analytics - Image Segmentation
Srinivasan Kannan
More... about 7 years ago