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
(FEATURE) Intrinsic Merit of Music - Learning Weighted Automata from Music - Example
Srinivasan Kannan
More... over 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 662. (FEATURE) Intrinsic Merit of Music - Learning Weighted Automata from Music - Example - 14 May 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1.MusicWeightedAutomaton.py has been updated to define a new function audio_notes_to_samples() to read a music clip of certain duration by librosa, convert the waveform to notes, create notes string samples from them and return samples as word frequency dictionary and numeric encoded sample notes strings 2.SplearnArray is instantiated from encoded notes strings of variable size equal to number of columns in the notes matrix 3.Weighted automaton is learnt from these Sample dictionary and numeric encoded notes strings 4.automaton graph drawn in MusicWeightedAutomaton.gv.pdf shows a condensed weighted automata for an example music clip 5.Logs for this are committed to testlogs/MusicWeightedAutomaton.log.14May2019
Srinivasan Kannan
More... over 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE) Music Pattern Mining - Weighted Automaton Learning from Music
Srinivasan Kannan
More... over 7 years ago
----------------------------------------------------------------------------------------------------------------------------------------------- 661.(THEORY and FEATURE) Music Pattern Mining - Weighted Automaton Learning from Music - 11 May 2019 - related to all sections on Intrinsic Merit of Music ----------------------------------------------------------------------------------------------------------------------------------------------- 1.Previous sections described ascertaining merit of music by various analyses of waveforms e.g MFCC, learning DFA from music notes. Apart from these there are standard measures viz., zero crossings, chroma which quantify genre or merit of music (https://librosa.github.io/librosa/feature.html). 2.Of these learning a deterministic finite automaton from strings of music notes is a qualitative intrinsic merit measure of music which does not involve any statistical processing. 3.Learning DFA is a computational learning theoretic problem (PAC Learning, Angluin Model etc.,). Weighted Deterministic Finite State Automata are generalizations of DFA in which state transitions have an associate numeric weightage in addition to alphabets from a language. 4.For example, weighted DFA below has the weighted transitions: (state1, a, state2) - transition from state1 to state2 on symbol a of weight 1 (state2, 2b, state3) - transition from state2 to state3 on symbol b of weight 2 5.Weighted automata are apt formalisms to represent music notes strings as DFA because notes have subdivisions - sharp and flat - which can be likened to weightage for state transitions (https://en.wikipedia.org/wiki/Musical_note - CDEFGABC - ascending and descending graphic). 6.For example string of music notes - "CD(sharp)DEEGAF(flat)F(sharp)" - and its accepting weighted music automaton could be: state1, C, state2 state2, Dsharp, state3 ... state10, Fflat, state5 state6, Fsharp, state2 ... where sharp and flat are weights for the transition symbol. 7.Scikit Learn SPLearn (scikit-splearn) provides a weighted automaton learning framework based on Spectral Theory. 8.This commit implements a Music Weighted Automaton python code in MusicWeightedAutomaton.py which accepts a set of training string of music notes and fits them to a weighted automaton using scikit-splearn. It makes predictions later based on training data. 9.Learnt weighted automaton for music notes string is drawn as DOT graph file by graphviz in MusicWeightedAutomaton.gv and is rendered to a pdf - MusicWeightedAutomaton.gv.pdf 10.Weighting the transition in music notes has an another advantage: Some portions of the music waveform are more impressive and catchy which makes them to be listened again - https://en.wikipedia.org/wiki/Earworm - and transitions for these could have heavy weights.
Srinivasan Kannan
More... over 7 years ago
Theory - Pattern Grammars, Topological Handwriting Recognition
Srinivasan Kannan
More... over 7 years ago
Theory - Computational Geometric Factorization - Parallel Planar Point Location by Parallel Wavelet Tree Construction - References
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE) DeepLearning Convolution BackPropagation - handwriting recognition - utility function
Srinivasan Kannan
More... over 7 years ago
------------------------------------------------------------------------------------------------------------------------ 660. (FEATURE) DeepLearning Convolution BackPropagation - handwriting recognition - utility function - 5 May 2019 ------------------------------------------------------------------------------------------------------------------------ 1.DeepLearning_ConvolutionNetwork_BackPropagation.py has been changed to define a new utility function handwriting_recognition() which loads two images of handwritings by CV2, thresholds them, finds contours of the two images and approximates the contours by Douglas-Peucker Polynomial (DP) and computes the hausdorff distance between the two DP approximate polynomials of the handwritten images. 2.testlogs/DeepLearning_ConvolutionNetwork_BackPropagation.log.5May2019 shows how 2 pairs of handwritten digits fare against hausdorff istance measure - 2 pictures of handwritten digit 1 (similar), 2 pictures of handwritten digits 1 and 8 (dissimilar) 3.Logs show DP polynomials of contours of 1 and 1 are close enough while DP polynomials of contours of 1 and 8 are far apart: Handwriting Recognition [[[1279 719]]] [[[1276 719]]
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Music Pattern Mining - Music Synthesis by Automaton - Read from textfile
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------- 659. (FEATURE) Music Pattern Mining - Music Synthesis by Automaton - Read from textfile - 30 April 2019 -------------------------------------------------------------------------------------------------------- 1.Hardcoded states2notes DFA has been removed and DFA is read from a textfile NotesStateMachine.txt and ast.literal_eval()-ed to a dictionary. 2.Scaling of notes have been commented and note values are magnified (multiplied by 1000). Waveform timeseries can be visually analyzed by loading .wav file to a player and displaying waveform oscilloscope in visual effects. 3.Logs for this have been committed to testlogs/AudioToBitMatrix.log.MusicSynthesizer.30April2019
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE) Video EventNet Causality Graph Complexity Measures, Graph Minors, Merit of Large Scale Visuals
Srinivasan Kannan
More... over 7 years ago
------------------------------------------------------------------------------------------------------- 658. (THEORY and FEATURE) Video EventNet Causality Graph Complexity Measures, Graph Minors, Merit of Large Scale Visuals - 29 April 2019 ------------------------------------------------------------------------------------------------------- 1.This commit invokes networkx Graph API and creates NetworkX EventNet causality graph object for an example video. Connectivity of the Video EventNet is printed. 2.Example video in matrovska format which records the facebook profile of the author https://www.facebook.com/shrinivaasan.ka (containing NeuronRain commits twitter feeds) is comverted to mp4 by ffmpeg: ffmpeg -i ExampleVideo_Facebook_GRAFIT_29April2019.mkv -strict experimental ExampleVideo_Facebook_GRAFIT_29April2019.mp4 3.Usually most videos have set of frames which are consecutive and closely related. These frames create straightline linear paths of vertices of degree 2 in the EventNet Graph underneath it. 4.These straightline paths can be coalesced or edge-contracted to single vertex forming Graph Minors of EventNet. 5.Connectivity of a Video is a necessary but not sufficient condition for merit. 6.Rationale for Video EventNet Connectivity as Merit Measure: Video of disconnected underlying EventNet causality has scattered non-coherent information which is not appealing to viewer.
Srinivasan Kannan
More... over 7 years ago
(THEORY) Depth-2 Majority Function, Theoretical EVMs, Set Partitions, People Analytics, Axiom of Choice - Partition distance measures
Srinivasan Kannan
More... over 7 years ago
(THEORY) Depth-2 Majority Function, Theoretical EVMs, Set Partitions, People Analytics, Axiom of Choice - Ramanujan Congruences, Central Limit Theorem
Srinivasan Kannan
More... over 7 years ago
(THEORY) Depth-2 Majority Function, Theoretical EVMs, Set Partitions, People Analytics, Axiom of Choice
Srinivasan Kannan
More... over 7 years ago
Updated NeuronRain Usecases
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Medical Imageing Analytics - ElectroCardioGram (ECG), EventNet from Video
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------- 656, (FEATURE) Medical Imageing Analytics - ElectroCardioGram (ECG), EventNet from Video - 23 April 2019 -------------------------------------------------------------------------------------------------------- 1.Extracting patterns in medical imagery (ECG, MRI, Echo Cardio Gram Scans etc.,) is one of the most obvious analytics that can be performed on a Medical Image dataset which has widespread requirement in healthcare. 2.ImageNet Keras-Theano Python implementation has been augmented with a medical_imageing() function which takes an image source (presently ECG) and thresholds the image to retrieve contours from it by CV2 findContours(). 3.Image Contours are prominent features in an image which are more pronounced and distinguish it from other images. 4.Two example ECG images (Normal and Infarction) are compared by creating contours (i.e extract the ECG waveform from image) and computing the hausdorff distance between the two image contours: 5.Example comparison from logs in testlogs/ImageGraph_Keras_Theano.log.MedicalImageing.23April2019: Distance between Normal ECG and Normal ECG: (0.0, 0, 0) Distance between Normal ECG and Infarction ECG: (1741.9586677071302, 0, 0) 6.EventNet edges and vertices numbering in text files Video_EventNetEdges.txt and Video_EventNetVertices.txt has been corrected.
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE-BUGFIX) EventNet creation from Video - bug resolutions, corrections to EventNet vertices and edges text files
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------- 655. (THEORY and FEATURE-BUGFIX) EventNet creation from Video - bug resolutions, corrections to EventNet vertices and edges text files - 21 April 2019 -------------------------------------------------------------------------------------------------------- 1.ImageNet Keras Theano Python Implementation had some bugs in creating EventNet vertices and edges text files which have been resolved in this commit. 2.Frame prevfix has been included in EventNet edges text file. 3.Extracting EventNet from video facilitates computing usual dense subgraph complexity measures for Video similar to TextGraphs - e.g k-core, cycles, strongly connected components, algebraic connectivity etc.,
Srinivasan Kannan
More... over 7 years ago
Theory - EventNet and Video Intrinsic Merit, Process Calculi - references
Srinivasan Kannan
More... over 7 years ago
Theory - Parallel Planar Point Location by Wavelet Trees - References
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE) EventNet creation from Video, Actor Model, PetriNets, Video EventNet Tensor Products Intrinsic Merit
Srinivasan Kannan
More... over 7 years ago
------------------------------------------------------------------------------------------------------ 654. (THEORY and FEATURE) EventNet creation from Video, Actor Model, PetriNets, Video EventNet Tensor Products Intrinsic Merit - 19 April 2019 - related to all sections on Causality-Logical Time and Merit ------------------------------------------------------------------------------------------------------ 1.ImageNet Keras-Theano Python implementation has been changed to write EventNet Vertices and Edges files from causally related frame events of the video which can be read by GraphViz or NeuronRain EventNet Python implementation 2.EventNet differs from traditional PetriNet in following respects: 2.1 PetriNets have vertices having tokens called "Places" while EventNet has vertices having "actors" or "participants" in an event (Actor-Event Model) 2.2 PetriNets have transitions which move the tokens from Place-a to Place-b while EventNet has directed edges which connect a causing event vertex to an effect event vertex which is akin to flow of tokens. 2.3 EventNet is an Actor Model where actors interact in an event, modify local states and send messages to other actors. 3.Video EventNet vertices and edges are written to Video_EventNetEdges.txt and Video_EventNetVertices.txt
Srinivasan Kannan
More... over 7 years ago
(FEATURE and THEORY) Computational Geometric Factorization - Tile Search - Spark 2.4 -SparkContext range()
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------- 653. (FEATURE and THEORY) Computational Geometric Factorization - Tile Search - Spark 2.4 -SparkContext range() - 16 April 2019 -------------------------------------------------------------------------------------------------------- 1.DiscreteHyperbolicFactorizationUpperbound_TileSearch_Optimized.py has been changed to replace xrange() by Spark Context range() function in Spark 2.4. 2.Python xrange() is sequential and local while Spark range() is resilient distributed dataset which can be computed on cloud in parallel 3.Some ray query functions have also been transitioned to Spark Context range(). 4.An example integer is factorized and logs in testlogs/DiscreteHyperbolicFactorizationUpperbound_TileSearch_Optimized.log.16April2019 show some accurate ray shooting by Hardy-Ramanujan approximate factoring and exact factors. 5.Spark 2.4 has Barrier Synchronization feature (mentioned experimental) which is BSP model (like Pregel) than PRAM - https://spark.apache.org/docs/latest/api/python/pyspark.html#pyspark.BarrierTaskContext.get 6.Greenwich Mean Time (gmtime()) is printed after every factor is found for duration analysis of finding all factors.
Srinivasan Kannan
More... over 7 years ago
Theory - Exact Lagrange Four Square Tile Cover of Rectangle, Set Partitions, Integer Factorization, Non-Linear Programming - Correction 2
Srinivasan Kannan
More... over 7 years ago
Theory - Exact Lagrange Four Square Tile Cover of Rectangle, Set Partitions, Integer Factorization, Non-Linear Programming - Correction
Srinivasan Kannan
More... over 7 years ago
Exact Tile Cover, Factorization and Non-linear Programming - Correction
Srinivasan Kannan
More... over 7 years ago