Open Hub
Open Hub

 Settings |  Report Duplicate

0
I Use This!
Very Low Activity

Commits : Listings

Analyzed about 7 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 - Query Index and Ranking - Sorting by Intrinsic Merit
Srinivasan Kannan
More... over 8 years ago
----------------------------------------------------------------------------------------------------------- Query Index and Ranking - Sorting by merit ----------------------------------------------------------------------------------------------------------- 1.QueryIndexAndRank.py has been changed to sort the results queried from LSH and ThoughtNet indices by descending order of Graph Tensor Neuron Network Intrinsic Merit. Earlier sorting was missing. Two dictionaries have been created for LSH and ThoughtNet index rankings and are sorted descending. 2.Locations of ThoughtNet Hypergraph have been updated in Indexing/ 3.Logs for the ranking have been captured in testlogs/QueryIndexAndRank.log.22June2018
Srinivasan Kannan
More... over 8 years ago
Theory - Quantum Computational Geometric Factorization - Update
Srinivasan Kannan
More... over 8 years ago
SAT Solver Update - Randomized Rounding
Srinivasan Kannan
More... over 8 years ago
Theory - Update to Quantum NC Computational Geometric Factorization
Srinivasan Kannan
More... over 8 years ago
3SAT least square solver update - increased LSMR accuracy
Srinivasan Kannan
More... over 8 years ago
----------------------------------------------------------------------------------------------------------- Random 3SAT Approximate LSQR/LSMR Solver update - Increase in Accuracy ----------------------------------------------------------------------------------------------------------- 1.CNFSATSolver.py has been updated to increase the accuracy of least squares approximation of SAT assignments by setting atol=btol=conlim=0 as prescribed by SciPy documentation in : https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.linalg.lsmr.html Also set the damp to a low value close to zero to solve the 3SAT as regularized least squares problem which is a generalization of least squares. 2.Random 3SAT instances of 1000 variables and Alpha=1.0 have been solved and MAXSAT approximation ratio for this converge to ~97% for few iterationsA 3.This increase in accuracy slows down LSMR. But by better least squares optimization which is parallelizable on cloud viz., LSRN - http://web.stanford.edu/group/SOL/software/lsrn/ which is recent parallel algorithm for solving least squares (LSRN paper - http://web.stanford.edu/group/SOL/papers/lsrn-sisc-2014.pdf) this slow down can be remedied.
Srinivasan Kannan
More... over 8 years ago
Theory - Quantum NC Computational Geometric Factorization - Update
Srinivasan Kannan
More... over 8 years ago
Theory - Quantum NC algorithm for Computational Geometric Factorization
Srinivasan Kannan
More... over 8 years ago
Theory - Quantum NC algorithm for Computational Geometric Factorization
Srinivasan Kannan
More... over 8 years ago
Theory - Quantum NC algorithm for Computational Geometric Factorization
Srinivasan Kannan
More... over 8 years ago
Theory - Computational Geometric Factorization, Ray Shooting Query Oracles, Transition from Quantum to Classical computation
Srinivasan Kannan
More... over 8 years ago
Added BestPractices.txt - Coding Conventions and Practices in NeuronRain
Srinivasan Kannan
More... over 8 years ago
Scheduler Analytics Update - ConvNet for graphic performance streamed image data
Srinivasan Kannan
More... over 8 years ago
--------------------------------------------------------------------------------------------------------------- Scheduler Analytics Update - New Convolution Network for analyzing stream of 2-dimensional graphics performance image data --------------------------------------------------------------------------------------------------------------- Pictorial Performance Pattern Mining by Convolution Network: --------------------------------------------------------------------------------------------------------------- Input to this Convolution Network is an image bitmap matrix obtained from some standard graphic performance analyzers - runqueuelat, BPF/bcc-tools, perf etc., Following example image is created by "perf timechart" on perf.data which captures the state,timeslice etc., information of list of processes in scheduler at any instant in the form of a gantt chart. Backpropagation-Convolution is then applied to this performance snapshot and convolution, maxpooling, final neuron layers are computed as usual. This convolution is iterated periodically for stream of performance snapshot bitmaps. Advantages of mining graphic images of performances captured periodically are: - number of dimensions of a performance snapshot image is 2 while psutils creates 1 dimensional array stream of data per process id. - 2 dimensional performance snapshots are system wide and capture all process id(s) at a time and not process id specific. - Recent advanced performance analyzers like BPF/bcc-tools provide a histogram equivalent of timechart in perf (runqueuelat) which is also an image bitmap. So support for graphic performance data analytics is futuristic. - Processing stream of performance snapshot image bitmaps by convolution to extract patterns is more universal/holistic than applying convolution to stream of process structure data - Specifically if remaining_clockticks-to-processes dynamic hash tables are available as stream of histogram images, applying convolution on this stream creates frequent recurring patterns of queueing in OS Scheduler (Described in GRAFIT course material: - https://github.com/shrinivaasanka/Grafit/blob/master/course_material/NeuronRain/AdvancedComputerScienceAndMachineLearning/AdvancedComputerScienceAndMachineLearning.txt, - https://gitlab.com/shrinivaasanka/Grafit/blob/master/course_material/NeuronRain/AdvancedComputerScienceAndMachineLearning/AdvancedComputerScienceAndMachineLearning.txt )
Srinivasan Kannan
More... over 8 years ago
Measuring Intrinsic Fitness - References
Srinivasan Kannan
More... over 8 years ago
SAT Solver Update - Damp variable for nonuniform_choice3
Srinivasan Kannan
More... over 8 years ago
---------------------------------------------------------------------------------------------------------- SAT Solver Update - nonuniform_choice3() damp - 2000 variables and Alpha=4.267 ---------------------------------------------------------------------------------------------------------- 1.CNFSATSolver.py - damp variable in nonuniform_choice3() function has been fine tuned to match the theoretical probability of 1/sqrt(m*n) - observed probability almost matches with theoretical 2.random 3SAT instances for Number of variables=100,200,300,400,500,600,700,800,900,1000,1100,1200,1300,1400,1500 and 2000 for alpha=4.267 have been solved and observed probability closely matches the random matrix probability 3.Logs for 2000 variables and Alpha=4.267 have been committed to testlogs/CNFSATSolver.2000clausesAlpha4.267.log.12June2018. 4.Observed MAXSAT approximation ratio for first few iterations hovers around 90-91-92% for all previous variable combinations. 5.Theoretically, if damp variable is perfectly tuned to create required bias in nonuniform choice, then MAXSAT ratio observed for unequal variable-clause combinations should also be in the range of 94-95% if not 100%.
Srinivasan Kannan
More... over 8 years ago
Theory - Computational Geometric Factorization and Segment Trees
Srinivasan Kannan
More... over 8 years ago
Ray Shooting Queries - Constants Changed
Srinivasan Kannan
More... over 8 years ago
---------------------------------------------------------------------------------------------------------------- 559. (THEORY and FEATURE-DONE) Computational Geometric Factorization - Ray Shooting Queries - Constant Changes - 8 June 2018 ---------------------------------------------------------------------------------------------------------------- 1.Constants in all ray shooting query functions have been changed so that gaps are reduced and possibility of exact hit increases. Specifically constant l for Cramer, Baker-Harman-Pintz, Zhang has been made small to widen xrange(). 2.2 integers have been factored and approximate factors for all 5 queries are printed after these new constant values. Zhang and Cramer queries were overshooting previously beyond N. None of the queries exactly shoot but are quite close to actual factors.
Srinivasan Kannan
More... over 8 years ago
Ray Shooting Queries - Corrections to Hardy-Ramanujan-PrimeNumberTheorem estimate of prime factors
Srinivasan Kannan
More... over 8 years ago
-------------------------------------------------------------------------------------------------------------- Computational Geometric Factorization - Hardy-Ramanujan-PNT Ray Shooting - Correction -------------------------------------------------------------------------------------------------------------- 1. Mysterious convergence of merit p/logp has been investigated. 2. This estimate equates the Prime number theorem merit difference between 2 prime factors and ratio of Prime counting function and Hardy-Ramanujan prime factors estimate as below: l*pf(n+1)/log(pf(n+1)) - l*pf(n)/log(pf(n)) = l*N/(logN * k*loglogN) 3. Earlier the constant multiples k and l were missing in some places which has been rectified. 4. Even after this, previous ratio converges after lot of iterations for l=10 and k=6. So suitable value of l is still elusive. 5. Two integers with large Omega (usually numbers ending in 5 have lot of factors) have been factored. While one integer has exact shoot from Hardy-Ramanujan, the other has close shave, but approximate factors are quite in proximity to actual factors. 6. This establishes utility of ray shooting queries as an Oracle access implementation which can be used both by classical P and quantum period-finding BQP (P^A versus BQP^A) Factoring Turing Machines which at most requires O(loglogN) queries.
Srinivasan Kannan
More... over 8 years ago
Feature-Ray Shooting Queries-Correction
Srinivasan Kannan
More... over 8 years ago
------------------------------------------------------------------------------------------------------------- Computational Geometric Factorization - Ray shooting queries - correction ------------------------------------------------------------------------------------------------------------- 1.DiscreteHyperbolicFactorizationUpperbound_TileSearch_Optimized.py has been updated to correct the number of primes estimate between two factors - constant for prime counting function (merit) O(p/logp) has been moved to numerator which was earlier in denominator incorrectly. 2.Two integers have been factored and logs have been committed to testlogs/. Both show exact ray shoot finding factors correctly matching with actual factors. These are on Single Core + Spark 2.3 with local[4]. This simulates quadcore on single core and some factors are printed out of order probably because of Spark Driver Executor Threads.
Srinivasan Kannan
More... over 8 years ago
Theory - Factorization - P and BQP - Relativization - Ray Shooting Queries and Period Finding
Srinivasan Kannan
More... over 8 years ago
Theory and implementation - Yitang Zhang Prime Gaps estimation for Ray shooting queries
Srinivasan Kannan
More... over 8 years ago
-------------------------------------------------------------------------------------------------------------- Computational Geometric Factorization - Ray Shooting Queries - Yitang Zhang Prime Gap Estimate -------------------------------------------------------------------------------------------------------------- 1.DiscreteHyperbolicFactorizationUpperbound_TileSearch_Optimized.py has been updated to include a new function for ray shooting queries which implements recent breakthrough result in Prime Gap estimation - there are infinitely many twin primes separated by gap < 70000000 - by [Yitang Zhang] which refines [Goldston-Pintz-Yildirim] Sieve. This might have effect only for finding primes between prime factors of large integers and summing up the gap for spacing between any two prime factors. 2.Constants k and l for multiplying estimation bounds have been changed slightly. 3.logs for this have been committed to testlogs/ for an 8-digit 26-bit integer. 4.None of the five ray shooting queries in the bouquet shoot exactly. But Hardy-Ramanujan is still the best and Baker-Harman-Pintz estimate is also close enough to actual factors. 5.Ray Shooting's success rate depends on the number and multiplicity of the prime factors - Integers of large Big Omega and Small Omega would have lot of prime factors taking a point-blank hit. In other words large multiplicative partition (Factorisatio Numerorum) number of an integer is a pre-requisite for exact ray shooting. Finding the values of the constants for estimation functions which are normal orders, is presently done by trial and error. Correct values for these constants are also crucial for exact shooting. 6.Another weird observation is the convergence of the Hardy-Ramanujan-PrimeNumberTheorem estimate after some iterations and this converged integer sometimes matches closely with actual factors. This has happened in earlier logs too.
Srinivasan Kannan
More... over 8 years ago
AsFer Design Document - BQP and NC - Decoherence and Philosophical Implications of Factorization in NC
Srinivasan Kannan
More... over 8 years ago
AsFer Design Document - Theory - Oracle version of Condorcet Jury Theorem circuit and conflicts
Srinivasan Kannan
More... over 8 years ago