A

asfer-github-code

Settings | Report Duplicate

0

I Use This!

Very Low Activity

Commits : Listings

Analyzed about 4 hours ago. based on code collected 1 day ago.

Commit Message	Contributor	Files Modified	Lines Added	Lines Removed	Code Location	Date
Nov 05, 2025 — Nov 05, 2026 Showing page 106 of 130 Search / Filter on:
) Software Analytics Deep Learning code has been updated to include all Deep Learning implementations: RNN LSTM, RNN GRU, BackPropagation and ConvolutionNetwork+BackPropagation ) logs for this have been committed to software_analytics/testlogs/	Srinivasan Kannan	More...				over 9 years ago
) Software Analytics Deep Learning code has been updated to include all Deep Learning implementations: RNN LSTM, RNN GRU, BackPropagation and ConvolutionNetwork+BackPropagation ) logs for this have been committed to software_analytics/testlogs/	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
) Implemented Korner Entropy intrinsic merit for Recursive Gloss Overlap graph. ) function korner_entropy() finds maximal independent sets for all vertices and chooses the minimum entropy such that a vertex is in a maximal independent set. *) Logs for this have been committed to testlogs/	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
---------------------------------------------------------------------------------------------------------------------- Recursive Lambda Function Growth + Tensor Neuron (NTN) Intrinsic Merit ---------------------------------------------------------------------------------------------------------------------- ) This commit rewrites the Recursive Lambda Function Growth implementation by adding a new lambda function growth function. ) This new function creates a definition graph with Recursive Gloss Overlap algorithm from a text and computes all pairs shortest paths in the graph ) For each such shortest path edges, an AVL lambda function composition balanced tree is grown (i.e every path is a lambda function composition tree). ) Tensor Neuron Lambda Function is computed for each subtree root by following relation evaluated on a stack: function(operand1, operand2) where operand1, operand2 and function are successively popped from AVL composition tree stack representation ) Tensor Neuron triplet (operand1, function, operand2) is evaluated presently as maximum Wu-Palmer similarity of Synsets for cartesian product of operand1 and operand2. But it can be augmented with any arbitrary weighting scheme. Cartesian product of Synsets for 2 entities is the simplest 2-dimensional tensor neuron. ) Sum of tensor neuron weights for all random walks composition tree evaluation is printed as merit of the text. *) Logs for this have been committed to testlogs/	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
------------------------------------------------------------------------------------------------------------- Deep Learning Recurrent Neural Network Gated Recurrent Unit (RNN GRU) Implementation ------------------------------------------------------------------------------------------------------------- ) This commit implements Gated Recurrent Unit algorithm (most recent advance in deep learning - published in 2014) which is a simplification of RNN LSTM by reducing number of gates (input, cell, forget, output are mapped to cell, reset, update) ) Logs for this have been uploaded to testlogs/ showing convergence of gates and state.	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
) Refactored the tobit() function - moved it to a new directory image_pattern_mining and reimplemented it with a new image_to_bitmatrix() function in ImageToBitMatrix.py by enumerating through the pixel rows and applying map(tobit) ) This removed the opaqueness related 255 appearing for all pixels in bitmap. ) Each pixel is multiplied by a fraction. Mapping it to binary digits 0,1 based on [0-127],[128-255] intervals for pixel values does not work well with BackPropagation of MaxPooling in Convolution Network. Instead multiples of a small fraction makes the MaxPooling layer quite receptive to changes in image pixel values. ) Added few handwriting recognition example images for numbers 1(two fonts),2 and 8. *) Logs for this have been committed to testlogs/	Srinivasan Kannan	More...				over 9 years ago
Updated Requirements.txt	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
) Updated python-src/DeepLearning_ConvolutionNetwork_BackPropagation.py tobit() function to map each pixel to a decimal value < 1 ) Increased BackPropagation iteration to 100. *) Logs for this have been committed to testlogs/. With this final neurons for max pooling layer are sensitive to changes in image pixels	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
--------------------------------------------------------------------------------------------------------------------- DeepLearning JPEG files with Python Imaging Library (PILlow) --------------------------------------------------------------------------------------------------------------------- ) Added import of PIL Image library to open any image file and convert into a matrix in Integer mode ) This matrix is input to Convolution and BackPropagation code *) Logs of Convolution and BackPropagation for sample images have been committed to testlogs/	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
---------------------------------------------------------------------------------------------- Convolution Network Final Neuron Layer Integrated with BackPropagation Implementation ---------------------------------------------------------------------------------------------- ) New python module has been added to repository to invoke BackPropagation algorithm implementation in final neuron layer of convolution network. ) This creates a multilayer perceptron from maxpooling layer variables as input layer and backpropagates for few iterations to do weight updates. ) Maxpooling layer is a matrix of 55=25 variables. With 25 variable input layer, backpropagation is done on 25252=1250 activation edges with weights(25 inputs * 25 hidden + 25 hidden * 25 outputs). ) Logs for this have been added to testlogs/ ) Presently expected output layer has been hardcoded to some constant.	Srinivasan Kannan	More...				over 9 years ago
---------------------------------------------------------------------------------------------------------------------------------------- Major rewrite of BackPropagation Implementation for arbitrary number of Neuron input variables layer - 2 March 2017 ---------------------------------------------------------------------------------------------------------------------------------------- ) BackPropagation algorithm code has been rewritten for arbitrary number of input layer, hidden layer and output layer variables. ) Some Partial Differential Equations functions have been merged into one function. ) This was executed with 6 variable input layer neuron, 6 variable hidden layer and 6 variable output layer with 266=72 weights for each of 72 activations. ) Logs for this have been committed to testlogs/ *) For 100000 iterations, error tapers to ~10^-29 as below -------------------------------------------------------------------------------------------------------------- Error after Backpropagation- iteration : 99998 1.05827458078e-29 Layers in this iteration: Input Layer: [0.23, 0.11, 0.05, 0.046, 0.003, 0.1] Hidden middle Layer: [0.07918765738490302, 0.17639329000539292, 0.3059104560477687, 0.06819897810162112, 0.13596389556531566, 0.18092367420130773] Output Layer: [0.29999999999999977, 0.53, 0.11000000000000039, 0.09000000000000002, 0.010000000000004578, 0.2099999999999999] Weights updated in this iteration: [0.048788506679748, 0.11426171324845727, 0.24603459114714846, 0.0795780558315572, 0.17784209754093078, 0.18901665972998824, 0.11262698024008909, 0.20899193850252168, 0.5190303819914409, 0.3056462222980023, 0.5043023226568699, 0.6597581902164429, 0.16317511417689112, 0.6856596163043989, 1.0387023103855846, 0.2606362555325616, 0.9902995852922012, 1.0606233996394616, 0.03792106334721366, 0.08894920940249851, 0.19690317303590538, 0.06956852060077344, 0.15685074966243606, 0.1617685764906898, 0.07393365494416165, 0.11785754559950794, 0.32914500297950156, 0.2691055311488234, 0.4255090307780099, 0.5588219323805761, 0.17422407739745155, 0.7118075853635217, 0.335920938970278, 0.08421750448989332, 0.18955175416917805, 0.2131459459234997, 0.18585916895619456, 0.2871421205369935, 0.4097382856269691, 0.09927665626493291, 0.2820455537957722, 0.44170499922267087, 0.4181874495051114, 0.6343547541074734, 0.5019659876526656, 0.6341824150669009, 0.9311854573920668, 0.415739542966714, 0.8116648622103757, 0.7642211240783323, -0.27787045024304763, -0.03796350343576473, -0.06897746410924578, -0.12106324365840128, 0.09688102161356603, 0.0833653488478167, 0.025844281679456543, 0.01924251728820474, 0.1166534752079898, 0.22544113767193752, 0.2565459528164421, 0.2598246679566054, -0.2448286735461752, 0.48545168615654455, -0.12033616818940464, -0.17996884551601777, 0.09161370830913909, 0.16523481156548922, 0.22079944846849042, 0.4246703906819317, 0.05813513760785996, 0.49187786590191596] Recomputing Neural Network after backpropagation weight update Error after Backpropagation- iteration : 99999 1.05827458078e-29 Layers in this iteration: Input Layer: [0.23, 0.11, 0.05, 0.046, 0.003, 0.1] Hidden middle Layer: [0.07918765738490302, 0.17639329000539292, 0.3059104560477687, 0.06819897810162112, 0.13596389556531566, 0.18092367420130773] -------------------------------------------------------------------------------------------------------------------	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
--------------------------------------------------------------------------------------------- Streaming Algorithms Implementations for Spark Streaming Parquet Data Storage --------------------------------------------------------------------------------------------- ) All Streaming_<algorithm> implementations have been updated to have Spark Streaming Parquet file storage as input generators. ) New Streaming_SocialNetworkAnalyis.py file has been added to do sentiment analysis for Streaming Data from Spark Parquet file storage ) Streaming_AbstractGenerator.py has been updated to filter each Spark Parquet DataFrame to remove grammatical connectives and extract only keywords from streamed content ) Logs for all Streaming_<algorithm>.py and Streaming_SocialNetworkAnalysis.py executed against Spark Streaming Parquet data have been committed to testlogs/ ) Streaming_SocialNetworkAnalysis.py presently iterates through word stream generated by Streaming_AbstractGenerator.py and does Markov Random Fields Clique potential sentiment analysis. ) Streaming_AbstractGenerator.py Spark Parquet __iter__() clause presently returns word stream which can be optionally made to return sentences or phrases of specific length	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago
-------------------------------------------------------------------------------------------- Java Spark Generic Streaming and Streaming Abstract Generator Integration -------------------------------------------------------------------------------------------- ) Enabled URLSocket in Spark Generic Streaming and crawled an example facebook and twitter streaming data stored into Parquet files and Spark metastore ) Added Spark Streaming Data Source and Parquet Data Storage to Streaming_AbstractGenerator.py and updated iterator to read DataFrames from word.parquet Spark Streaming storage. ) Spark DataFrames are iterated by accessing word column ordinal and yielded to any application of interest. ) As an example Streaming HyperLogLogCounter implementation has been updated to read from Spark Parquet streaming data storage and cardinality is computed ) Example Spark Generic Streaming logs with URLSockets and HyperLogLogCounter with Spark Parquet data storage have been added to testlogs in python-src and java-src/bigdata_analytics/spark_streaming/testlogs ) Thus an end-to-end Java+Spark+Python streaming ETSL pipeline has been implemented where: - Extract is done by JSoup crawl - Transform is done by Spark RDD transformations - Store is done by Spark Hive and Parquet storage - Load is done by Streaming Abstract Generator iterator/facade pattern	Srinivasan Kannan	More...				over 9 years ago
Updated AsFer Design Document	Srinivasan Kannan	More...				over 9 years ago

←
1
2
…
102
103
104
105
106
107
108
109
110
…
129
130
→