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
--------------------------------------------------------------------------------------------------------------------------------------------- 698. (FEATURE) GIS Remote Sensing Image Analytics - Image Segmentation - 13 August 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1.New function image_segmentation() has been defined in ImageGraph_Keras_Theano.py which segments an image into similar regions or patches 2.As opposed to RGB patches and extract_patches_2d() of scikit learn, segmentation of an image is generic and segments/patches of arbitrary irregular shapes and containing similar pixels can be extracted 3.Example code in OpenCV2 documentation - https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_watershed/py_watershed.html - has been used as reference 4.This example filters the image into foreground and background by: 4.1 applying OTSU threshold filter 4.2 morphological opening for removal of noise 4.3 Background - Dilation to increase object boundary to background and demarcate background accurately 4.4 Foreground - Distance transform and then thresholding is applied to demarcate the foreground 4.5 Markers - Connected components are found from foreground of 4.4 4.6 Watershed - watershed segmentation algorithm is applied on the marked image to find segments. 5.Watershed algorithm works by considering an image as topographic surface of valleys and dams between valleys. Valleys in image are filled with water till no two adjacent valleys merge creating "dams" or segments bounding the valleys. 6.Watershed segmentation is graphically illustrated in http://www.cmm.mines-paristech.fr/~beucher/wtshed.html 7.EventNet Tensor Products Merit algorithm for visuals described earlier depends on bounding box segmentation which are squares. Instead segmentation could be the way to go if ImageNet and Keras support it. 8.Two segmented satellite images of Chennai Metropolitan Area Urban Sprawl are committed at: 8.1 testlogs/RemoteSensingGIS/ChennaiUrbanSprawl_Page-10-Image-15_segmented.jpg 8.2 testlogs/RemoteSensingGIS/ChennaiUrbanSprawl_Page-9-Image-13_segmented.jpg
Srinivasan Kannan
More... about 7 years ago
(THEORY-FEATURE) Named Entity Recognition and Text Graph Algorithms in NeuronRain AstroInfer - a non-statistical alternative
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Named Entity Recognition (NER)-HMM-Viterbi-CRF - Rewrite
Srinivasan Kannan
More... about 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 696. (FEATURE) Named Entity Recognition (NER)-HMM-Viterbi-CRF - Rewrite - 12 August 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1.Named Entity Recognition by Conditional Random Fields Hidden Markov Model has been updated significantly for resolving some errors in Viterbi computation 2.Prior Emission Probabilities computation by skew normal distribution has been updated to renormalize by dividing the skewnorm probability by sum of probabilities 3.New part of speech "adverb" has been included in transition probabilities,emission probabilities and states 4.weighted sknorm has been done away with and replaced by previous renormalized skewnorm distribution 5.Error in argmax() computation has been corrected - label and maximum are reset in the loop 6.logs for this rewrite have been committed to testlogs/NamedEntityRecognition_HMMViterbi_CRF.log.12August2019 7.Transition probabilities are updated as below: transition_probabilities={ 'noun':{'noun':0.0, 'verb':0.3, 'object':0.2, 'adjective':0.1, 'adverb':0.1, 'conjunction':0.3}, 'verb':{'noun':0.1, 'verb':0.0, 'object':0.4, 'adjective':0.2, 'adverb':0.1,'conjunction':0.3}, 'object':{'noun':0.0, 'verb':0.1, 'object':0.4, 'adjective':0.1, 'adverb':0.1,'conjunction':0.3}, 'adjective':{'noun':0.4, 'verb':0.2, 'object':0.2, 'adjective':0.0, 'adverb':0.1, 'conjunction':0.1}, 'adverb':{'noun':0.1, 'verb':0.4, 'object':0.1, 'adjective':0.0, 'adverb':0.1, 'conjunction':0.3}, 'conjunction':{'noun':0.2, 'verb':0.4, 'object':0.1, 'adjective':0.1, 'adverb':0.1, 'conjunction':0.1} } 7.It has to be mentioned here that previous state transition probabilities priors are hardcoded heuristics and based on the following rationale: 7.1 verb follows nouns more often 7.2 object follows verb more often 7.3 object follows object more often 7.4 noun follows adjective more often 7.5 verb follows adverb more often 7.6 verb follows conjunction more often (e.g infinitives) 8.Previous heuristics are minimal and could be replaced by more sophisticated probability distribution learnt from a training dataset once and loaded everytime.
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Linear and Logistic Regressions - Rewrite - Intrinsic merit purview update
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Linear and Logistic Regressions - Rewrite - Intrinsic merit purview update
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Gradient Ascent and Descent - Rewrite
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 695. (FEATURE) Gradient Ascent and Descent - Rewrite - 4 August 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1.PerceptronAndGradient.py has been rewritten for computing Gradient Ascent and Descent for arbitrary number of variables. 2.Earlier loop for partial derivative delta computation for just 2 variables has been generalized for any number of variables 3.Gradient Ascent has also been added based on computed value of weight updated perceptron (+ instead of -) 4.Logs for Ascent and Descent are at testlogs/PerceptronAndGradient.log.4August2019 5.Hardcoded variable,weights,bias,rho,output values have been removed and made as arguments.
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Linear and Logistic Regressions - Rewrite
Srinivasan Kannan
More... over 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 694. (FEATURE) Linear and Logistic Regressions - Rewrite - 4 August 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1.LinearAndLogisticRegression.py has been rewritten to accept arbitrary number of weights and variables for computing Linear and Logistic Regressions. 2.Linear And Logistic Regressions are estimators of Intrinsic Economic Merit of Nations in Economic Networks. Thus far following classes of intrinsic merit have been researched and implemented in NeuronRain: 1. Text 2. Audio - Speech 3. Audio - Music 4. Visuals - Images 5. Visuals - Videos 6. People - Experiential Intrinsic Merit 7. People - Log Normal Least Energy Merit - Education/Wealth/Valour e.g Sports IPR (Elo ratings in Chess, Player rankings in sports) 8. People - HR Analytics - Attrition model and Economic merit of nations is the 9th class of intrinsic merit applicable to Economic Networks. Regression is one amongst the set of econometric techniques for ascertaining growth of economies: 9. Nations - GDP, HDI etc., 3. Logs for the previous are at testlogs/LinearAndLogisticRegression.log.4August2019
Srinivasan Kannan
More... over 7 years ago
Updated Best Practices
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Time Series Analysis and Fraud Analytics - Credit Card Transactions Dataset
Srinivasan Kannan
More... over 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 693. (FEATURE) Time Series Analysis and Fraud Analytics - Credit Card Transactions Dataset - 3 August 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1.FraudAnalytics.py has been updated to import TimeSeries ARMA and ARIMA functions from Streaming_TimeSeriesData.py and compute ARMA and ARIMA polynomials on amounts withdrawn timeseries in credit card transactions dataset. 2.Logs for these projections are at testlogs/FraudAnalytics.log.3August2019 3.Though timeseries is a perfect fit for analyzing credit card amounts withdrawn as timeseries, projection curves for ARMA and ARIMA in logs are skewed widely for multiple iterations of random weights in autoregression and moving averages but values of ARMA and ARIMA are uniformly same almost.
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Time Series Analysis - ARMA and ARIMA implementations - Rewrite 2
Srinivasan Kannan
More... over 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 692. (FEATURE) Time Series Analysis - ARMA and ARIMA implementations - Rewrite 2 - 3 August 2019 --------------------------------------------------------------------------------------------------------------------------------------------- 1. ARMA and ARIMA functions in Streaming_TimeSeriesData.py have been again refactored to implement lag operator and binomial expansion of lag factor in ARIMA polynomial 2. autoregression_factored() has been rewritten to invoke a new function lag_factor_binomial_expansion() which in turn expands the factor power term in ARIMA polynomial by binomial series expansion of (1-L)^d. 3. lag_factor_binomial_expansion() loop invokes functions binomial_term() which computes the numerator and denominator factorial of each binomial term and lag() function for each binomial term. 4. Example logs for Stock quotes timeseries data is at Streaming_StockMarketData.ARIMATimeSeries.log.3August2019 which has comparative projections of rewritten ARMA and ARIMA functions. 5. ARIMA has been invoked by parameters (2,2,2)
Srinivasan Kannan
More... over 7 years ago
(THEORY and FEATURE) Bounded Error Factorization (Approximate Factors),
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Time Series Analysis - ARMA and ARIMA implementations - Rewrite
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 691. (FEATURE) Time Series Analysis - ARMA and ARIMA implementations - Rewrite - 2 August 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1. ARMA and ARIMA related functions in Streaming_TimeSeriesData.py have been rewritten based on ARMA and ARIMA polynomials in https://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average 2. Projection iterations have been removed and instead are done in the invoking source file e.g Streaming_StockMarketData.py for randomly chosen regression weights 3. ARMA polynomial is computed as: arma=timeseries[len(timeseries)-1] - autoregression(time_series_data[pprime:]) - moving_averages(time_series_data[q:], 5) which subtracts the autoregression and moving averages from the latest timeseries point. ARMA parameters pprime and q are sliced to be as distant from the latest timeseries point 4. ARIMA polynomial is computed as: arima=autoregression_factored(time_series_data[len(time_series_data)-d:],p) - moving_averages(time_series_data[q:], 5) which computes autoregression_factored differently as: 5.Lag operator for lag d is applied to slice the timeseries data from latest point t to d as t-d and autoregression_factored() is invoked to compute regressions from slice t-d-p to t-d 6.Finally the autoregression is subtracted from latest timeseries point Xt and moving averages is added to it 7.Logs for rewrite are available for few iterations at testlogs/Streaming_StockMarketData.ARIMATimeSeries.log.2August2019 8.R+rpy2 function plots have been commented
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Time Series Analysis - ARIMA implementation
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 690. (FEATURE) Time Series Analysis - ARIMA implementation - 1 August 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1.Time Series implementation has been made to a new python source file containing time series algorithms implementations for ARMA and ARIMA - Streaming_TimeSeriesData.py 2.Auto Regressive Integrated Moving Averages (ARIMA) has been implemented as a function ARIMA in Streaming_TimeSeriesData.py and factoring has been implemented for a parameter lag in autoregression_factored() 3.ARIMA differs from ARMA in factoring the autoregression polynomial into two factor polynomials one which is dependent on a lag operator. 4.Because Yahoo finance and Google Finance stock data are no longer available, ystockquote does not work and throws urllib exception and hence a hardcoded array of timeseries stock close data has been used. 5.ARMA and ARIMA functions from Streaming_TimeSeriesData.py are invoked in Streaming_StockMarketData.py. 6.Logs for ARIMA and R+rpy2 graphics plot have been committed to DJIA_ARIMA_Time_Series.pdf and testlogs/Streaming_StockMarketData.ARIMATimeSeries.log.1August2019
Srinivasan Kannan
More... over 7 years ago
Approximate Polynomials and Approximate Degree of Boolean Functions, Algebraic Circuits and KRW Conjecture for Boolean composition of Majority and VoterSAT/VoterPTF
Srinivasan Kannan
More... over 7 years ago
(FEATURE) Fraud Analytics - Credit Card Dataset - Pandas DataFrame Correlation Coefficient Matrix
Srinivasan Kannan
More... over 7 years ago
-------------------------------------------------------------------------------------------------------------------------------------------- 689. (FEATURE) Fraud Analytics - Credit Card Dataset - Pandas DataFrame Correlation Coefficient Matrix - 28 July 2019 -------------------------------------------------------------------------------------------------------------------------------------------- 1.Fraud Analytics implementation has been updated to instantiate Pandas DataFrame from credit card dataset and print Pandas Correlation matrix for all the columns. 2.Columns V1-V28 are Obfuscated Names of the cardholders which are numeric unique id(s) 3.Correlation between unique identities/names and amount drawn is a behavioural pattern and any deviation is an alarm (most swipes are periodic and personality oriented). 4.Correlation matrix logs are committed to testlogs/FraudAnalytics.log.28July2019 5.Correlation between Timestamp and Amount drawn is also measure of fraud.
Srinivasan Kannan
More... over 7 years ago
(THEORY) Approximate Polynomials and Approximate Degree of Boolean Functions, Algebraic Circuits and KRW Conjecture for Boolean composition of Majority and VoterSAT/VoterPTF
Srinivasan Kannan
More... over 7 years ago
(THEORY) Approximate Polynomials and Approximate Degree of Boolean Functions, Algebraic Circuits and KRW Conjecture for Boolean composition of Majority and VoterSAT/VoterPTF
Srinivasan Kannan
More... over 7 years ago
(THEORY) Approximate Polynomials and Approximate Degree of Boolean Functions, Algebraic Circuits and KRW Conjecture for Boolean composition of Majority and VoterSAT/VoterPTF
Srinivasan Kannan
More... over 7 years ago
(THEORY) Approximate Polynomials and Approximate Degree of Boolean Functions, Algebraic Circuits and KRW Conjecture for Boolean composition of Majority and VoterSAT/VoterPTF
Srinivasan Kannan
More... over 7 years ago
(THEORY) Approximate Polynomials and Approximate Degree of Boolean Functions, Dual Polynomials, Hardness Amplification, KRW Conjecture, Boolean Composition of Majority and VoterSAT/VoterPTF, Communication Complexity, Linear Program Primal/Dual
Srinivasan Kannan
More... over 7 years ago
(FEATURE) People Analytics - Kaggle LinkedIn DataSet - Experiential and Degree Intrinsic Merit
Srinivasan Kannan
More... over 7 years ago
------------------------------------------------------------------------------------------------------------------------------------------- 686. (FEATURE) People Analytics - Kaggle LinkedIn DataSet - Experiential and Degree Intrinsic Merit - 27 July 2019 ------------------------------------------------------------------------------------------------------------------------------------------- 1.In continuation to previous commit for linkedin dataset analytics, this commit implements computation of lognormal experiential intrinsic merit and degree based experiential intrinsic merit for all profiles in linkedin dataset. 2.Pandas DataFrame has been expanded to include two new columns - lognormal_experiential_merits and degree_experiential_merits - and correlation matrix is computed for this 7 rows * 7 columns. 3.Experience of the n-th profile is derived from dataset columns as: experience = avg_n_pos_per_prev_tenure[n] * avg_pos_len[n] * n_prev_tenures[n] + tenure_len[n] 4.Correlation coefficients for all pairs of 7 variables from logs:
Srinivasan Kannan
More... over 7 years ago