prashant-jonny / accord Goto Github PK

One-Way ANOVA doesn't works with different sample sizes. This is already fixed 
on the development trunk and should be available on 2.7.1

The Binomial test yields an incorrect P-Value for the Binomial Two-Tailed test 
but is correct for both One Tailed results compared to SciPy and R.

What steps will reproduce the problem?

1. use Constructor: BinomialTest(int successes, int trials, double    
hypothesizedProbability, OneSampleHypothesis alternate);
2. No. of trials 235
3. No. of successes 51
4. HypothesizedProbability 1/6


What is the expected output? What do you see instead?
Two Tailed
Statistic = 51
Accord.Net = 1.96404518910672
R = 0.04375
SciPy= 0.043747970182413443

What version of the product are you using? On what operating system?

v2.8.2

Please provide any additional information below.

I've already tried to submit issue but seems like it was timeout, so - once 
again:

1. corner detector type - FastCornersDetector
2. feature detector type - FastRetinaKeypointDetector
3. point matcher type - KNearestNeighborMatching<byte[]>(2, 
Distance.BitwiseHamming)

a bit of code:

cornerDetector.Threshold = 18;
List<FastRetinaKeypoint> descriptors = freakDetector.ProcessImage(newImage);
var union1 = descriptors.ToArray(); // length == 2, got it from newImage 
(attached)
var union2 = lastFeaturePoints1; // length == 1, got it from oldImage (attached)
var knMatches = matcher.Match(union2, union1); // exception here

Exception details:
System.IndexOutOfRangeException was unhandled by user code
  HResult=-2146233080
  Message=Index was outside the bounds of the array.
  Source=Accord.Math
  StackTrace:
       at Accord.Math.Matrix.Bottom[T](T[] values, Int32 count, Boolean inPlace) in C:\Work\Accord.NET\Sources\Accord.Math\Matrix\Matrix.Selection.cs:line 1657
       at Accord.MachineLearning.KNearestNeighbors`1.Compute(T input, Double[]& scores) in C:\Work\Accord.NET\Sources\Accord.MachineLearning\KNearestNeighbors`1.cs:line 326
       at Accord.MachineLearning.KNearestNeighbors`1.Compute(T input, Double& response) in C:\Work\Accord.NET\Sources\Accord.MachineLearning\KNearestNeighbors`1.cs:line 305
       at Accord.Imaging.KNearestNeighborMatching`1.Match(IFeaturePoint`1[] points1, IFeaturePoint`1[] points2) in C:\Work\Accord.NET\Sources\Accord.Imaging\KNearestNeighborMatching.cs:line 128
       at Exonya.EdgeDetector.Accord.MainWindow.ProcessImageFREAK(UnmanagedImage rgbImage) in c:\Work\Exonya.EdgeDetector1\Exonya.EdgeDetector.Accord\MainWindow.xaml.cs:line 274
       at Exonya.EdgeDetector.Accord.MainWindow.video_NewFrame(Object sender, NewFrameEventArgs eventArgs) in c:\Work\Exonya.EdgeDetector1\Exonya.EdgeDetector.Accord\MainWindow.xaml.cs:line 130
       at AForge.Video.DirectShow.VideoCaptureDevice.OnNewFrame(Bitmap image)
       at AForge.Video.DirectShow.VideoCaptureDevice.Grabber.BufferCB(Double sampleTime, IntPtr buffer, Int32 bufferLen)
  InnerException: 



Windows 7, Accord.NET 2.9.0.0


thanks

The C4.5 may end with a IndexOutOfRange after the 2.8.2 update.

What steps will reproduce the problem?
1.
2.
3.

What is the expected output? What do you see instead?


What version of the product are you using? On what operating system?

v2.8.1
Please provide any additional information below.

i have downloaded the rar file. how to include these dlls in the project/how to 
install them please help

Add support for probit regression using the already existent LinkFunctions.

What steps will reproduce the problem?
Issue linked from StackOverflow post:
http://stackoverflow.com/questions/13052167/using-accord-nets-codification-objec
t-to-codify-second-data-set/


What is the expected output? What do you see instead?
Expected Result is "lorem"


Please provide any additional information below.

Attached is a source code file which reproduces the issue. 

It is entirely possible I am doing something wrong.

Hello,

I'm using Accord.NET framework to recognize some patterns using 
MulticlassSupportVectorMachine class.

Here is my code:

Parallel.For(0, list.Length, i =>
            {

                // .. load some stuff ..
                // here we get a NullReferenceException inside createCache() method.
                int num = SupportVectorMachine.Compute(data, config.ComputeMethod, out responses);

                if (!SupportVectorMachine.IsProbabilistic)
                {
                    // Normalize responses
                    double max = responses.Max();
                    double min = responses.Min();

                    responses = Accord.Math.Tools.Scale(min, max, 0, 1, responses);
                }
                // .. do something ..
                }
            });


If I run this Parallel.For loop in a normal way (e.g using for or foreach), the 
methods runs fine.

Can anyone help me, please?

Hi,

As far as I know (maybe it would need more research) the FREAK detector is more 
powerfull, faster and consume less memory than SURF and SIFT. 

Also, I don't think it is patented because in openCV it is not included in the 
"nonfree" DLL. 

It could be a very good alternative for users who want to use Accord.NET in 
commercial application but can't because most of the very powerful algorithm 
use SURF. All of this could be replaced by the FREAK detector.

Here is a comparison of 3 detector and FREAK is the winner:

http://computer-vision-talks.com/2012/08/a-battle-of-three-descriptors-surf-frea
k-and-brisk/

What steps will reproduce the problem?
1. Create a new GridSearch object with 2*2 parameters
2. Specify a Fitting function throwing a ConvergenceException once. Other calls 
return a normal double value e.g [0.2, 0.5, 0.1]
3. Call Compute() on the GridSearch object

What is the expected output? What do you see instead?
As the best error rate in the received GridSearchResult 0.1 is expected. 
Instead, the Error property of the result is set to double.NaN

What version of the product are you using? On what operating system?
Accord 2.2.8


Please provide any additional information below.

double.NaN is set as error rate if a ConvergenceException is thrown while 
computing the Fitting function. This error rate is added to the errors array. 
At the end, the best error rate is estimated by the Min function of the System 
Library. If a value in the error array is set to double.NaN, this rate is 
selected as the Min value

 var doubles = new[] {0.5, double.NaN, 1.0, 3.0};
 Console.WriteLine(doubles.Min()); 
 //writes "n. Def"

Please describe the new feature you are interested and how it would
contribute to the project.

Trimmed mean option parameter the Levene's Test. It is a common parameter of 
the Levene's test in many statistical libraries.

Add examples for the PCA/LDA analysis using images.

What steps will reproduce the problem?
1. Try to traverse automatically a KDtree<double> using the foreach(...).
2. The output is empty.


What is the expected output? What do you see instead?
A list of KDTreeNode<double>.

What version of the product are you using? On what operating system?
2.8.1

Please provide any additional information below.
Looking at the source code, the condition stack.Count == 0 is wrong. it should 
be stack.Count != 0.

What steps will reproduce the problem?
1. create the GoldfarbIdnaniQuadraticSolver with the following constraints

A   B   C   D   Op  Val
0.0732  0.0799  0.1926  0.0047  =   0.098
1   1   1   1   =   1
1   0   0   0   >=  0
0   1   0   0   >=  0
0   0   1   0   >=  0
0   0   0   1   >=  0
1   0   0   0   >=  0.5

2. minimize with
Q:
0.12264004  0.011579293 0.103326825 0.064073439
0.011579293 0.033856    0.014311947 0.014732381
0.103326825 0.014311947 0.17715681  0.067615114
0.064073439 0.014732381 0.067615114 0.11539609
and 
d:
0   0   0   0

What is the expected output? What do you see instead?
the expected results are
A   B   C   D
0.5 0.336259542 0.163740458 0

The output are
A   B   C   D
0.5 0.3996  0.1543  -0.0539

What version of the product are you using? On what operating system?
the version is 2.8.1.0 Last Published on 12/21/2012
OS is Window XP

Please provide any additional information below.
development Tools is Visual Studio 2010 Professional
more Properties from the solver:
Deletions: 2
Iterations: 7
Lagrangian[0]: 0.47610163141385442
Lagrangian[1]: 0.018357076997984511
Lagrangian[2]: 0.0
Lagrangian[3]: 0.0
Lagrangian[4]: 0.0
Lagrangian[5]: 0.057217556502738381
Lagrangian[6]: 0.057966723535120945
Solution[0]: 0.5
Solution[1]: 0.39959370168854663
Solution[2]: 0.15434219180400793
Solution[3]: -0.053935893492554586
Value: 0.03092710642619409

ID3 (and possibly C4.5) generate a leaf node with Null output if a discrete 
variable is said to have multiple possible values but some of the possible 
values do not have any corresponding output. 

It should be investigated if those algorithms handle this case and implement 
the solution (use the mode?) appropriately.

'''What steps will reproduce the problem?'''
* 1.Create two images with fully transparent areas that can be combined
* 2.the first image blends properly
* 3.the second images alpha channel is blended in. if its fully transparent 
those bits should not copy

'''What is the expected output? What do you see instead?'''
100% transparent bits should not blend int

'''What version of the product are you using? On what operating system?'''
Accord 2.1.4

'''Please provide any additional information below:'''
Attached code fixes for 100% alpha, but doesn't attempt anything fancy for 
partial transparency
{{{
                        // validate source pixel's coordinates
                        if ((ox >= 0) && (oy >= 0) && (ox < width) && (oy < height))
                        {
                            int c = oy * srcStride + ox * srcPixelSize;

                            // fill destination image with pixel from source image
                            if (srcPixelSize == 4 && src[c + 3] == 0)
                            {
                                // nothing to copy in
                                //NOP
                            }
                            else if (dst[3] > 0)
                            {
                                // there is a pixel from the other image here, blend
                                double d1 = distance(ox, oy, center1.X, center1.Y);
                                double d2 = distance(ox, oy, center2.X, center2.Y);
                                double f = Accord.Math.Tools.Scale(0, dmax, 0, 1, d1 - d2);

                                if (f < 0) f = 0;
                                if (f > 1) f = 1;
                                double f2 = (1.0 - f);

                                dst[0] = (byte)(src[c] * f2 + dst[0] * f);
}}}

KNearestNeighbor should not assume zero-based consecutive class labels. If the 
labels (the outputs parameter of the constructor) happens to be something like 
{4, 2, 5} (does not have zero and does not include all labels from 0 - 5), the 
method Compute crashes.

Attached is a modified version of the class that works with this use case 
without the problem.

User code contribution of Dynamic Time Warping by user david_r

I think Transform method should be like this.

Two small fixes: removed dependence on Accord.Audio.Formats (as it is part of 
Accord.Audio) and fixed the package name to Accord.DirectAudio. Without these 
fixes, package cannot be installed via NuGet.

What steps will reproduce the problem?
1. Instead of example.xls file i want to attach my excel file to the SVM 
Classifier, When i try to attach my file (attached), the problem i had faced is 
the dataTable object is unable to typecast or read it properly to store in type 
double Matrix. Please suggest me is there any specific file format for 
example.xls as both sheet include 3 column X, Y and G.

What is the expected output? What do you see instead?
I need that SVM will read a generic xls file, with a specified file format of 
column.

What version of the product are you using? On what operating system?
Window 7 , Latest Version.

Please provide any additional information below.

In order to use some models with categoric data, such as logistic regression, 
it is first necessary to convert categories into factors. It would be nice to 
have a more easier or standard way to create such variables.

These lines aren't needed (does nothing).

What steps will reproduce the problem?
1.create an array of imputs for more than 10000 values in a row 
2.more than 500 rows 
3.SequentialMinimalOptimization

What is the expected output? What do you see instead?
out of memory exeption
Locating source for 
'D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\M
ulticlassSupportVectorLearning.cs'. Checksum: MD5 {56 e9 4d 58 59 a2 7e ab 57 5 
af dc 91 88 72 a6}
The file 
'D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\M
ulticlassSupportVectorLearning.cs' does not exist.
Looking in script documents for 
'D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\M
ulticlassSupportVectorLearning.cs'...
Looking in the projects for 
'D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\M
ulticlassSupportVectorLearning.cs'.
The file was not found in a project.
Looking in directory 'C:\Program Files (x86)\Microsoft Visual Studio 
11.0\Common7\IDE\vc7\atlmfc'...
Looking in directory 'C:\Program Files (x86)\Microsoft Visual Studio 
11.0\Common7\IDE\vc7\crt'...
The debug source files settings for the active solution indicate that the 
debugger will not ask the user to find the file: 
D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\Mu
lticlassSupportVectorLearning.cs.
The debugger could not locate the source file 
'D:\Projects\Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\M
ulticlassSupportVectorLearning.cs'.

What version of the product are you using? On what operating system?
windows 7 .net 4.5 , accord version 2.7.1

Please provide any additional information below.
It only gives me an error when i add more than about 500 samples (inputs)Error 
is from visual studio 12 express

Ensure BagOfWords and BagOfVisualWords are fully serializable.

Evaluate the feasibility of producing an Android compatible version using 
Mono/Xamarin.

What steps will reproduce the problem?
1.Start Visual C# 2010 and Open the Samples.sln installed in C:/Program 
Files/Accord.Net/Framework/Samples and Launch the solver Project
2.The Form shows Objective Min 2x² + xy + y² - 5y.
 Put the constraints as :
-x - 3y >= -2
-x - y >= 0
x >= 0
y >= 0

3.Click Compute

What is the expected output? What do you see instead?
the only feasible point is x=0,y=0 which should be the solution

the solution given by the Form is: 
Solution:

 2x² + xy + y² - 5y = 0.0947319461552687

 x = -0.44079027173455
 y = 0.44079027173455

which does not respect the constraint x>=0


What version of the product are you using? On what operating system?
I am using the version 2.6.1 of the library Accord.Net on windows XP and using 
visual c# 2010

Please provide any additional information below.

What steps will reproduce the problem?
1. Variances and Degrees of Freedom a calculated from the attached data sets 
Sample 1 (Random_50_0.5.txt) has a variance of 1.05766555271071 and DoF of 49. 
and Sample 2 (Random_50_0.7.txt) has a variance of 1.16570834301777 and DoF of 
49. 

2. Enter data in Constructor FTest(Sample1.Variance, Sample2.Variance, 
Sample1.DoF, Sample2.DoF, TwoSampleHypothesis.ValuesAreDifferent);

3. Compare these with One Tailed Tests using Constructors: 
One Tailed - Greater:
FTest(Sample1.Variance, Sample2.Variance, 
Sample1.DoF, Sample2.DoF, TwoSampleHypothesis.FirstValueIsGreaterThanSecond);
One Tailed - Less:
FTest(Sample1.Variance, Sample2.Variance, 
Sample1.DoF, Sample2.DoF, TwoSampleHypothesis.FirstValueIsSmallerThanSecond));

What is the expected output? What do you see instead?

Results: 
Two Tailed p-value: 1.26772493058357
One Tail Greater p-value: 0.633862465291784
One Tail Less p-value: 0.366137534708217

For the two tailed test I would have expected 0.732275069416434 (i.e. double 
the One Tail Less, this value was confirmed by R and SciPy).

What version of the product are you using? On what operating system?

Windows8 (Visual Studio 2008), 64.bit, Accord.Net Version 2.8.0

Please provide any additional information below.

Add backward compatibility to the .NET Framework v3.5

What steps will reproduce the problem?
1.
2.
3.

What is the expected output? What do you see instead?
The SVM problem when used for Regression set, it has some problem

What version of the product are you using? On what operating system?
window 7

Please provide any additional information below.

The learning of Logistic Regression models is done in the same fashion as 
AForge.NET's Neural Networks: a Run method has to be called to run iterations 
(or batch of iterations) manually, until the user decides convergence has been 
reached.

The Cox's models, the Hidden Markov Models and Hidden Conditional Random 
Fields, on the other hand, control the iteration themselves. You specify the 
number of iterations and tolerance threshold and call them once.

It is needed to decide which is most appropriated and enforce this as a common 
standard for all learning methods.

It would be useful to have an implementation of the implicitly restarted 
Lanczos method algorithm (http://en.wikipedia.org/wiki/Lanczos_algorithm). It 
could improve the efficiency of methods which require only the largest or the 
first largest Eigenvalues to be computed.

It would be interesting to add a feature extractor for computing histogram of 
oriented gradients in the Accord.Imaging namespace.

[http://accord-net.origo.ethz.ch/node/184 Dan originally wrote]:

In the correlationMatching.cs, there is a formula for Normalized Cross 
Correlation matrix like:


                   for (int i = 0; i < windowSize; i++)
                   {
                       for (int j = 0; j < windowSize; j++)
                       {
                           sum1 += w1[i, j] * w2[i, j];
                           sum2 += w2[i, j] * w2[i, j];
                       }
                   }
                   matrix[n1, n2] = sum1 / System.Math.Sqrt(sum2);


But in reference pages, I found that sum2 should be like : sum2 += w1[i, j] * 
w1[i, j] * w2[i, j] * w2[i, j];

It would be interesting to have implementations of the AdaBoost learning 
algorithm, specially if it could be used to learn Haar Cascade Classifiers 
which could be used in the Vision namespace.

What steps will reproduce the problem?
1.
2.
3.

What is the expected output? What do you see instead?


What version of the product are you using? On what operating system?


Please provide any additional information below.

getting started gives the information about how to install executable installer 
and not about how to install rar file. please help me. I am in urgent need of 
it.

What steps will reproduce the problem?
Running a PLS-Regression with 3 dependent variables and 12 predictor variables, 
and then computing the MultivariateLinearRegression obtained from the PLS.

What is the expected output? What do you see instead?
It should compute the regression coheficients correctly, instead it crashes 
with a index out of range exception in the compute method.

What version of the product are you using? On what operating system?
v4.7 on Windows 7

Please provide any additional information below.
It seems you are taking the wrong dimensions in the compute function of the 
MultivariateLinearRegression.
int M = coefficients.GetLength(1); 
it should be:
int M = coefficients.GetLength(0);

Some classes from the Analysis namespace, such as LogisticRegressionAnalysis, 
do not have source code examples.

It should be OK to add support for discrete-valued observations, but it will 
also be necessary to make a notice that this may decrease performance.

What steps will reproduce the problem?
1. Used the following data:

       List<double[]> _samples = new List<double[]>();

        double[] A = { 250, 260, 230, 270 };
        double[] B = { 310, 330, 280, 360 };
        double[] C = { 250, 230, 220, 260 };
        double[] D = { 340, 270, 300, 320 };
        double[] E = { 250, 240, 270, 290 };
            _samples.Add(A);
            _samples.Add(B);
            _samples.Add(C);
            _samples.Add(D);
            _samples.Add(E);

     LeveneTest result = new LeveneTest(_samples.ToArray(), true);

What is the expected output? What do you see instead?

Expect Output for result: 
statistic= ~0.7247
pValue = ~0.5886
This was outputted by R (Library: lawstat, levene.test) and propriety 
statistics software.


Actual Output for result:
statistic= ~0.7247
pValue = ~0.8228


What version of the product are you using? On what operating system?

2.8.1

Please provide any additional information below

Data Works fine for the newly added Bartlett's test

Here's code for Accord.MachineLearning.DecisionTrees.Learning.computeInfoGain:
{
  threshold = 0;
  if (tree.Attributes[attributeIndex].Nature == DecisionAttributeKind.Discrete)
    return entropy - computeInfo(input, output, attributeIndex, out partitions);

  return entropy - computeInfo(input, output, attributeIndex, out partitions, out threshold);
}

when the attribute is continous, split gain is SUBSTRACTED from entropy, but 
the method that computes split gain for continous attributes (computeInfo) 
already returns NEGATIVE value, so split gain should be ADDED to entropy. 
Otherwise the c4.5 algorithm uses the worst of best possible splits.
so the last line should be as follows:

return entropy + computeInfo(input, output, attributeIndex, out partitions, out 
threshold);

What steps will reproduce the problem?
1.Install Accord & AForge
2.Create project and add reference to the Accord.dll
3. The interfaces aren't recognized the the version information is 0.0.0

What is the expected output? What do you see instead?
It should show the versioning information in the reference. Instead I get a 
message the the file cannot be referenced

What version of the product are you using? On what operating system?
Win 7

Please provide any additional information below.

What steps will reproduce the problem?
1. I created an Id3Learning tree
2. After using id3Learning.Run() I attempted to serialize the tree.

What is the expected output? What do you see instead?
The program threw an exception stating that DecisionAttributeCollection is not 
serializeable.


What version of the product are you using? On what operating system?
4.0.30319 on Windows 7 64 bit

Please provide any additional information below.
I was able to fix the error by adding the [Serializeable] flag to the 
DecisionAttributeCollection class.

Add minimum neighbors threshold and rectangle averaging in the Haar object 
detector.

What steps will reproduce the problem?

1.linear filtering

2.cross scale center-surround differences

3.winner-take-all networks

What is the expected output? What do you see instead?
saliency maps can be obtained.

What version of the product are you using? On what operating system?

I am using SaliencyToolbox2.2 which implemented by matlab code on widowns 7.

Please provide any additional information below.

[1] L.Itti, C.Koch, E. Niebur, A Model of Saliency-Based Visual Attention for 
Rapid Scene Analysis, IEEE Transactions on Pattern Analysis and Machine 
Intelligence 20(11) (1998)1254-1259.

[2] http://ilab.usc.edu/toolkit/

need to change the indexing variable in the Translate method as follows:

        public int[] Translate(params string[] data)
        {
            int[] result = new int[data.Length];

            for (int i = 0; i < data.Length; i++)
            {
                for (int j = 0; j < Columns.Count; j++)
                {
                    //Options options = this.Columns[i];
                    Options options = this.Columns[j]; // new index***
                    if (options.Mapping.TryGetValue(data[i], out result[i]))
                        break;
                }
            }

            return result;
        }

Thanks for sharing your project.... very helpful!
Scott

What steps will reproduce the problem?
1. Use a SVM algorithm with SMO and Bootstrapping
2. At some point Bootstrapping can select all negative or all positive samples. 
This breaks the code in 
Accord.NET\Sources\Accord.MachineLearning\VectorMachines\Learning\SequentialMini
malOptimization.cs at line 479
3. It tries to find an upper and lower bound for the + en - samples but if 
there are no + samples, an IndexOutOfRangeException is thrown here. 

What is the expected output? What do you see instead?
I'm not too sure what the algorithm is supposed to do in this case. At least 
catch the Exception?

What version of the product are you using? On what operating system?
Accord.NET v2.7.0 on Windows Vista

Please provide any additional information below.

What steps will reproduce the problem?

1. Decision tree with 6 continuous variables, 1 discrete  
2. Learn the tree
3. Exception at C45Learning.cs:274

The reason for this is that the maxGainPartition array contains only one value, 
and the algorithm expects two. Causes an IndexOutOfBounds exception.

It seems all partitions are assumed to be split in two but for some strange 
reason my data produces one partition with only one subset. 

What version of the product are you using? On what operating system?

2.8.1, Windows 7, 64-bit.

Attached is my training data in a flat format. First column is irrelevant, last 
column is output (yes/no). All except the penultimate column are continuous 
decision variables.

What steps will reproduce the problem?

1. Use HaarObjectDetector class to detect objects in an image
2. Set the MinSize property to something larger than the default
3. Observe that rectangles returned by the ProcessFrame method can be smaller 
than the value of MinSize

What is the expected output? What do you see instead?

I would expect that only rectangles (indicating objects discovered) that are 
larger than MinSize are returned by ProcessFrame.

What version of the product are you using? On what operating system?

2.8.1 on both Windows 7 & Windows 8

Please provide any additional information below.

Changing line 448 of HaarObjectDetector.cs from: 

"if (window.Width < minSize.Width && window.Height < minSize.Height &&
                    window.Width > maxSize.Width && window.Height > maxSize.Height)"

to:

"if ((window.Width < minSize.Width || window.Height < minSize.Height))"

Solves the issue for me, although I'm not sure if this is in line with the 
original intent

            //Some description please