Entropy Calculation

I've been implementing a number of entropy encoding schemes. To analyze their performance. I wrote a class that can calculate the 'zero-order' information entropy of data. Because the data sets I've been working on are quite large, it needed to be fast. I couldn't find any code worth using for the task, so I produced a single Java class that does the job very effectively.

I have released the source code into the public domain:
CodingFrequencies.java

Usage Example

Using the class is very simple:

//define some data int[] values = {7, 7, 3, 3, 3, 2, 7}; //analyze its frequencies CodingFrequencies freqs = CodingFrequencies.fromValues(values); //outputs: 1.4488156357251847 System.out.println( freqs.binaryEntropy() ); //outputs: 3 System.out.println( freqs.getFrequency(7) ); //outputs: [1, 3, 3] System.out.println( Arrays.toString(freqs.getFrequencies()) );

The source code is fully documented. For more information see the comments in the supplied code.

Features and Characteristics

• The code has been designed to make frequency analysis as fast as possible in most cases. Further optimizations are possible, but the implementation should be fast enough for most purposes.
• In instances where only the frequency distribution is of interest (not its relationship with the actual values) a compact representation can be generated that reduces memory consumption and speeds up some methods.
• Entropy calculations can be made on both byte[] and int[] data.
• No restriction is placed on the range of int values that can be analyzed.
• After an instance of the CodingFrequencies class has been constructed, information entropy can be be returned in any base with no further computation