Calculate the block entropy

Question

I once needed to write a function that calculates the block entropy of a given symbol series for a given block size and was surprised how short the result was. Thus I am challenging you to codegolf such a function. I am not telling you what I did for now (and in which language), but I will in a week or so, if nobody came up with the same or better ideas first.

Definition of the block entropy:

Given a symbol sequence A = A_1, …, A_n and a block size m:

• A block of size m is a segment of m consecutive elements of the symbol sequence, i.e., A_i, …, A_(i+m−1) for any appropriate i.
• If x is a symbol sequence of size m, N(x) denotes the number of blocks of A which are identical to x.
• p(x) is the probability that a block from A is identical to a symbol sequence x of size m, i.e., p(x) = N(x)/(n−m+1)
• Finally, the block entropy for block size m of A is the average of −log(p(x)) over all blocks x of size m in A or (which is equivalent) the sum of −p(x)·log(p(x)) over every x of size m occurring in A. (You can choose any reasonable logarithm you want.)

Restricions and clarifications:

• Your function should take the symbol sequence A as well as the block size m as an argument.
• You may assume that the symbols are represented as zero-based integers or in another convenient format.
• Your program should be capable of taking any reasonable argument in theory and in reality should be able to calculate the example case (see below) on a standard computer.
• Built-in functions and libraries are allowed, as long as they do not perform big portions of the procedure in one call, i.e., extracting all blocks of size m from A, counting the number of occurrences of a given block x or calculating the entropies from a sequence of p values – you have to do those things yourself.

Test:

[2, 3, 4, 1, 2, 3, 0, 0, 3, 2, 3, 0, 2, 2, 4, 4, 4, 1, 1, 1, 0, 4, 1,
2, 2, 4, 0, 1, 2, 3, 0, 2, 3, 2, 3, 2, 0, 1, 3, 4, 4, 0, 2, 1, 4, 3,
0, 2, 4, 1, 0, 4, 0, 0, 2, 2, 0, 2, 3, 0, 0, 4, 4, 2, 3, 1, 3, 1, 1,
3, 1, 3, 1, 0, 0, 2, 2, 4, 0, 3, 2, 2, 3, 0, 3, 3, 0, 0, 4, 4, 1, 0,
2, 3, 0, 0, 1, 4, 4, 3]

The first block entropies of this sequence are (for the natural logarithm):

• m = 1: 1.599
• m = 2: 3.065
• m = 3: 4.067
• m = 4: 4.412
• m = 5: 4.535
• m = 6: 4.554

Answer 1

Mathematica - 81 78 75 72 67 65 62 56 bytes

I haven't golfed anything in Mathematica before, so I suppose there's room for improvement. This one doesn't quite conform to the rules due to the Partition and Tally functions, but it's quite neat so I thought I'd post it anyway.

f=N@Tr[-Log[p=#2/Length@b&@@@Tally[b=##~Partition~1]]p]&

This works with any set of symbols, and can be used like

sequence = {2, 3, 4, 1, 2, 3, 0, 0, 3, 2, 3, 0, 2, 2, 4, 4, 4, 1, 1, 
   1, 0, 4, 1, 2, 2, 4, 0, 1, 2, 3, 0, 2, 3, 2, 3, 2, 0, 1, 3, 4, 4, 
   0, 2, 1, 4, 3, 0, 2, 4, 1, 0, 4, 0, 0, 2, 2, 0, 2, 3, 0, 0, 4, 4, 
   2, 3, 1, 3, 1, 1, 3, 1, 3, 1, 0, 0, 2, 2, 4, 0, 3, 2, 2, 3, 0, 3, 
   3, 0, 0, 4, 4, 1, 0, 2, 3, 0, 0, 1, 4, 4, 3};
f[sequence, 3]

> 4.06663

Here is a somewhat ungolfed version:

f[sequence_, m_] := (
    blocks = Partition[sequence, m, 1];
    probabilities = Apply[#2/Length[blocks] &, Tally[blocks], {1}];
    N[Tr[-Log[probabilities]*probabilities]]
)

It will probably run faster if I apply N directly to the result of Tally.

By the way, Mathematica does actually have an Entropy function, that reduces this to 28 bytes, but that's definitely against the rules.

f=N@Entropy@Partition[##,1]&

On the other hand, here is a 128 byte version that reimplements Partition and Tally:

f=N@Tr[-Log[p=#2/n&@@@({#[[i;;i+#2-1]],1}~Table~{i,1,(n=Length@#-#2+1)}//.{p___,{s_,x_},q___,{s_,y_},r___}:>{p,{s,x+y},q,r})]p]&

Ungolfed:

f[sequence_, m_] := (
    n = Length[sequence]-m+1; (*number of blocks*)
    blocks = Table[{Take[sequence, {i, i+m-1}], 1},
                   {i, 1, n}];
    blocks = b //. {p___, {s_, x_}, q___, {s_, y_}, r___} :> {p,{s,x+y},q,r};
    probabilities = Apply[#2/n &, blocks, {1}];
    N[Tr[-Log[probabilities]*probabilities]]
)

Answer 2

Perl, 140 bytes

The following Perl script defines a function E that takes the symbol sequence, followed by segment size as arguments.

sub E{$m=pop;$E=0;%h=();$"=',';$_=",@_,";for$i(0..@_-$m){next
if$h{$s=",@_[$i..$i+$m-1],"}++;$E-=($p=s|(?=$s)||g/(@_-$m+1))*log$p;}return$E}

Ungolfed version with test

sub E { # E for "entropy"
    # E takes the sequence and segment size as arguments
    # and returns the calculated entropy.
    $m = pop;    # get segment size (last argument)
    $E = 0;      # initialize entropy
    %h = ();     # hash that remembers already calculated segments
    $" = ',';#"  # comma is used as separator
    $_ = ",@_,"; # $_ takes sequence as string, with comma as delimiters
    for $i (0 .. @_-$m) {
        $s = ",@_[$i..$i+$m-1],"; # segment
        next if$h{$s}++;          # check, if this segment is already calculated
        $p = s|(?=\Q$s\E)||g / (@_ - $m + 1); # calculate probability
             # N(x) is calculated using the substitution operator
             # with a zero-width look-ahead pattern
             # (golfed version without "\Q...\E", see below)
        $E -= $p * log($p); # update entropy
    }
    return $E
}

# Test

my @A = (
    2, 3, 4, 1, 2, 3, 0, 0, 3, 2, 3, 0, 2, 2, 4, 4, 4, 1, 1, 1, 0, 4, 1,
    2, 2, 4, 0, 1, 2, 3, 0, 2, 3, 2, 3, 2, 0, 1, 3, 4, 4, 0, 2, 1, 4, 3,
    0, 2, 4, 1, 0, 4, 0, 0, 2, 2, 0, 2, 3, 0, 0, 4, 4, 2, 3, 1, 3, 1, 1,
    3, 1, 3, 1, 0, 0, 2, 2, 4, 0, 3, 2, 2, 3, 0, 3, 3, 0, 0, 4, 4, 1, 0,
    2, 3, 0, 0, 1, 4, 4, 3
);

print "m = $_: ", E(@A, $_), "\n" for 1 .. @A;

Result:

m = 1: 1.59938036027528
m = 2: 3.06545141203611
m = 3: 4.06663334311518
m = 4: 4.41210802885304
m = 5: 4.53546705894451
m = 6: 4.55387689160055
m = 7: 4.54329478227001
m = 8: 4.53259949315326
m = 9: 4.52178857704904
...
m = 97: 1.38629436111989
m = 98: 1.09861228866811
m = 99: 0.693147180559945
m = 100: 0

Symbols:

The symbols are not restricted to integers, because pattern matching based on strings is used. The string representation of a symbol must not contain the comma, because it is uses as delimiter. Of course, different symbols must have different string representations.

In the golfed version, the string representation of the symbols should not contain specials characters of patterns. The additional four bytes \Q...\E are not needed for numbers.

Answer 3

Python 127 152B 138B

import math
def E(A,m):N=len(A)-m+1;R=range(N);return sum(math.log(float(N)/b) for b in [sum(A[i:i+m]==A[j:j+m] for i in R) for j in R])/N

Adjusted to not break the rules any more and have a slightly cuter algorithm. Adjusted to be smaller

Older version:

import math
def E(A,m):
 N=len(A)-m+1
 B=[A[i:i+m] for i in range(N)]
 return sum([math.log(float(N)/B.count(b)) for b in B])/N

My first ever Python script! See it in action: http://pythonfiddle.com/entropy

Answer 4

Python with Numpy, 146 143 Bytes

As promised, here is my own solution. It requires an input of non-negative integers:

from numpy import*
def e(A,m):
    B=zeros(m*[max(A)+1]);j=0
    while~len(A)<-j-m:B[tuple(A[j:j+m])]+=1;j+=1
    return -sum(x*log(x)for x in B[B>0]/j)

The disadvantage is that this bursts your memory for a large m or max(A).

Here is the mostly ungolfed and commented version:

from numpy import *
def e(A,m):
    B = zeros(m*[max(A)+1])          # Generate (max(A)+1)^m-Array of zeroes for counting.
    for j in range(len(A)-m+1):
        B[tuple(A[j:j+m])] += 1      # Do the counting by directly using the array slice
                                     # for indexing.
    C = B[B>0]/(len(A)-m+1)          # Flatten array, take only non-zero entries,
                                     # divide for probability.
    return -sum(x*log(x) for x in C) # Calculate entropy

Answer 5

MATLAB

function E =BlockEntropy01(Series,Window,Base )

%-----------------------------------------------------------
% Calculates BLOCK ENTROPY of Series
% Series: a Vector of numbers
% Base: 2 or 10 (affects logarithm of the Calculation)
% for 2 we use log2, for 10 log10
% Windows: Length of the "Sliding" BLOCK
% E: Entropy
%-----------------------------------------------------------
% For the ENTROPY Calculations
% http://matlabdatamining.blogspot.gr/2006/....
% 11/introduction-to-entropy.html
% BlogSpot: Will Dwinnell
%-----------------------------------------------------------
% For the BLOCK ENTROPY
% http://codegolf.stackexchange.com/...
% questions/24316/calculate-the-block-entropy
%-----------------------------------------------------------
% Test (Base=10)
% Series=[2, 3, 4, 1, 2, 3, 0, 0, 3, 2, 3, 0, ....
%     2, 2, 4, 4, 4, 1, 1, 1, 0, 4, 1,2, 2, 4, 0, ....
%     1, 2, 3, 0, 2, 3, 2, 3, 2, 0, 1, 3, 4, 4, 0, ....
%     2, 1, 4, 3,0, 2, 4, 1, 0, 4, 0, 0, 2, 2, 0, ....
%     2, 3, 0, 0, 4, 4, 2, 3, 1, 3, 1, 1,3, 1, 3, 1, ....
%     0, 0, 2, 2, 4, 0, 3, 2, 2, 3, 0, 3, 3, 0, 0, 4, ...
%     4, 1, 0,2, 3, 0, 0, 1, 4, 4, 3]';
%
% Results 
%
% Window=1: 1.599
% Window=2: 3.065
% Window=3: 4.067
% Window=4: 4.412
% Window=5: 4.535
% Window=6: 4.554
%-----------------------------------------------------------
n=length(Series);
D=zeros(n,Window); % Pre Allocate Memory
for k=1:Window;    D(:,k)=circshift(Series,1-k);end
D=D(1:end-Window+1,:); % Truncate Last Part
%
% Repace each Row with a "SYMBOL"
% in this Case a Number ...............
[K l]=size(D);
for k=1:K; MyData(k)=polyval(D(k,:),Base);end
clear D
%-----------------------------------------------------------
% ENTROPY CALCULATIONS on MyData
% following  Will Dwinnell
%-----------------------------------------------------------
UniqueMyData = unique(MyData);
nUniqueMyData = length(UniqueMyData);
FreqMyData = zeros(nUniqueMyData,1); % Initialization
for i = 1:nUniqueMyData
    FreqMyData(i) = ....
        sum(double(MyData == UniqueMyData(i)));
end
% Calculate sample class probabilities
P = FreqMyData / sum(FreqMyData);
% Calculate entropy in bits
% Note: floating point underflow is never an issue since we are
%   dealing only with the observed alphabet
if Base==10
    E= -sum(P .* log(P));
elseif BASE==2
    E= -sum(P .* log2(P));
else
end
end

WITH TEST SCRIPT 
%-----------------------------------------------------------
Series=[2, 3, 4, 1, 2, 3, 0, 0, 3, 2, 3, 0, ....
    2, 2, 4, 4, 4, 1, 1, 1, 0, 4, 1,2, 2, 4, 0, ....
    1, 2, 3, 0, 2, 3, 2, 3, 2, 0, 1, 3, 4, 4, 0, ....
    2, 1, 4, 3,0, 2, 4, 1, 0, 4, 0, 0, 2, 2, 0, ....
    2, 3, 0, 0, 4, 4, 2, 3, 1, 3, 1, 1,3, 1, 3, 1, ....
    0, 0, 2, 2, 4, 0, 3, 2, 2, 3, 0, 3, 3, 0, 0, 4, ...
    4, 1, 0,2, 3, 0, 0, 1, 4, 4, 3]';
Base=10;
%-----------------------------------------------------------
for Window=1:6
    E =BlockEntropy01(Series,Window,Base )
end