Related
Recently I needed to do weighted random selection of elements from a list, both with and without replacement. While there are well known and good algorithms for unweighted selection, and some for weighted selection without replacement (such as modifications of the resevoir algorithm), I couldn't find any good algorithms for weighted selection with replacement. I also wanted to avoid the resevoir method, as I was selecting a significant fraction of the list, which is small enough to hold in memory.
Does anyone have any suggestions on the best approach in this situation? I have my own solutions, but I'm hoping to find something more efficient, simpler, or both.
One of the fastest ways to make many with replacement samples from an unchanging list is the alias method. The core intuition is that we can create a set of equal-sized bins for the weighted list that can be indexed very efficiently through bit operations, to avoid a binary search. It will turn out that, done correctly, we will need to only store two items from the original list per bin, and thus can represent the split with a single percentage.
Let's us take the example of five equally weighted choices, (a:1, b:1, c:1, d:1, e:1)
To create the alias lookup:
Normalize the weights such that they sum to 1.0. (a:0.2 b:0.2 c:0.2 d:0.2 e:0.2) This is the probability of choosing each weight.
Find the smallest power of 2 greater than or equal to the number of variables, and create this number of partitions, |p|. Each partition represents a probability mass of 1/|p|. In this case, we create 8 partitions, each able to contain 0.125.
Take the variable with the least remaining weight, and place as much of it's mass as possible in an empty partition. In this example, we see that a fills the first partition. (p1{a|null,1.0},p2,p3,p4,p5,p6,p7,p8) with (a:0.075, b:0.2 c:0.2 d:0.2 e:0.2)
If the partition is not filled, take the variable with the most weight, and fill the partition with that variable.
Repeat steps 3 and 4, until none of the weight from the original partition need be assigned to the list.
For example, if we run another iteration of 3 and 4, we see
(p1{a|null,1.0},p2{a|b,0.6},p3,p4,p5,p6,p7,p8) with (a:0, b:0.15 c:0.2 d:0.2 e:0.2) left to be assigned
At runtime:
Get a U(0,1) random number, say binary 0.001100000
bitshift it lg2(p), finding the index partition. Thus, we shift it by 3, yielding 001.1, or position 1, and thus partition 2.
If the partition is split, use the decimal portion of the shifted random number to decide the split. In this case, the value is 0.5, and 0.5 < 0.6, so return a.
Here is some code and another explanation, but unfortunately it doesn't use the bitshifting technique, nor have I actually verified it.
A simple approach that hasn't been mentioned here is one proposed in Efraimidis and Spirakis. In python you could select m items from n >= m weighted items with strictly positive weights stored in weights, returning the selected indices, with:
import heapq
import math
import random
def WeightedSelectionWithoutReplacement(weights, m):
elt = [(math.log(random.random()) / weights[i], i) for i in range(len(weights))]
return [x[1] for x in heapq.nlargest(m, elt)]
This is very similar in structure to the first approach proposed by Nick Johnson. Unfortunately, that approach is biased in selecting the elements (see the comments on the method). Efraimidis and Spirakis proved that their approach is equivalent to random sampling without replacement in the linked paper.
Here's what I came up with for weighted selection without replacement:
def WeightedSelectionWithoutReplacement(l, n):
"""Selects without replacement n random elements from a list of (weight, item) tuples."""
l = sorted((random.random() * x[0], x[1]) for x in l)
return l[-n:]
This is O(m log m) on the number of items in the list to be selected from. I'm fairly certain this will weight items correctly, though I haven't verified it in any formal sense.
Here's what I came up with for weighted selection with replacement:
def WeightedSelectionWithReplacement(l, n):
"""Selects with replacement n random elements from a list of (weight, item) tuples."""
cuml = []
total_weight = 0.0
for weight, item in l:
total_weight += weight
cuml.append((total_weight, item))
return [cuml[bisect.bisect(cuml, random.random()*total_weight)] for x in range(n)]
This is O(m + n log m), where m is the number of items in the input list, and n is the number of items to be selected.
I'd recommend you start by looking at section 3.4.2 of Donald Knuth's Seminumerical Algorithms.
If your arrays are large, there are more efficient algorithms in chapter 3 of Principles of Random Variate Generation by John Dagpunar. If your arrays are not terribly large or you're not concerned with squeezing out as much efficiency as possible, the simpler algorithms in Knuth are probably fine.
It is possible to do Weighted Random Selection with replacement in O(1) time, after first creating an additional O(N)-sized data structure in O(N) time. The algorithm is based on the Alias Method developed by Walker and Vose, which is well described here.
The essential idea is that each bin in a histogram would be chosen with probability 1/N by a uniform RNG. So we will walk through it, and for any underpopulated bin which would would receive excess hits, assign the excess to an overpopulated bin. For each bin, we store the percentage of hits which belong to it, and the partner bin for the excess. This version tracks small and large bins in place, removing the need for an additional stack. It uses the index of the partner (stored in bucket[1]) as an indicator that they have already been processed.
Here is a minimal python implementation, based on the C implementation here
def prep(weights):
data_sz = len(weights)
factor = data_sz/float(sum(weights))
data = [[w*factor, i] for i,w in enumerate(weights)]
big=0
while big<data_sz and data[big][0]<=1.0: big+=1
for small,bucket in enumerate(data):
if bucket[1] is not small: continue
excess = 1.0 - bucket[0]
while excess > 0:
if big==data_sz: break
bucket[1] = big
bucket = data[big]
bucket[0] -= excess
excess = 1.0 - bucket[0]
if (excess >= 0):
big+=1
while big<data_sz and data[big][0]<=1: big+=1
return data
def sample(data):
r=random.random()*len(data)
idx = int(r)
return data[idx][1] if r-idx > data[idx][0] else idx
Example usage:
TRIALS=1000
weights = [20,1.5,9.8,10,15,10,15.5,10,8,.2];
samples = [0]*len(weights)
data = prep(weights)
for _ in range(int(sum(weights)*TRIALS)):
samples[sample(data)]+=1
result = [float(s)/TRIALS for s in samples]
err = [a-b for a,b in zip(result,weights)]
print(result)
print([round(e,5) for e in err])
print(sum([e*e for e in err]))
The following is a description of random weighted selection of an element of a
set (or multiset, if repeats are allowed), both with and without replacement in O(n) space
and O(log n) time.
It consists of implementing a binary search tree, sorted by the elements to be
selected, where each node of the tree contains:
the element itself (element)
the un-normalized weight of the element (elementweight), and
the sum of all the un-normalized weights of the left-child node and all of
its children (leftbranchweight).
the sum of all the un-normalized weights of the right-child node and all of
its chilren (rightbranchweight).
Then we randomly select an element from the BST by descending down the tree. A
rough description of the algorithm follows. The algorithm is given a node of
the tree. Then the values of leftbranchweight, rightbranchweight,
and elementweight of node is summed, and the weights are divided by this
sum, resulting in the values leftbranchprobability,
rightbranchprobability, and elementprobability, respectively. Then a
random number between 0 and 1 (randomnumber) is obtained.
if the number is less than elementprobability,
remove the element from the BST as normal, updating leftbranchweight
and rightbranchweight of all the necessary nodes, and return the
element.
else if the number is less than (elementprobability + leftbranchweight)
recurse on leftchild (run the algorithm using leftchild as node)
else
recurse on rightchild
When we finally find, using these weights, which element is to be returned, we either simply return it (with replacement) or we remove it and update relevant weights in the tree (without replacement).
DISCLAIMER: The algorithm is rough, and a treatise on the proper implementation
of a BST is not attempted here; rather, it is hoped that this answer will help
those who really need fast weighted selection without replacement (like I do).
This is an old question for which numpy now offers an easy solution so I thought I would mention it. Current version of numpy is version 1.2 and numpy.random.choice allows the sampling to be done with or without replacement and with given weights.
Suppose you want to sample 3 elements without replacement from the list ['white','blue','black','yellow','green'] with a prob. distribution [0.1, 0.2, 0.4, 0.1, 0.2]. Using numpy.random module it is as easy as this:
import numpy.random as rnd
sampling_size = 3
domain = ['white','blue','black','yellow','green']
probs = [.1, .2, .4, .1, .2]
sample = rnd.choice(domain, size=sampling_size, replace=False, p=probs)
# in short: rnd.choice(domain, sampling_size, False, probs)
print(sample)
# Possible output: ['white' 'black' 'blue']
Setting the replace flag to True, you have a sampling with replacement.
More info here:
http://docs.scipy.org/doc/numpy/reference/generated/numpy.random.choice.html#numpy.random.choice
We faced a problem to randomly select K validators of N candidates once per epoch proportionally to their stakes. But this gives us the following problem:
Imagine probabilities of each candidate:
0.1
0.1
0.8
Probabilities of each candidate after 1'000'000 selections 2 of 3 without replacement became:
0.254315
0.256755
0.488930
You should know, those original probabilities are not achievable for 2 of 3 selection without replacement.
But we wish initial probabilities to be a profit distribution probabilities. Else it makes small candidate pools more profitable. So we realized that random selection with replacement would help us – to randomly select >K of N and store also weight of each validator for reward distribution:
std::vector<int> validators;
std::vector<int> weights(n);
int totalWeights = 0;
for (int j = 0; validators.size() < m; j++) {
int value = rand() % likehoodsSum;
for (int i = 0; i < n; i++) {
if (value < likehoods[i]) {
if (weights[i] == 0) {
validators.push_back(i);
}
weights[i]++;
totalWeights++;
break;
}
value -= likehoods[i];
}
}
It gives an almost original distribution of rewards on millions of samples:
0.101230
0.099113
0.799657
I'm looking for a way of getting X points in a fixed sized grid of let's say M by N, where the points are not returned multiple times and all points have a similar chance of getting chosen and the amount of points returned is always X.
I had the idea of looping over all the grid points and giving each point a random chance of X/(N*M) yet I felt like that it would give more priority to the first points in the grid. Also this didn't meet the requirement of always returning X amount of points.
Also I could go with a way of using increments with a prime number to get kind of a shuffle without repeat functionality, but I'd rather have it behave more random than that.
Essentially, you need to keep track of the points you already chose, and make use of a random number generator to get a pseudo-uniformly distributed answer. Each "choice" should be independent of the previous one.
With your first idea, you're right, the first ones would have more chance of getting picked. Consider a one-dimensional array with two elements. With the strategy you mention, the chance of getting the first one is:
P[x=0] = 1/2 = 0.5
The chance of getting the second one is the chance of NOT getting the first one 0.5, times 1/2:
P[x=1] = 1/2 * 1/2 = 0.25
You don't mention which programming language you're using, so I'll assume you have at your disposal random number generator rand() which results in a random float in the range [0, 1), a Hashmap (or similar) data structure, and a Point data structure. I'll further assume that a point in the grid can be any floating point x,y, where 0 <= x < M and 0 <= y < N. (If this is a NxM array, then the same applies, but in integers, and up to (M-1,N-1)).
Hashmap points = new Hashmap();
Point p;
while (items.size() < X) {
p = new Point(rand()*M, rand()*N);
if (!points.containsKey(p)) {
items.add(p, 1);
}
}
Note: Two Point objects of equal x and y should be themselves considered equal and generate equal hash codes, etc.
As we know from programming, sometimes a slight change in a problem can
significantly alter the form of its solution.
Firstly, I want to create a simple algorithm for solving
the following problem and classify it using bigtheta
notation:
Divide a group of people into two disjoint subgroups
(of arbitrary size) such that the
difference in the total ages of the members of
the two subgroups is as large as possible.
Now I need to change the problem so that the desired
difference is as small as possible and classify
my approach to the problem.
Well,first of all I need to create the initial algorithm.
For that, should I make some kind of sorting in order to separate the teams, and how am I suppose to continue?
EDIT: for the first problem,we have ruled out the possibility of a set being an empty set. So all we have to do is just a linear search to find the min age and then put it in a set B. SetA now has all the other ages except the age of setB, which is the min age. So here is the max difference of the total ages of the two sets, as high as possible
The way you described the first problem, it is trivial in the way that it requires you to find only the minimum element (in case the subgroups should contain at least 1 member), otherwise it is already solved.
The second problem can be solved recursively the pseudo code would be:
// compute sum of all elem of array and store them in sum
min = sum;
globalVec = baseVec;
fun generate(baseVec, generatedVec, position, total)
if (abs(sum - 2*total) < min){ // check if the distribution is better
min = abs(sum - 2*total);
globalVec = generatedVec;
}
if (position >= baseVec.length()) return;
else{
// either consider elem at position in first group:
generate(baseVec,generatedVec.pushback(baseVec[position]), position + 1, total+baseVec[position]);
// or consider elem at position is second group:
generate(baseVec,generatedVec, position + 1, total);
}
And now just start the function with generate(baseVec,"",0,0) where "" stand for an empty vector.
The algo can be drastically improved by applying it to a sorted array, hence adding a test condition to stop branching, but the idea stays the same.
My task:
Create a program to copy a picture (given as input) using primitives only (like triangle or something). The program should use evolutionary algorithm to create output picture.
My question:
I need to invent an algorithm to create populations and check them (how much - in % - they match the input picture).
I have an idea; you can find it below.
So what I want from you: advice (if you find my idea not so bad) or inspiration (maybe you have a better idea?)
My idea:
Let's say that I'll use only triangles to build the output picture.
My first population is P pictures (generated by using T randomly generated triangles - called Elements).
I check by my fitness function every pictures in population and choose E of them as elite and rest of population just remove:
To compare 2 pictures we check every pixel in picture A and compare his R,G,B with
the same pixel (the same coordinates) in picture B.
I use this:
SingleDif = sqrt[ (Ar - Br)^2 + (Ag - Bg)^2 + (Ab - Bb)^2]
then i sum all differences (from all pixels) - lets call it SumDif
and use:
PictureDif = (DifMax - SumDif)/DifMax
where
DifMax = pictureHeight * pictureWidth * 255*3
The best are used to create the next population in this way:
picture MakeChild(picture Mother, picture Father)
{
picture child;
for( int i = 0; i < T; ++i )
{
j //this is a random number from 0 to 1 - created now
if( j < 0.5 ) child.element(i) = Mother.element(i);
else child.element(i) = Father.element(i)
if( j < some small % ) mutate( child.element(i) );
}
return child;
}
So it's quite simple. Only the mutation needs a comment: So there is always some small probability that element X in child will be different than X in his parent. To do this we make random changes in element in child (change his colour by random number, or add random number to his (x,y) coordinate - or his node).
So this is my idea... I didn't test it, didn't code it.
Please check my idea - what do you think about it?
I would make the number of patches of each child dynamic and get the mutation operation to insert/delete patches with some (low) probability. Of course this could result in a lot of redundancy and bloat in the child's genome. In these situations, it is usually a good idea to use the length of an individual's genome as a parameter of the fitness function so that individuals get rewarded (with a higher fitness value) for using fewer patches. So for example if the PictureDif of individuals A and B are the same but the A has fewer patches than B, then A has a higher fitness.
Another issue is the reproductive operator that you proposed (namely, the crossover operation). In order for the evolutionary process to work efficiently, you need to achieve a reasonable exploration and exploitation balance. One way of doing this is by having a set of reproductive operators that exhibit a good fitness correlation [1] which means the fitness of a child must be close to the fitness of its parent(s).
In the case of single parent reproduction you only need to find the right mutation parameters. However, when it comes to multi-parent reproduction (crossover) one of the frequently used techniques is to produce 2 children (instead of 1) from the same 2 parents. For the first child, each gene comes from the mother with the probability of 0.2 and from the father with the probability of 0.8, and for the second child the other way around. Of course after the crossover, you can do the mutation.
Oh and one more thing, for the mutation operators, when you say
... make random changes in element in child (change his colour by random number, or add random number to his (x,y) coordinate - or his node)
it's a good idea to use a Gaussian distribution to change the colour, coordinate etc.
[1] Evolutionary Computation: A unified approach by Kenneth A. De Jong, page 69
You have a biased random number generator that produces a 1 with a probability p and 0 with a probability (1-p). You do not know the value of p. Using this make an unbiased random number generator which produces 1 with a probability 0.5 and 0 with a probability 0.5.
Note: this problem is an exercise problem from Introduction to Algorithms by Cormen, Leiserson, Rivest, Stein.(clrs)
The events (p)(1-p) and (1-p)(p) are equiprobable. Taking them as 0 and 1 respectively and discarding the other two pairs of results you get an unbiased random generator.
In code this is done as easy as:
int UnbiasedRandom()
{
int x, y;
do
{
x = BiasedRandom();
y = BiasedRandom();
} while (x == y);
return x;
}
The procedure to produce an unbiased coin from a biased one was first attributed to Von Neumann (a guy who has done enormous work in math and many related fields). The procedure is super simple:
Toss the coin twice.
If the results match, start over, forgetting both results.
If the results differ, use the first result, forgetting the second.
The reason this algorithm works is because the probability of getting HT is p(1-p), which is the same as getting TH (1-p)p. Thus two events are equally likely.
I am also reading this book and it asks the expected running time. The probability that two tosses are not equal is z = 2*p*(1-p), therefore the expected running time is 1/z.
The previous example looks encouraging (after all, if you have a biased coin with a bias of p=0.99, you will need to throw your coin approximately 50 times, which is not that many). So you might think that this is an optimal algorithm. Sadly it is not.
Here is how it compares with the Shannon's theoretical bound (image is taken from this answer). It shows that the algorithm is good, but far from optimal.
You can come up with an improvement if you will consider that HHTT will be discarded by this algorithm, but in fact it has the same probability as TTHH. So you can also stop here and return H. The same is with HHHHTTTT and so on. Using these cases improves the expected running time, but are not making it theoretically optimal.
And in the end - python code:
import random
def biased(p):
# create a biased coin
return 1 if random.random() < p else 0
def unbiased_from_biased(p):
n1, n2 = biased(p), biased(p)
while n1 == n2:
n1, n2 = biased(p), biased(p)
return n1
p = random.random()
print p
tosses = [unbiased_from_biased(p) for i in xrange(1000)]
n_1 = sum(tosses)
n_2 = len(tosses) - n_1
print n_1, n_2
It is pretty self-explanatory, and here is an example result:
0.0973181652114
505 495
As you see, nonetheless we had a bias of 0.097, we got approximately the same number of 1 and 0
The trick attributed to von Neumann of getting two bits at a time, having 01 correspond to 0 and 10 to 1, and repeating for 00 or 11 has already come up. The expected value of bits you need to extract to get a single bit using this method is 1/p(1-p), which can get quite large if p is especially small or large, so it is worthwhile to ask whether the method can be improved, especially since it is evident that it throws away a lot of information (all 00 and 11 cases).
Googling for "von neumann trick biased" produced this paper that develops a better solution for the problem. The idea is that you still take bits two at a time, but if the first two attempts produce only 00s and 11s, you treat a pair of 0s as a single 0 and a pair of 1s as a single 1, and apply von Neumann's trick to these pairs. And if that doesn't work either, keep combining similarly at this level of pairs, and so on.
Further on, the paper develops this into generating multiple unbiased bits from the biased source, essentially using two different ways of generating bits from the bit-pairs, and giving a sketch that this is optimal in the sense that it produces exactly the number of bits that the original sequence had entropy in it.
You need to draw pairs of values from the RNG until you get a sequence of different values, i.e. zero followed by one or one followed by zero. You then take the first value (or last, doesn't matter) of that sequence. (i.e. Repeat as long as the pair drawn is either two zeros or two ones)
The math behind this is simple: a 0 then 1 sequence has the very same probability as a 1 then zero sequence. By always taking the first (or the last) element of this sequence as the output of your new RNG, we get an even chance to get a zero or a one.
Besides the von Neumann procedure given in other answers, there is a whole family of techniques, called randomness extraction (also known as debiasing, deskewing, or whitening), that serve to produce unbiased random bits from random numbers of unknown bias. They include Peres's (1992) iterated von Neumann procedure, as well as an "extractor tree" by Zhou and Bruck (2012). Both methods (and several others) are asymptotically optimal, that is, their efficiency (in terms of output bits per input) approaches the optimal limit as the number of inputs gets large (Pae 2018).
For example, the Peres extractor takes a list of bits (zeros and ones with the same bias) as input and is described as follows:
Create two empty lists named U and V. Then, while two or more bits remain in the input:
If the next two bits are 0/0, append 0 to U and 0 to V.
Otherwise, if those bits are 0/1, append 1 to U, then write a 0.
Otherwise, if those bits are 1/0, append 1 to U, then write a 1.
Otherwise, if those bits are 1/1, append 0 to U and 1 to V.
Run this algorithm recursively, reading from the bits placed in U.
Run this algorithm recursively, reading from the bits placed in V.
This is not to mention procedures that produce unbiased random bits from biased dice or other biased random numbers (not just biased bits); see, e.g., Camion (1974).
I discuss more on randomness extractors in a note on randomness extraction.
REFERENCES:
Peres, Y., "Iterating von Neumann's procedure for extracting random bits", Annals of Statistics 1992,20,1, p. 590-597.
Zhou, H. And Bruck, J., "Streaming algorithms for optimal generation of random bits", arXiv:1209.0730 [cs.IT], 2012.
S. Pae, "Binarization Trees and Random Number Generation", arXiv:1602.06058v2 [cs.DS].
Camion, Paul, "Unbiased die rolling with a biased die", North Carolina State University. Dept. Of Statistics, 1974.
Here's one way, probably not the most efficient. Chew through a bunch of random numbers until you get a sequence of the form [0..., 1, 0..., 1] (where 0... is one or more 0s). Count the number of 0s. If the first sequence is longer, generate a 0, if the second sequence is longer, generate a 1. (If they're the same, try again.)
This is like what HotBits does to generate random numbers from radioactive particle decay:
Since the time of any given decay is random, then the interval between two consecutive decays is also random. What we do, then, is measure a pair of these intervals, and emit a zero or one bit based on the relative length of the two intervals. If we measure the same interval for the two decays, we discard the measurement and try again
HotBits: How It Works
I'm just explaining the already proposed solutions with some running proof. This solution will be unbiased, no matter how many times we change the probability. In a head n tail toss, the exclusivity of consecutive head n tail or tail n head is always unbiased.
import random
def biased_toss(probability):
if random.random() > probability:
return 1
else:
return 0
def unbiased_toss(probability):
x = biased_toss(probability)
y = biased_toss(probability)
while x == y:
x = biased_toss(probability)
y = biased_toss(probability)
else:
return x
# results with contain counts of heads '0' and tails '1'
results = {'0':0, '1':0}
for i in range(1000):
# on every call we are changing the probability
p = random.random()
results[str(unbiased_toss(p))] += 1
# it still return unbiased result
print(results)