Use of the class `population`#

In pygmo, a population is a storage for candidate solutions to some problem (sometimes called individuals). It thus contains a problem and a number of decision vectors (chromosomes) and fitness vectors, each associated to a unique ID as to allow some degree of tracking.

>>> import pygmo as pg
>>> prob = pg.problem(pg.rosenbrock(dim = 4))
>>> pop1 = pg.population(prob)
>>> pop2 = pg.population(prob, size = 5, seed= 723782378)

In the code above, after the trivial import of the pygmo package, we first define a problem from rosenbrock, and then we construct two populations (i.e. two sets of candidate solutions). The first population, pop1, is empty, while the second population is initialized with five candidate solutions, hence also causing five fitness evaluations. The candidate solutions in pop2 are randomly created within the box-bounds of the problem using the random seed passed as kwarg.

>>> print(len(pop1))
0
>>> print(pop1.problem.get_fevals())
0
>>> print(len(pop2))
5
>>> print(pop2.problem.get_fevals())
5

The full inspection of a population is possible, as usual, via its methods or glancing to the screen print of the entire class:

>>> print(pop2) 
...

List of individuals:
#0:
    ID:                     10207561574636104601
    Decision vector:        [-0.777137, 7.91467, -4.31933, 5.92765]
    Fitness vector:         [470010]
#1:
    ID:                     15637512834538262525
    Decision vector:        [8.94985, 0.924838, 4.39905, -1.17683]
    Fitness vector:         [670339]
#2:
    ID:                     15983142956381075935
    Decision vector:        [-0.291054, 4.99031, 1.34008, -0.00609471]
    Fitness vector:         [58271.1]
#3:
    ID:                     9198455452901607935
    Decision vector:        [2.18419, -1.04727, 6.35101, 6.39632]
    Fitness vector:         [121365]
#4:
    ID:                     4553447107323210017
    Decision vector:        [7.50729, -1.14561, 5.98053, -3.48833]
    Fitness vector:         [487030]

Individuals, i.e. new candidate solutions can be put into a population calling its push_back() method:

>>> pop1.push_back(x = [0.1,0.2,0.3,0.4]) # correct size
>>> len(pop1) == 1
True
>>> pop1.problem.get_fevals() == 1
True
>>> pop1.push_back(x = [0.1,0.2,0.3]) # wrong size
Traceback (most recent call last):
  File ".../lib/python3.6/doctest.py", line 1330, in __run
    compileflags, 1), test.globs)
  File "<doctest default[3]>", line 1, in <module>
    pop1.push_back([0.1,0.2,0.3])
ValueError:
function: check_decision_vector
where: /Users/darioizzo/Documents/pagmo2/include/pagmo/problem.hpp, 1835
what: Length of decision vector is 3, should be 4

Some consistency checks are done by push_back(), e.g. on the decision vector length.

Note

Decision vectors that are outside of the box bounds are allowed to be pushed back into a population

The snippet above will trigger fitness function evaluations as the decision vector is always associated to a fitness vector in a population. If the fitness vector associated to a chromosome is known, you may still push it back in a population and avoid triggering a fitness re-evaluation by typing:

>>> pop1.push_back(x = [0.2,0.3,1.3,0.2], f = [11.2])
>>> len(pop1) == 2
True
>>> pop1.problem.get_fevals() == 1
True

When designing user-defined algorithms (UDAs) it is often important to be able to change some individual decision vector:

>>> pop1.problem.get_fevals() == 1
True
>>> print(pop1.get_x()[0]) 
[0.1  0.2  0.3  0.4]
>>> pop1.set_x(0, [1.,2.,3.,4.])
>>> pop1.problem.get_fevals() == 2
True
>>> print(pop1.get_f()[0])
[2705.]
>>> pop1.set_xf(0, [1.,2.,3.,4.], [8.43469444])
>>> pop1.problem.get_fevals() == 2
True
>>> print(pop1.get_f()[0])
[8.43469444]

Note

Using the method set_xf() or:func:~pygmo.population.push_back() it is possible to avoid triggering fitness function evaluations, but it is also possible to inject spurious information into the population (i.e. breaking the relation between decision vectors and fitness vectors imposed by the problem)

The best individual in a population can be extracted as:

>>> # The decision vector
>>> pop1.get_x()[pop1.best_idx()] 
array([1.,  2.,  3.,  4.])
>>> # The fitness vector
>>> pop1.get_f()[pop1.best_idx()]
array([8.43469444])

The best individual that ever lived in a population, i.e. the champion can also be extracted as:

>>> # The decision vector
>>> pop1.champion_x 
array([1.,  2.,  3.,  4.])
>>> # The fitness vector
>>> pop1.champion_f
array([8.43469444])

Note

The champion is not necessarily identical to the best individual in the current population as it actually keeps the memory of all past individuals that were at some point in the population

Use of the class population

Use of the class population#

Use of the class `population`#