Ruby 2.0
Com o conjunto de operadores [U, D, F, L], o menor número de seqüências selecionado para classificar a lista é o número de itens na lista menos a subsequência comum mais longa. Para adicionar o operador R, basta inverter a string e aplicar a mesma regra. Infelizmente, minimizar a seleção de cadeias não é o mesmo que minimizar o número de operações. Por exemplo, para uma entrada de 8 dd ww aa bb cc xx yy zz
, a resposta correta é ww D D D dd D D D
, mas o menor número de operações (que atende aos outros critérios da pergunta) seria cc F bb F aa F
. Isso significa que uma porção muito maior do conjunto de possíveis caminhos de classificação precisa ser explorada.
Esta solução usa uma estratégia de pesquisa profunda e poda alfa-beta. É importante diminuir o valor alfa rapidamente para minimizar a profundidade da pesquisa, caso contrário, a árvore de pesquisa explode exponencialmente. Por exemplo, para determinar o caminho de classificação com a pontuação mínima para o exemplo introdutório do OP, classificando os meses em ordem de calendário para ordem lexical, provavelmente levará algumas décadas com o método de pontuação atual deste programa. O programa encontra o número mínimo de seleções de string, 8, muito rapidamente. Infelizmente, isso ainda deixa uma árvore enorme para atravessar.
Estou usando a classificação gnome como minha função de pontuação porque:
- é simples de entender e modificar
- a pontuação geralmente converge para o alfa ideal rapidamente
- essa implementação é mais rápida que minha implementação da função LCS
- jogará melhor que a função LCS
O número 4 seria suficiente. Tudo além é um bônus.
Para uma pesquisa profunda, a ordem na qual as operações são exploradas tem um impacto significativo no tempo de pesquisa. Como qualquer conjunto não vazio de N itens pode ser classificado com operações ≤ N-1 F (primeiro) ou L (ast), essas operações são tentadas primeiro.
# gnome sort
def gnomeSort(a)
selects = 0
previous = nil
i = 1
while i < a.size
if a[i-1] <= a[i]
# the array a[0..i] is sorted
i += 1 # take another bite
else
if a[i] != previous
previous = a[i]
selects += 1
end
a[i], a[i-1] = a[i-1], a[i]
if (i > 1)
i -= 1
end
end
end
return selects
end
def score(a)
return gnomeSort(a.dup)
end
# squeeze out unnecessary operands
def consolidate(a)
# separate operands and operators
x = [] # operands
f = [] # operators
a.each_slice(2) { |a,b|
x << a
f << b
}
n = x.size # number of (operand operator) pairs
if n>=2
# replace all R operands with the lexically lower operand
# from the right or left
f.each_with_index{|v,i|
if v=='R'
leftOperand = x[i-1]
rightOperand = x[i+1]
# handle left & right edge cases
leftOperand = rightOperand.succ if i==0
rightOperand = leftOperand.succ if i>=n-1
x[i] = [leftOperand, rightOperand].min
end
}
# replace repeated operands with <nil>
x = x.chunk{|e|e}.map{|v|v[1].fill(nil,1)}.flatten
end
return [x, f]
end
@solutions = []
@operation = []
@operation[3] = ->(a, i) {
# swap a[i] and a[i-1]
return nil if i<1 || i>=a.size
v = a[i]
a[i-1], a[i] = a[i], a[i-1]
return [ v, 'U' ]
}
@operation[0] = ->(a, i) {
# move a[i] after a.last
return nil if i+1>=a.size
a.push(v=a.delete_at(i))
return [ v, 'L' ]
}
@operation[4] = ->(a, i) {
# reverse the whole array
v = a[i]
a.reverse!
return [ v, 'R' ]
}
@operation[1] = ->(a, i) {
# move a[i] before a.first
return nil if i<=0
a.unshift(v=a.delete_at(i))
return [ v, 'F' ]
}
@operation[2] = ->(a, i) {
# swap a[i] and a[i+1]
return nil if i<0 || i+1>=a.size
v = a[i]
a[i], a[i+1] = a[i+1], a[i]
return [ v, 'D' ]
}
def alphaSort(depth, a, selected, selects, sortPath)
depth += 1
return if selects > @alpha
return if selects>@alpha || selects+depth>a.size+1
if a.each_cons(2).all?{ |x, y| x <= y }
# found a sort path
@alpha = selects
@solutions << sortPath.flatten.compact
else
selectsFromHere = score(a)
if @alpha > selects+selectsFromHere
@alpha = selects+selectsFromHere
else
end
@operation.each do |op|
a.each_index do |i|
b = a.dup
branch = sortPath.dup << op[b,i]
alphaSort(depth, b, a[i], selects+(selected==a[i] ? 0 : 1), branch)
end
end
end
end
# input
a = ARGF.read.scan(/\w+/m) # alternative, $*[0].scan(/\w+/m)
a.shift # ignore the item count
# depth-first search of sort operations
@alpha = [a.size-1, score(a), score(a.reverse)+1].min + 1
alphaSort(0, a, nil, 0, [])
# winnow the set of solutions
# determine the minimum number of string selects to solve
# short-circuit if selects to solve is 0 (already sorted)
@solutions.map!{|v|consolidate v}
minSelects = @solutions.map{|v|v[0].compact.size}.min
if !minSelects
puts
exit
end
# keep only solutions with the minimum number of string selects
@solutions.reject!{ |v| v[0].compact.size > minSelects }
# determine the minimum number of moves in the remaining solutions
minMoves = @solutions.map{|v|v[1].size}.min
# keep only solutions with the minimum number of moves
@solutions.reject!{ |v| v[1].size > minMoves }
# beauty contest
# turn into strings
solutions = @solutions.map{|v|v[0].zip(v[1]).flatten.compact*' '}
# keep the shortest strings
minLength = solutions.map{|v|v.size}.min
solutions.reject!{ |v| v.size > minLength }
# print the string that "that comes first alphabetically"
puts solutions.sort.first
Ele passa neste conjunto de testes perl TAP :
use strict;
use warnings;
use Test::More qw(no_plan);
#use Test::More tests => 61;
# solution executable
my $solver = 'ruby2.0 sortshort.rb';
my $nonTrivial = 1;
# "happy" path
# examples from OP
is( `echo 2 zz abc | $solver 2>&1`, "zz D\n", 'OP example #1');
is( `echo 3 cc bb aa | $solver 2>&1`, "aa R\n", 'OP example #2');
is( `echo 4 abc def cd ccc | $solver 2>&1`, "abc L R\n", 'OP example #3');
is( `echo 6 rr mm nn oo qq pp | $solver 2>&1`, "pp U rr L\n", 'OP example #4');
# example from bizangles
is( `echo 6 xx aa dd bb ee cc | $solver 2>&1`, "dd L ee L xx L\n", 'wascally wabbit, challenges deep diver (from bizangles)');
SKIP: {
skip('non-trivial tests', 2) unless $nonTrivial;
# 7 item example; bizangles' python solution (circa 2014-08-22) requires higher sys.setrecursionlimit() and takes about 5 minutes
is( `echo 7 aa bb ee cc dd ff gg | $solver 2>&1`, "ee D D\n", 'shallow');
# minimizing the number of selects scores better than minimizing moves
# minimizing moves => cc F bb F aa F
# minimizing selects => dd D D D D ww D D D D, ww D D D dd D D D, ww L U U U dd D D D, etc.
# minimizing selects, then moves => ww D D D dd D D D
is( `echo 8 dd ww aa bb cc xx yy zz | $solver 2>&1`, "ww D D D dd D D D\n", 'joker, minimize selects before moves');
}
# exhaustive variations on a theme with 1 item ["aa"]
is( `echo 1 aa | $solver 2>&1`, "\n", 'permutations of 1, #1');
# exhaustive variations on a theme with 2 items ["ab", "c"]
is( `echo 2 ab c | $solver 2>&1`, "\n", 'permutations of 2, #1');
# test OP's requirement that a string be selected before reverse operation
is( `echo 2 c ab | $solver 2>&1`, "c D\n", 'permutations of 2, #2');
# exhaustive variations on a theme with 3 items ["five", "four", "three"]
is( `echo 3 five four three | $solver 2>&1`, "\n", 'permutations of 3, #1');
is( `echo 3 five three four | $solver 2>&1`, "four U\n", 'permutations of 3, #2');
is( `echo 3 four five three | $solver 2>&1`, "five F\n", 'permutations of 3, #3');
is( `echo 3 four three five | $solver 2>&1`, "five F\n", 'permutations of 3, #4');
is( `echo 3 three five four | $solver 2>&1`, "three L\n", 'permutations of 3, #5');
is( `echo 3 three four five | $solver 2>&1`, "five R\n", 'permutations of 3, #6');
# selected variations on a theme with 5 items ["aa", "bb", "cc", "dd", "ee"]
is( `echo 5 aa bb cc dd ee | $solver 2>&1`, "\n", 'permutations of 5, #1, already sorted');
# two sort paths of length 1
is( `echo 5 aa bb cc ee dd | $solver 2>&1`, "dd U\n", 'permutations of 5, #2, single U or D');
is( `echo 5 aa bb ee cc dd | $solver 2>&1`, "ee L\n", 'permutations of 5, #4, single L');
is( `echo 5 bb cc aa dd ee | $solver 2>&1`, "aa F\n", 'permutations of 5, #31, single F');
is( `echo 5 ee dd cc bb aa | $solver 2>&1`, "aa R\n", 'permutations of 5, #120, reverse sorted');
# exhaustive variations on a theme with 4 items ["aa", "bb", "cc", "dd"]
# sort paths of length 0
is( `echo 4 aa bb cc dd | $solver 2>&1`, "\n", 'permutations of 4, #1');
# sort paths of length 1
is( `echo 4 aa bb dd cc | $solver 2>&1`, "cc U\n", 'permutations of 4, #2');
is( `echo 4 aa cc bb dd | $solver 2>&1`, "bb U\n", 'permutations of 4, #3');
is( `echo 4 aa dd bb cc | $solver 2>&1`, "dd L\n", 'permutations of 4, #5');
is( `echo 4 bb aa cc dd | $solver 2>&1`, "aa F\n", 'permutations of 4, #7');
is( `echo 4 bb cc aa dd | $solver 2>&1`, "aa F\n", 'permutations of 4, #9');
is( `echo 4 bb cc dd aa | $solver 2>&1`, "aa F\n", 'permutations of 4, #10');
is( `echo 4 dd aa bb cc | $solver 2>&1`, "dd L\n", 'permutations of 4, #19');
is( `echo 4 dd cc bb aa | $solver 2>&1`, "aa R\n", 'permutations of 4, #24');
# sort paths of length 2
is( `echo 4 aa cc dd bb | $solver 2>&1`, "bb F D\n", 'permutations of 4, #4');
is( `echo 4 aa dd cc bb | $solver 2>&1`, "aa L R\n", 'permutations of 4, #6');
is( `echo 4 bb aa dd cc | $solver 2>&1`, "aa F cc U\n", 'permutations of 4, #8');
is( `echo 4 bb dd aa cc | $solver 2>&1`, "aa F cc U\n", 'permutations of 4, #11');
is( `echo 4 bb dd cc aa | $solver 2>&1`, "bb D D R\n", 'permutations of 4, #12');
is( `echo 4 cc aa bb dd | $solver 2>&1`, "cc D D\n", 'permutations of 4, #13');
is( `echo 4 cc aa dd bb | $solver 2>&1`, "bb F aa F\n", 'permutations of 4, #14');
is( `echo 4 cc bb aa dd | $solver 2>&1`, "dd F R\n", 'permutations of 4, #15');
is( `echo 4 cc bb dd aa | $solver 2>&1`, "dd F R\n", 'permutations of 4, #16');
is( `echo 4 cc dd aa bb | $solver 2>&1`, "bb F aa F\n", 'permutations of 4, #17');
is( `echo 4 cc dd bb aa | $solver 2>&1`, "cc D R\n", 'permutations of 4, #18');
is( `echo 4 dd aa cc bb | $solver 2>&1`, "aa L R\n", 'permutations of 4, #20');
is( `echo 4 dd bb aa cc | $solver 2>&1`, "cc F D R\n", 'permutations of 4, #21');
is( `echo 4 dd bb cc aa | $solver 2>&1`, "bb D R\n", 'permutations of 4, #22');
is( `echo 4 dd cc aa bb | $solver 2>&1`, "aa D R\n", 'permutations of 4, #23');
# variations on a theme with 4 items ["aaaaa", "bbbb", "ccc", "dd"]
# force choice by string length
is( `echo 4 ccc dd aaaaa bbbb | $solver 2>&1`, "ccc L dd L\n", 'permutations of 4, #17');
is( `echo 4 dd bbbb aaaaa ccc | $solver 2>&1`, "ccc F D R\n", 'permutations of 4, #21');
is( `echo 4 bbbb aaaaa dd ccc | $solver 2>&1`, "bbbb D dd D\n", 'permutations of 4, #8');
is( `echo 4 bbbb dd aaaaa ccc | $solver 2>&1`, "dd L bbbb D\n", 'permutations of 4, #11');
is( `echo 4 ccc aaaaa dd bbbb | $solver 2>&1`, "ccc L dd L\n", 'permutations of 4, #14');
is( `echo 4 ccc dd bbbb aaaaa | $solver 2>&1`, "dd F R\n", 'permutations of 4, #18');
is( `echo 4 dd aaaaa ccc bbbb | $solver 2>&1`, "aaaaa L R\n", 'permutations of 4, #20');
is( `echo 4 dd bbbb ccc aaaaa | $solver 2>&1`, "ccc R D\n", 'permutations of 4, #22');
is( `echo 4 dd ccc aaaaa bbbb | $solver 2>&1`, "bbbb R D\n", 'permutations of 4, #23');
# identical items in list
is( `echo 2 aa aa | $solver 2>&1`, "\n", '1 repeat #1');
is( `echo 3 aa aa bb | $solver 2>&1`, "\n", '1 repeat #2');
is( `echo 3 aa bb aa | $solver 2>&1`, "aa F\n", '1 repeat #3');
is( `echo 3 bb aa aa | $solver 2>&1`, "aa R\n", '1 repeat #4');
is( `echo 4 aa cc bb aa| $solver 2>&1`, "aa L R\n", '1 repeat #5');
is( `echo 5 cc bb aa bb cc | $solver 2>&1`, "aa F cc L\n", '2 repeats');
# "sad" path
# not explicitly excluded, so cover this case
# exhaustive variations on a theme with 0 items []
is( `echo 0 | $solver 2>&1`, "\n", 'permutations of 0, #1');
# "bad" path
# none!
exit 0;
A
que não existe.