I am encountering a issue that I need to add up the second number of each list. For example, suppose I have a list of lists like below,
(list (list -4
(list (list -1 4) (list 1 7)))
(list 1 (list (list -2 5) (list 3 3)))
(list 3 12))
Then my job is to add up 4 + 7 + 5 + 3 + 12 = 31. However, the list can have multiple sub lists. But the second item inside a list can either be a number or a list. If it is a list, then we need to dig deeper into this list until we get a number.
Thanks!
Solution
(define (atom? x)
(and (not (null? x))
(not (pair? x))))
(define (my-and x y)
(and x y))
(define (every? l)
(foldr my-and #t l))
(define (flat-list? l)
(cond ((null? l) #t)
((every? (map atom? l)) #t)
(else #f)))
(define (add-only-seconds l)
(define (l-sec-add l acc)
(cond ((null? l) acc)
((atom? l) acc)
((flat-list? l) (+ (second l) acc))
((list? l) (apply + acc (map (lambda (x) (l-sec-add x 0)) l)))))
(l-sec-add l 0))
Example test
(define example-list (list (list -4
(list (list -1 4) (list 1 7)))
(list 1 (list (list -2 5) (list 3 3)))
(list 3 12)))
(add-only-seconds example-list) ;; 31
I used common-lisp-typical functions atom? and every?.
Since and cannot be used in foldr, I defined my-add to make add a function which can be used infoldr`.
Let's consider this list:
(define parts '(("a" "b" "c" "d" "e" "1")
("x" "y" "z" "a")
("q" "w" "e" "x")
("1" "2" "3" "4" "q")))
I need to make an hash where every first element is a key and its value is a list with references to this key appeared in another lists. This an example of my desired result:
(define desired-result '#hash(("a" . ("x"))
("x" . ("q"))
("q" . ("1"))
("1" . ("a"))))
As you can see "a" (first in first list) is mentioned by "x" ("a" in present in second list beginning with "x"). "x" is mentioned by "q", etc.
I have come up with this code to get a more complete view of the "references" thing, but it not what I need and it is also ugly (and possibly slow) see the complete code:
#lang racket/base
(require racket/list)
(define parts '(("a" "b" "c" "d" "e" "1")
("x" "y" "z" "a")
("q" "w" "e" "x")
("1" "2" "3" "4" "q")))
(define my-hash (make-hash '()))
; This will initialize a key for every single element in parts
(for-each (λ (x)
(hash-set! my-hash x '()))
(remove-duplicates (flatten parts)))
(define (put x y)
(hash-set! my-hash x y))
(define (get x)
(hash-ref my-hash x))
(define (insert a n)
(let ([aList (get a)]
[nList (get n)])
(unless (member n aList)
(put a (append aList (list n))))
(unless (member a nList)
(put n (append nList (list a))))))
(define (iterate l)
(let ([a (car l)]
[r (cdr l)])
(for-each (λ (n) (insert a n)) r)))
(for-each iterate parts)
my-hash
This will result in:
'#hash(("c" . ("a"))
("e" . ("a" "q"))
("2" . ("1"))
("a" . ("b" "c" "d" "e" "1" "x"))
("w" . ("q"))
("4" . ("1"))
("y" . ("x"))
("d" . ("a"))
("3" . ("1"))
("1" . ("a" "2" "3" "4" "q"))
("b" . ("a"))
("q" . ("w" "e" "x" "1"))
("x" . ("y" "z" "a" "q"))
("z" . ("x")))
There are surely better ways to obtain this (and I am curious if someone can suggest some) and I know I can get to desired-result from this, but this will be even uglier.
PS:
This is not a school assignment
I know hashes are not ordered
My solution additionally uses hash set for performance to test if an element is a member of the leading elements or not.
(define (process parts)
(define leading-list (map first parts))
(define leading-set (list->set leading-list))
(define in-leading-set? (curry set-member? leading-set))
(define my-hash (make-hash (map (curryr cons empty) leading-list)))
(for-each
(λ (lst)
(for-each
(λ (e)
(hash-set! my-hash e (cons (first lst) (hash-ref my-hash e))))
(filter in-leading-set? (rest lst))))
parts)
my-hash)
Here's the output
> (process parts)
'#hash(("1" . ("a")) ("x" . ("q")) ("q" . ("1")) ("a" . ("x")))
This has a caveat that, there might be some elements mapping to an empty list. For example:
> (define parts2 '(("a" "b")))
> (process parts2)
'#hash(("a" . ()))
If you don't prefer them, you could post-process by filtering them out.
This might be what you want:
(define dwim
(lambda (parts)
(let loop ((todo parts)
(done '())
(result '()))
(if (null? todo)
result
(let* ((key (caar todo))
(value
(fold
(lambda (lst previous)
(if (member key lst)
(cons (car lst) previous)
previous))
'()
(append (cdr todo) done))))
(loop (cdr todo)
(cons (car todo) done)
(cons (cons key value) result)))))))
This seems straightforward, but I can't seem to find a solution. I want to replace an item within a list of a list with something, but if that item appears multiple times then you randomly replace one of them, but not both. I want to do this in ISL+.
I created the function flatten which appends all sublists :
(check-expect (flatten '((a b) (c) (d e f g) (h i j)))
(list 'a 'b 'c 'd 'e 'f 'g 'h 'i 'j))
(define (flatten lol)
(foldr append empty lol))
I also made rewrite, which replaces the value at index n with whatever you choose
(check-expect (rewrite '(x x x - x x x x) 3 'x)
(list 'x 'x 'x 'x 'x 'x 'x 'x))
(define (rewrite ls n val)
(cond
[(empty? ls) (error "error")]
[(= n 0) (cons val (rest ls))]
[else (cons (first ls) (rewrite (rest ls) (sub1 n) val))]))
The problem is I don't know how to apply this to a list of list and I also don't know how to randomly replace one of items if it occurs more than once. This is what I have for the final product, but it's probably not the way to go:
(define (fullreplace b)
(local [
;makes a list of nested lists of each index the element occurs
;problem is that it makes a list of nested lists so I can't use flatten either
(define (position ls ele n)
(cond [(empty? ls) 0]
[(equal? ele (first ls)) (list n (position (rest ls) ele (add1 n))) ]
[else (position (rest ls) ele (+ 1 n))]))]
;lol-full? checks if the item occurs in the list of lists at all
(if (lol-full? b) b (rewrite (flatten b)
(position (flatten b) '- 0)
"item replaced"))))
;just used for testing
(define lol2 (list
(list 2 2 2 2)
(list 4 '- 4 '-)
(list '- 8 8 8)
(list 16 '- '- 16)))
(fullreplace lol2) may return this or where any of the other '- are located:
(list
(list 2 2 2 2)
(list 4 '- 4 2)
(list '- 8 8 8)
(list 16 '- '- 16))
I've been working on this awhile so any new insight would go a long way. Thank you
The "random" part is what makes this problem pathological. If you could just replace the first occurrence, it would be easy. But to replace a random occurence, you must first know how many occurrences there are. So before you go replacing stuff, you have to go a-counting:
(define (count/recursive val tree)
(cond ((equal? val tree)
1)
(else (foldl (λ (next-value total)
(cond ((equal? val next-value)
(add1 total))
((list? next-value)
(+ total (count/recursive val next-value)))
(else total))) 0 tree))))
Then you need a function that can replace the nth occurrence of a value:
(define (replace/recursive val replace-with n tree)
(cond ((equal? val tree)
replace-with)
(else
(cdr
(foldl (λ (next-value total/output-tree)
(local ((define total (car total/output-tree))
(define output-tree (cdr total/output-tree)))
(cond ((equal? next-value val)
(cons (add1 total)
(cons (if (= total n) replace-with next-value) output-tree)))
((list? next-value)
(cons (+ total (count/recursive val next-value))
(cons (replace/recursive val replace-with (- n total) next-value)
output-tree)))
(else (cons total (cons next-value output-tree)))))) (cons 0 empty) tree)))))
Finally, you use random to pick the instance you will replace, using count/recursive to limit how high of a number random picks:
(define original '((x x (x y x) a b (((c x z x) x) y x x))))
(replace/recursive 'x '- (random (count/recursive 'x original)) original)
How to replace all occurences of a value with another value:
(define (replace-all needle new-value haystack)
(cond ((equal? needle haystack) new-value)
((pair? haystack)
(cons (replace-all needle new-value (car haystack))
(replace-all needle new-value (cdr haystack))))
(else haystack)))
The only thing to change is to check if the first part constituted a change. If it did you don't do the replace on the other half. Use equal? to compare structure.
It's not random. It will replace the first occurence it finds either by doing car before cdr or cdr before car.
I get NZEC with the following code for INVCNT
; for lists of length > 2 inversions are the same as the number of elements
; against which the first is greater + the inversions of the remaining
(define (inversions l)
(cond ((< (length l) 2) 0)
(else (+ (length (filter (lambda (x) (> (car l) x)) (cdr l)))
(inversions (cdr l))))))
(use-modules (ice-9 rdelim))
(define (call-n-times proc n)
(if (= 0 n)
'()
(cons (proc) (call-n-times proc (- n 1)))))
(define (solve)
(write-line (inversions (call-n-times read (read)))))
(call-n-times solve (read))
Any hints, please?
Filtering accross a very long list can run you into the maximum recusion error (specs say up to ten million) Instead of using '(length (filter ...' use a fold
(define (inversion L)
(let loop ((accumulator 0) (L L))
(if (null? L)
accumulator
(loop
(+ accumulator
(fold
(lambda (init next)
(if (> (car l) next)
(+ init 1)
init))
0
(cdr L)))
(cdr L)))))
Second though this would be easier to read pulling out that fold into it's own function
(define (inversions-from-car L)
(fold
(lambda (init next)
(if (> (car l) next)
(+ init 1)
init))
0
(cdr L)))
(define (inversion L)
(let loop ((accumulator 0) (L L))
(if (null? L)
accumulator
(loop
(+ accumulator
(inversions-from-car L)
(cdr L)))))
This looks like a good problem to play with data structures, because as written, it's n^2 complexity.
I think you can get it down to n(log n)
Say create a sorted tree on the list of value paired with the # of nodes to the left.
for this set
'(2 3 8 6 1) -> '(1 2 3 6 8) ->
(*tree (*entry 3 2 2)
(*tree (*entry 2 1 1)
(*tree (*entry 1 0 1)
()
())
())
(*tree (*entry 8 1 1)
(*tree (*entry 6 0 1)
()
())
()))
*tree and *entry are just type-tage
*tree should have an entry, a left and a right
*entry should have a value, #left, and number)
Start by finding the the FIRST in the orginal list with a zero accumulator
'(2 3 8 6 1)
If the value of the enrty matched to FIRST, add #left to the accumulator
If the value is entry is more than FIRST recurse on the left branch of the tree with accumulator
If the value of the entry is less then FIRST , recurse on the right branch with #left added to the accumulator
If it's a null-tree throw an error
Then you need to update the tree.
If the value of the entry equal to FIRST, mutate the entry to reduce the number by one
If the value is entry is more then FIRST, mutate the entry toreduce #left by one and recurse on the left branch
If the value of the entry is less than first , recurse on the right branch
If it's a null-tree throw an error
You can combine these rules into a single traversal
Additionally add the rule that if #left is 0 and number is zero, then if the right branch is null mutate this tree to the empty-tree else the right-branch.
Here's a rough (untested version of the idea)
(define (rev-sorted-list->count-list L) ;;sort should be resverse of
;; final desired order
(let loop ((value (car L)) (count 1) (L (cdr L)) (acc '()))
(cond ((null? L) '())
((= value (car l))
(loop value (+ 1 count) (cdr L) acc))
(else
(loop (car l) 1 (cdr L) (cons (cons value count) acc))))))
(define (make-tree count c-L)
(let* ((middle (ceiling (+ 1 count) 2))
(left-count (- middle 1))
(right-count (-count middle))
(left (if (= 0 left-count)
null-tree
(make-tree left-count c-L)))
(entry+right
(let loop ((index 1) (L c-L))
(if (= index middle)
L
(loop (+ 1 index) (cdr L)))))
(entry
(make-entry
(caar entry+right)
left-count
(cdar entry+right))))
(build-tree
entry
left
(if (= 0 right-count)
null-tree
(make-tree right-count (cdr entry+right))))))
;;form left branches from starting points
;;;form right from stopping points
;;never mutating c-L or copies
;;if count = 0 then null tree
(define (build-tree entry left right)
(list '*tree entry left right)
(define (entry tree)
(cadr tree)
(define (left-branch tree)
(caddr tree))
(define (right-branch tree)
(cadddr tree))
(define null-tree (list '*tree '()))
(define (null-tree? tree)
(null? (entry tree)))
(define (make-entry value Nleft count)
(let ((vec (make-vector 3)))
(begin (vector-set! vec 0 value)
(vector-set! vec 1 Nleft)
(vector-set! vec 2 count)
vec)))
;;might meessage passing function here
(define (entry-value entry)
(vector-ref entry 0))
(define (entry-Nleft entry)
(vector-ref entry 1))
(define (entry-Nleft-set! entry int)
(vector-set! entry 1 int))
(define (entry-count entry)
(vector-ref entry 2))
(define (entry-count-set! entry int)
(vector-set! entry 2 int))
(define (inversions! Test-List Search-Tree)
(let loop ((acc 0) (L Test-list) (T Search-tree))
(cond ((null? L) acc)
((null-tree? T) (error "null tree "
"in inner loop of inversion!"))
((= (car L) (entry-value (entry T)))
(entry-count-set! (entry T)
(- (entry-count (entry T)) 1))
(if (and (= 0 (entry-count (entry T)))
(= 0 (entry-Nleft (entry T))))
(set-cdr! T (right-branch T))
'skip)
(loop (+ acc (entry-Nleft (entry T)))
(cdr L)
Search-tree))
((< (car L) (entry-value (entry T)))
(entry-Nleft-set! (entry T)
(- (entry-Nleft (entry T)) 1))
(loop acc L (left-branch T)))
((> (car L) (entry-value (entry T)))
(loop (+ acc (entry-Nleft (entry T)))
L
(right-branch T))))))
I hacked together several code snippets from various sources and created a crude implementation of a Wolfram Blog article at http://bit.ly/HWdUqK - for those that are mathematically inclined, it is very interesting!
Not surprisingly, given that I'm still a novice at Racket, the code takes too much time to calculate the results (>90 min versus 49 seconds for the author) and eats up a lot of memory. I suspect it is all about the definition (expListY) which needs to be reworked.
Although I have it working in DrRacket, I am also having problems byte-compiling the source, and still working on it
(Error message: +: expects type <number> as 1st argument, given: #f; other arguments were: 1 -1)
Anybody want to take a stab at improving the performance and efficiency? I apologize for the unintelligible code and lack of better code comments.
PS: Should I be cutting and pasting the code directly here?
Probably similar to soegaard's solution, except this one rolls its own "parser", so it's self contained. It produces the complete 100-year listing in a bit under 6 seconds on my machine. There's a bunch of tricks that this code uses, but it's not really something that would be called "optimized" in any serious way: I'm sure that it can be made much faster with some memoization, care for maximizing tree sharing etc etc. But for such a small domain it's not worth the effort... (Same goes for the quality of this code...)
BTW#1, more than parsing, the original solution(s) use eval which does not make things faster... For things like this it's usually better to write the "evaluator" manually. BTW#2, this doesn't mean that Racket is faster than Mathematica -- I'm sure that the solution in that post makes it grind redundant cpu cycles too, and a similar solution would be faster.
#lang racket
(define (tuples list n)
(let loop ([n n])
(if (zero? n)
'(())
(for*/list ([y (in-list (loop (sub1 n)))] [x (in-list list)])
(cons x y)))))
(define precedence
(let ([t (make-hasheq)])
(for ([ops '((#f) (+ -) (* /) (||))] [n (in-naturals)])
(for ([op ops]) (hash-set! t op n)))
t))
(define (do op x y)
(case op
[(+) (+ x y)] [(-) (- x y)] [(*) (* x y)] [(/) (/ x y)]
[(||) (+ (* 10 x) y)]))
(define (run ops nums)
(unless (= (add1 (length ops)) (length nums)) (error "poof"))
(let loop ([nums (cddr nums)]
[ops (cdr ops)]
[numstack (list (cadr nums) (car nums))]
[opstack (list (car ops))])
(if (and (null? ops) (null? opstack))
(car numstack)
(let ([op (and (pair? ops) (car ops))]
[topop (and (pair? opstack) (car opstack))])
(if (> (hash-ref precedence op)
(hash-ref precedence topop))
(loop (cdr nums)
(cdr ops)
(cons (car nums) numstack)
(cons op opstack))
(loop nums
ops
(cons (do topop (cadr numstack) (car numstack))
(cddr numstack))
(cdr opstack)))))))
(define (expr ops* nums*)
(define ops (map symbol->string ops*))
(define nums (map number->string nums*))
(string-append* (cons (car nums) (append-map list ops (cdr nums)))))
(define nums (for/list ([i (in-range 10 0 -1)]) i))
(define year1 2012)
(define nyears 100)
(define year2 (+ year1 nyears))
(define years (make-vector nyears '()))
(for ([ops (in-list (tuples '(+ - * / ||) 9))])
(define r (run ops nums))
(when (and (integer? r) (<= year1 r) (< r year2))
(vector-set! years (- r year1)
(cons ops (vector-ref years (- r year1))))))
(for ([solutions (in-vector years)] [year (in-range year1 year2)])
(if (pair? solutions)
(printf "~a = ~a~a\n"
year (expr (car solutions) nums)
(if (null? (cdr solutions))
""
(format " (~a more)" (length (cdr solutions)))))
(printf "~a: no combination!\n" year)))
Below is my implementation. I tweaked and optimized a thing or two in your code, in my laptop it takes around 35 minutes to finish (certainly an improvement!) I found that the evaluation of expressions is the real performance killer - if it weren't for the calls to the procedure to-expression, the program would finish in under a minute.
I guess that in programming languages that natively use infix notation the evaluation would be much faster, but in Scheme the cost for parsing and then evaluating a string with an infix expression is just too much.
Maybe someone can point out a suitable replacement for the soegaard/infix package? or alternatively, a way to directly evaluate an infix expression list that takes into account operator precedence, say '(1 + 3 - 4 & 7) - where & stands for number concatenation and has the highest precedence (for example: 4 & 7 = 47), and the other arithmetic operators (+, -, *, /) follow the usual precedence rules.
#lang at-exp racket
(require (planet soegaard/infix)
(planet soegaard/infix/parser))
(define (product lst1 lst2)
(for*/list ([x (in-list lst1)]
[y (in-list lst2)])
(cons x y)))
(define (tuples lst n)
(if (zero? n)
'(())
(product lst (tuples lst (sub1 n)))))
(define (riffle numbers ops)
(if (null? ops)
(list (car numbers))
(cons (car numbers)
(cons (car ops)
(riffle (cdr numbers)
(cdr ops))))))
(define (expression-string numbers optuple)
(apply string-append
(riffle numbers optuple)))
(define (to-expression exp-str)
(eval
(parse-expression
#'here (open-input-string exp-str))))
(define (make-all-combinations numbers ops)
(let loop ((opts (tuples ops (sub1 (length numbers))))
(acc '()))
(if (null? opts)
acc
(let ((exp-str (expression-string numbers (car opts))))
(loop (cdr opts)
(cons (cons exp-str (to-expression exp-str)) acc))))))
(define (show-n-expressions all-combinations years)
(for-each (lambda (year)
(for-each (lambda (comb)
(when (= (cdr comb) year)
(printf "~s ~a~n" year (car comb))))
all-combinations)
(printf "~n"))
years))
Use it like this for replicating the results in the original blog post:
(define numbers '("10" "9" "8" "7" "6" "5" "4" "3" "2" "1"))
(define ops '("" "+" "-" "*" "/"))
; beware: this takes around 35 minutes to finish in my laptop
(define all-combinations (make-all-combinations numbers ops))
(show-n-expressions all-combinations
(build-list 5 (lambda (n) (+ n 2012))))
UPDATE :
I snarfed Eli Barzilay's expression evaluator and plugged it into my solution, now the pre-calculation of all combinations is done in around 5 seconds! The show-n-expressions procedure still needs some work to avoid iterating over the whole list of combinations each time, but that's left as an exercise for the reader. What matters is that now brute-forcing the values for all the possible expression combinations is blazing fast.
#lang racket
(define (tuples lst n)
(if (zero? n)
'(())
(for*/list ((y (in-list (tuples lst (sub1 n))))
(x (in-list lst)))
(cons x y))))
(define (riffle numbers ops)
(if (null? ops)
(list (car numbers))
(cons (car numbers)
(cons (car ops)
(riffle (cdr numbers)
(cdr ops))))))
(define (expression-string numbers optuple)
(string-append*
(map (lambda (x)
(cond ((eq? x '&) "")
((symbol? x) (symbol->string x))
((number? x) (number->string x))))
(riffle numbers optuple))))
(define eval-ops
(let ((precedence (make-hasheq
'((& . 3) (/ . 2) (* . 2)
(- . 1) (+ . 1) (#f . 0))))
(apply-op (lambda (op x y)
(case op
((+) (+ x y)) ((-) (- x y))
((*) (* x y)) ((/) (/ x y))
((&) (+ (* 10 x) y))))))
(lambda (nums ops)
(let loop ((nums (cddr nums))
(ops (cdr ops))
(numstack (list (cadr nums) (car nums)))
(opstack (list (car ops))))
(if (and (null? ops) (null? opstack))
(car numstack)
(let ((op (and (pair? ops) (car ops)))
(topop (and (pair? opstack) (car opstack))))
(if (> (hash-ref precedence op)
(hash-ref precedence topop))
(loop (cdr nums)
(cdr ops)
(cons (car nums) numstack)
(cons op opstack))
(loop nums
ops
(cons (apply-op topop (cadr numstack) (car numstack))
(cddr numstack))
(cdr opstack)))))))))
(define (make-all-combinations numbers ops)
(foldl (lambda (optuple tail)
(cons (cons (eval-ops numbers optuple) optuple) tail))
empty (tuples ops (sub1 (length numbers)))))
(define (show-n-expressions all-combinations numbers years)
(for-each (lambda (year)
(for-each (lambda (comb)
(when (= (car comb) year)
(printf "~s ~a~n"
year
(expression-string numbers (cdr comb)))))
all-combinations)
(printf "~n"))
years))
Use it like this:
(define numbers '(10 9 8 7 6 5 4 3 2 1))
(define ops '(& + - * /))
; this is very fast now!
(define all-combinations (make-all-combinations numbers ops))
(show-n-expressions all-combinations numbers
(build-list 5 (lambda (n) (+ n 2012))))
As Óscar points out, the problem is that soegaard/infix is slow for this type of problem.
I found a standard shunting-yard parser for infix expressions on GitHub and wrote the following program in Racket:
#lang racket
(require "infix-calc.scm")
(define operators '("*" "/" "+" "-" ""))
(time
(for*/list ([o1 (in-list operators)]
[o2 (in-list operators)]
[o3 (in-list operators)]
[o4 (in-list operators)]
[o5 (in-list operators)]
[o6 (in-list operators)]
[o7 (in-list operators)]
[o8 (in-list operators)]
[o9 (in-list operators)]
[expr (in-value
(apply string-append
(list "1" o1 "2" o2 "3" o3 "4" o4 "5" o5 "6" o6 "7" o7 "8" o8 "9" o9 "10")))]
#:when (= (first (calc expr)) 2012))
expr))
After a little less than 3 minutes the results are:
Welcome to DrRacket, version 5.2.900.2--2012-03-29(8c22c6c/a) [3m].
Language: racket; memory limit: 128 MB.
cpu time: 144768 real time: 148818 gc time: 25252
'("1*2*3+4*567*8/9-10"
"1*2+34*56+7+89+10"
"1*23+45*6*7+89+10"
"1+2+3/4*5*67*8+9-10"
"1+2+3+4*567*8/9-10"
"1+2+34*56+7+8+9*10"
"1+23+45*6*7+8+9*10"
"1-2+345*6-7*8+9-10"
"12*34*5+6+7*8-9*10"
"12*34*5+6-7-8-9-10"
"1234+5-6+789-10")
The infix parser was written by Andrew Levenson.
The parser and the above code can be found here:
https://github.com/soegaard/Scheme-Infix-Calculator
this isn't a complete answer, but i think it's an alternative to the library Óscar López is asking for. unfortunately it's in clojure, but hopefully it's clear enough...
(def default-priorities
{'+ 1, '- 1, '* 2, '/ 2, '& 3})
(defn- extend-tree [tree priorities operator value]
(if (seq? tree)
(let [[op left right] tree
[old new] (map priorities [op operator])]
(if (> new old)
(list op left (extend-tree right priorities operator value))
(list operator tree value)))
(list operator tree value)))
(defn priority-tree
([operators values] (priority-tree operators values default-priorities))
([operators values priorities] (priority-tree operators values priorities nil))
([operators values priorities tree]
(if-let [operators (seq operators)]
(if tree
(recur
(rest operators) (rest values) priorities
(extend-tree tree priorities (first operators) (first values)))
(let [[v1 v2 & values] values]
(recur (rest operators) values priorities (list (first operators) v1 v2))))
tree)))
; [] [+ & *] [1 2 3 4] 1+23*4
; [+ 1 2] [& *] [3 4] - initial tree
; [+ 1 [& 2 3]] [*] [4] - binds more strongly than + so replace right-most node
; [+ 1 [* [& 2 3] 4]] [] [] - descend until do not bind more tightly, and extend
(println (priority-tree ['+ '& '*] [1 2 3 4])) ; 1+23*4
(println (priority-tree ['& '- '* '+ '&] [1 2 3 4 5 6])) ; 12 - 3*4 + 56
the output is:
(+ 1 (* (& 2 3) 4))
(+ (- (& 1 2) (* 3 4)) (& 5 6))
[update] adding the following
(defn & [a b] (+ b (* 10 a)))
(defn all-combinations [tokens length]
(if (> length 0)
(for [token tokens
smaller (all-combinations tokens (dec length))]
(cons token smaller))
[[]]))
(defn all-expressions [operators digits]
(map #(priority-tree % digits)
(all-combinations operators (dec (count digits)))))
(defn all-solutions [target operators digits]
(doseq [expression
(filter #(= (eval %) target)
(all-expressions operators digits))]
(println expression)))
(all-solutions 2012 ['+ '- '* '/ '&] (range 10 0 -1))
solves the problem, but it's slow - 28 minutes to complete. this is on a nice, fairly recent laptop (i7-2640M).
(+ (- (+ 10 (* 9 (& 8 7))) (& 6 5)) (* 4 (& (& 3 2) 1)))
(+ (- (+ (+ (* (* 10 9) 8) 7) 6) 5) (* 4 (& (& 3 2) 1)))
(- (- (+ (- (& 10 9) (* 8 7)) (* (& (& 6 5) 4) 3)) 2) 1)
(i only printed 2012 - see code above - but it would have evaluated the entire sequence).
so, unfortunately, this doesn't really answer the question, since it's no faster than Óscar López's code. i guess the next step would be to put some smarts into the evaluation and so save some time. but what?
[update 2] after reading the other posts here i replaced eval with
(defn my-eval [expr]
(if (seq? expr)
(let [[op left right] expr]
(case op
+ (+ (my-eval left) (my-eval right))
- (- (my-eval left) (my-eval right))
* (* (my-eval left) (my-eval right))
/ (/ (my-eval left) (my-eval right))
& (& (my-eval left) (my-eval right))))
expr))
and the running time drops to 45 secs. still not great, but it's a very inefficient parse/evaluation.
[update 3] for completeness, the following is an implementation of the shunting-yard algorithm (a simple one that is always left-associative) and the associated eval, butit only reduces the time to 35s.
(defn shunting-yard
([operators values] (shunting-yard operators values default-priorities))
([operators values priorities]
(let [[value & values] values]
(shunting-yard operators values priorities nil (list value))))
([operators values priorities stack-ops stack-vals]
; (println operators values stack-ops stack-vals)
(if-let [[new & short-operators] operators]
(let [[value & short-values] values]
(if-let [[old & short-stack-ops] stack-ops]
(if (> (priorities new) (priorities old))
(recur short-operators short-values priorities (cons new stack-ops) (cons value stack-vals))
(recur operators values priorities short-stack-ops (cons old stack-vals)))
(recur short-operators short-values priorities (list new) (cons value stack-vals))))
(concat (reverse stack-vals) stack-ops))))
(defn stack-eval
([stack] (stack-eval (rest stack) (list (first stack))))
([stack values]
(if-let [[op & stack] stack]
(let [[right left & tail] values]
(case op
+ (recur stack (cons (+ left right) tail))
- (recur stack (cons (- left right) tail))
* (recur stack (cons (* left right) tail))
/ (recur stack (cons (/ left right) tail))
& (recur stack (cons (& left right) tail))
(recur stack (cons op values))))
(first values))))
Interesting! I had to try it, it's in Python, hope you don't mind. It runs in about 28 seconds, PyPy 1.8, Core 2 Duo 1.4
from __future__ import division
from math import log
from operator import add, sub, mul
div = lambda a, b: float(a) / float(b)
years = set(range(2012, 2113))
none = lambda a, b: a * 10 ** (int(log(b, 10)) + 1) + b
priority = {none: 3, mul: 2, div: 2, add: 1, sub: 1}
symbols = {none: '', mul: '*', div: '/', add: '+', sub: '-', None: ''}
def evaluate(numbers, operators):
ns, ops = [], []
for n, op in zip(numbers, operators):
while ops and (op is None or priority[ops[-1]] >= priority[op]):
last_n = ns.pop()
last_op = ops.pop()
n = last_op(last_n, n)
ns.append(n)
ops.append(op)
return n
def display(numbers, operators):
return ''.join([
i for n, op in zip(numbers, operators) for i in (str(n), symbols[op])])
def expressions(years):
numbers = 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
operators = none, add, sub, mul, div
pools = [operators] * (len(numbers) - 1) + [[None]]
result = [[]]
for pool in pools:
result = [x + [y] for x in result for y in pool]
for ops in result:
expression = evaluate(numbers, ops)
if expression in years:
yield '%d = %s' % (expression, display(numbers, ops))
for year in sorted(expressions(years)):
print year