# Counting in pyramids

You should write a program or function which receives a list of distinct integers as input and outputs or returns the number of occurrences of the input numbers in the following upside-down number pyramid.

Starting from the original list in every step we create a new one with the maximal values of every pair of adjacent numbers (e.g. 5 1 2 6 becomes 5 2 6). We stop when there is only one number in the list.

The full pyramid for 5 1 2 6 is

5 1 2 6
5 2 6
5 6
6


The resulting number of occurrences are 3 1 2 4 (for 5 1 2 6 respectively).

## Input

• A list of one or more integers with no repetition. (e.g. 1 5 1 6 is invalid.)

## Output

• A list of positive integers. The ith element of the list is the number of occurrences of the ith input number in the pyramid.

## Examples

Input => Output

-5 => 1

8 4 => 2 1

5 9 7 => 1 4 1

1 2 3 9 8 6 7 => 1 2 3 16 3 1 2

6 4 2 1 3 5 => 6 4 2 1 3 5

5 2 9 1 6 0 => 2 1 12 1 4 1

120 5 -60 9 12 1 3 0 1200 => 8 2 1 3 16 1 4 1 9

68 61 92 58 19 84 75 71 46 69 25 56 78 10 89 => 2 1 39 2 1 27 6 5 1 6 1 2 14 1 12


This is code-golf so the shortest entry wins.

Bonus puzzle: can you solve the problem in O(n*log n) time?

• For a function submission, do I have to print them to STDOUT or simply output them ? Apr 26, 2015 at 8:01

# Python, 81

def f(L):
if L:i=L.index(max(L));L=f(L[:i])+[~i*(i-len(L))]+f(L[i+1:])
return L


A divide-and-conquer solution. The maximum element M seeps all the way down the pyramid, splitting it into a rectangle of M's and two subpyramids.

* * * M * *
* * M M *
* M M M
M M M
M M
M


So, the overall result is the output for the left sublist, followed by the area of the rectangle, followed by the output for the right sublist.

The input variable L is reused to stored the result so that the empty list is mapped to the empty list.

The constructs in solution are wordy in Python. Maybe some language with pattern-matching can implement the following pseudocode?

def f(L):
[] -> []
A+[max(L)]+B -> f(A)+[(len(A)+1)*(len(B)+1)]+f(B)

• I can do one byte shorter with Mathematica's pattern matching, but it doesn't even beat the existing Mathematica submission: f@{}=##&@@{};f@{a___,l_,b___}/;l>a~Max~b:={f@{a},Length@{a,0}Length@{b,0},f@{b}} Apr 26, 2015 at 11:03

# CJam, 23 22 bytes

Still looking for golfing options.

{]_,{)W$ew::e>~}%fe=~}  This is a CJam function (sort of). This expects the input numbers on stack and returns the corresponding output counts on the stack too. An example: 5 1 2 6 {]_,{)W$ew::e>~}%fe=~}~


leaves

3 1 2 4


on stack.

Pretty sure that this is not in O(n log n) time.

Code expansion:

]_                     e# Wrap the input numbers on stack in an array and take a copy
,{          }%       e# Take length of the copy and run the loop from 0 to length - 1
)W$e# Increment the iterating index and copy the parsed input array ew e# Get overlapping slices of iterating index + 1 size ::e> e# Get maximum from each slice ~ e# Unwrap so that there can be finally only 1 level array fe= e# For each of the original array, get the occurrence in this e# final array created by the { ... }% ~ e# Unwrap the count array and leave it on stack  Lets take a look at how it works by working up an example of 5 1 2 6 In the second row, 5 1 2 6 becomes 5 2 6 because 5, 2 and 6 are the maximum of [5 1], [1 2] and [2 6] respectively. In the third row, it becomes 5 6 because 5 and 6 are maximum of [5 2] and [2 6] respectively. This can also be written as maximum of [5 1 2] and [1 2 6] respectively. Similarly for last row, 6 is maximum of [5 1 2 6]. So we basically create the proper length slices starting from slice of length 1, which is basically the original numbers, each wrapped in an array, to finally a slice of length N for the last row, where N is the number of input integers. Try it online here # Pyth, 19 17 bytes m/smmeSb.:QhkUQdQ  Check out the Online Demonstration or the complete Test Suite (first 4 byte iterate over the examples). This one is a little bit more clever than the naive approach. Each number of the triangle can be represented as the maximal value of a connected subset of Q. In the first line it uses the subsets of length 1, the second line of the triangle uses the subsets of length 2, ... ### Explanation m/smmeSb.:QhkUQdQ implicit: Q = input() m UQ map each k in [0, 1, 2, ..., len(Q)-1] to: .:Qhk all subsets of Q of length (k + 1) meSb mapped to their maximum s join these lists together m Q map each d of Q to: / d its count in the computed list  To visualize this a little bit. m.:QhdUQ and the input [5, 1, 2, 6] gives me all possible subsets: [[, , , ], [[5, 1], [1, 2], [2, 6]], [[5, 1, 2], [1, 2, 6]], [[5, 1, 2, 6]]]  And mmeSk.:QhdUQ gives me each of their maxima (which corresponds exactly to the rows in the pyramid): [[5, 1, 2, 6], [5, 2, 6], [5, 6], ]  # Pyth, 23 22 bytes |u&aYGmeSd.:G2QQm/sYdQ  This is just the simple "do what you're told" approach. Check out the Online Demonstration or a complete Test Suite (first 4 byte iterate over the examples). ### Explanation meSd.:G2 maps each pair of [(G, G), (G, G), ...] to the maximal element. Y is an empty list, therefore aYG appends G to the Y. u...QQ repeatedly applies these two functions (len(Q) times) starting with G = Q and updating G after each run. m/sYdQ maps each element of the input list to their count in the flattened Y list. • your 17 byte version uses the same algorithm as mine, I guess that is now naive too :P Apr 26, 2015 at 16:23 # Mathematica, 72 bytes Last/@Tally[Join@@NestList[MapThread[Max,{Most@#,Rest@#}]&,#,Length@#]]&  # Python, 81 lambda L:[sum(x==max(L[i:j])for j in range(len(L)+1)for i in range(j))for x in L]  Each entry of the pyramid is the maximum of the sublist atop its upward cone. So, we generate all these sublists, indexed by intervals [i,j] with 0 < i < j <= len(L), and count how many time each element appears as a maximum. A shorter way to enumerate the subintervals would likely save characters. A single-index parametrization of the pairs [i,j] would be a plausible approach. # Pip, 56 + 1 = 57 bytes Not competing much with the CJam voodoo, I'm afraid. Looks like I need a better algorithm. Run with -s flag to get space-delimited output. l:gr:0*,#gg:0*g+1WrFir:{c:r@[a--a]c@($<l@c)}M1,#r++(gi)g


l:g                              l = input from cmdline args
r:0*,#g                          r = current row as a list of indices into l
g:0*g+1                          Repurpose g to store the frequencies
Wr                               Loop until r becomes empty
Fir:{c:r@[a--a]c@($<l@c)}M1,#r Redefine r (see below) and loop over each i in it ++(gi) Increment g[i] g Output g  The redefinition of r each time through works as follows: {c:r@[a--a]c@($<l@c)}M1,#r
{                   }M1,#r       Map this function to each a from 1 to len(r) - 1:
c:r@[a--a]                      c is a two-item list containing r[a] and r[a-1]
l@c              The values of l at the indices contained in c
$< Fold/less-than: true iff l[c] < l[c] c@( ) Return c if the former is greater, c otherwise  # APL (24) {+/⍵∘.={⍵≡⍬:⍵⋄⍵,∇2⌈/⍵}⍵}  This is a function that takes a list, like so;  {+/⍵∘.={⍵≡⍬:⍵⋄⍵,∇2⌈/⍵}⍵}68 61 92 58 19 84 75 71 46 69 25 56 78 10 89 2 1 39 2 1 27 6 5 1 6 1 2 14 1 12  Explanation: • {...}⍵: apply the following function to ⍵: • ⍵≡⍬:⍵: if ⍵ is empty, return ⍵ • 2⌈/⍵: generate the next list • ⍵,∇: return ⍵, followed by the result of applying this function to the next list • ⍵∘.=: compare each element in ⍵ to each element in the result of the function • +/: sum the rows (representing elements in ⍵) # Haskell, 78 bytes l=length f x=[l[b|b<-concat$take(l x)\$iterate(zipWith max=<<tail)x,a==b]|a<-x]


Usage: f [68,61,92,58,19,84,75,71,46,69,25,56,78,10,89]-> [2,1,39,2,1,27,6,5,1,6,1,2,14,1,12].

How it works

zipWith max=<<tail   -- apply 'max' on neighbor elements of a list
iterate (...) x      -- repeatedly apply the above max-thing on the
-- input list and build a list of the intermediate
-- results
take (l x) ...       -- take the first n elements of the above list
-- where n is the length of the input list
concat               -- concatenate into a single list. Now we have
-- all elements of the pyramid in a single list.
[ [b|b<-...,a==b] | a<-x]
-- for all elements 'a' of the input list make a
-- list of 'b's from the pyramid-list where a==b.
l                   -- take the length of each of these lists


# JavaScript, 109 bytes

I think I found an interesting way to go about this, but only after I was done realized the code was too long to compete. Oh well, posting this anyway in case someone sees further golfing potential.

f=s=>{t=[];for(i=-1;s.length>++i;){j=k=i;l=r=1;for(;s[--j]<s[i];l++);for(;s[++k]<s[i];r++);t[i]=l*r}return t}


I'm making use of the following formula here:

occurrences of i = (amount of consecutive numbers smaller than i to its left + 1) * (amount of consecutive numbers smaller than i to its right + 1)

This way one doesn't need to actually generate the entire pyramid or subsets of it. (Which is why I initially thought it would run in O(n), but tough luck, we still need inner loops.)

## MATLAB : (266 b)

• the correction of the code costs more bytes, i will struggle how to reduce it later.
v=input('');h=numel(v);for i=1:h,f=(v(i)>v(1));l=(v(i)>v(h));for j=h-1:-1:i+1,l=(v(i)>v(j))*(1+l);end,if(i>1),l=l+(v(i)>v(i-1))*l;end;for j=2:i-1,f=(v(i)>v(j))*(1+f);end,if(i<h),f=f+(v(i)>v(i+1))*f;end;s=f+l+1;if(i<h&&i>1),s=s-((v(i)>v(i+1))*(v(i)>v(i-1)));end;s
end


## INPUT

a vector must be of the form [a b c d ...]

• example:

[2 4 7 11 3]

## OUTPUT

pattern occurences.

s =

1

s =

2

s =

3

s =

8

s =

1


## EXPLANATION:

if [a b c d] is an input the program calculates result g h i j as

g= (a>b) + (a>b)(a>c) + (a>b)(a>c)*(a>d) = (a>b)(1+(a>c)(1+(a>c) )))

h= (b>a) + (b>c) + (b>a)(b>c) + (b>c)(b>d)+(b>a)(b>c)(b>d) = ... 'simplified'

i= (c>b) + (c>d) + (c>b)(c>d) + (c>b)(c>a)+(c>d)(c>b)(c>a) = ..

j= (d>c) + (d>c)(d>b) + (d>c)(d>b)*(d>a) = ...

# J (49)

I suppose there is some room for improvement...

[:+/~.="1 0[:;([:i.#)<@:(4 :'(}:>.}.)^:x y')"0 1]