heapq.py revision a0b3a00bc5f55cfbdc3d9b7925ee8a28fa2bdc55
1"""Heap queue algorithm (a.k.a. priority queue). 2 3Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for 4all k, counting elements from 0. For the sake of comparison, 5non-existing elements are considered to be infinite. The interesting 6property of a heap is that a[0] is always its smallest element. 7 8Usage: 9 10heap = [] # creates an empty heap 11heappush(heap, item) # pushes a new item on the heap 12item = heappop(heap) # pops the smallest item from the heap 13item = heap[0] # smallest item on the heap without popping it 14 15Our API differs from textbook heap algorithms as follows: 16 17- We use 0-based indexing. This makes the relationship between the 18 index for a node and the indexes for its children slightly less 19 obvious, but is more suitable since Python uses 0-based indexing. 20 21- Our heappop() method returns the smallest item, not the largest. 22 23These two make it possible to view the heap as a regular Python list 24without surprises: heap[0] is the smallest item, and heap.sort() 25maintains the heap invariant! 26""" 27 28# Code by Kevin O'Connor 29 30__about__ = """Heap queues 31 32[explanation by Fran�ois Pinard] 33 34Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for 35all k, counting elements from 0. For the sake of comparison, 36non-existing elements are considered to be infinite. The interesting 37property of a heap is that a[0] is always its smallest element. 38 39The strange invariant above is meant to be an efficient memory 40representation for a tournament. The numbers below are `k', not a[k]: 41 42 0 43 44 1 2 45 46 3 4 5 6 47 48 7 8 9 10 11 12 13 14 49 50 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 51 52 53In the tree above, each cell `k' is topping `2*k+1' and `2*k+2'. In 54an usual binary tournament we see in sports, each cell is the winner 55over the two cells it tops, and we can trace the winner down the tree 56to see all opponents s/he had. However, in many computer applications 57of such tournaments, we do not need to trace the history of a winner. 58To be more memory efficient, when a winner is promoted, we try to 59replace it by something else at a lower level, and the rule becomes 60that a cell and the two cells it tops contain three different items, 61but the top cell "wins" over the two topped cells. 62 63If this heap invariant is protected at all time, index 0 is clearly 64the overall winner. The simplest algorithmic way to remove it and 65find the "next" winner is to move some loser (let's say cell 30 in the 66diagram above) into the 0 position, and then percolate this new 0 down 67the tree, exchanging values, until the invariant is re-established. 68This is clearly logarithmic on the total number of items in the tree. 69By iterating over all items, you get an O(n ln n) sort. 70 71A nice feature of this sort is that you can efficiently insert new 72items while the sort is going on, provided that the inserted items are 73not "better" than the last 0'th element you extracted. This is 74especially useful in simulation contexts, where the tree holds all 75incoming events, and the "win" condition means the smallest scheduled 76time. When an event schedule other events for execution, they are 77scheduled into the future, so they can easily go into the heap. So, a 78heap is a good structure for implementing schedulers (this is what I 79used for my MIDI sequencer :-). 80 81Various structures for implementing schedulers have been extensively 82studied, and heaps are good for this, as they are reasonably speedy, 83the speed is almost constant, and the worst case is not much different 84than the average case. However, there are other representations which 85are more efficient overall, yet the worst cases might be terrible. 86 87Heaps are also very useful in big disk sorts. You most probably all 88know that a big sort implies producing "runs" (which are pre-sorted 89sequences, which size is usually related to the amount of CPU memory), 90followed by a merging passes for these runs, which merging is often 91very cleverly organised[1]. It is very important that the initial 92sort produces the longest runs possible. Tournaments are a good way 93to that. If, using all the memory available to hold a tournament, you 94replace and percolate items that happen to fit the current run, you'll 95produce runs which are twice the size of the memory for random input, 96and much better for input fuzzily ordered. 97 98Moreover, if you output the 0'th item on disk and get an input which 99may not fit in the current tournament (because the value "wins" over 100the last output value), it cannot fit in the heap, so the size of the 101heap decreases. The freed memory could be cleverly reused immediately 102for progressively building a second heap, which grows at exactly the 103same rate the first heap is melting. When the first heap completely 104vanishes, you switch heaps and start a new run. Clever and quite 105effective! 106 107In a word, heaps are useful memory structures to know. I use them in 108a few applications, and I think it is good to keep a `heap' module 109around. :-) 110 111-------------------- 112[1] The disk balancing algorithms which are current, nowadays, are 113more annoying than clever, and this is a consequence of the seeking 114capabilities of the disks. On devices which cannot seek, like big 115tape drives, the story was quite different, and one had to be very 116clever to ensure (far in advance) that each tape movement will be the 117most effective possible (that is, will best participate at 118"progressing" the merge). Some tapes were even able to read 119backwards, and this was also used to avoid the rewinding time. 120Believe me, real good tape sorts were quite spectacular to watch! 121From all times, sorting has always been a Great Art! :-) 122""" 123 124def heappush(heap, item): 125 """Push item onto heap, maintaining the heap invariant.""" 126 pos = len(heap) 127 heap.append(None) 128 while pos: 129 parentpos = (pos - 1) >> 1 130 parent = heap[parentpos] 131 if item >= parent: 132 break 133 heap[pos] = parent 134 pos = parentpos 135 heap[pos] = item 136 137def heappop(heap): 138 """Pop the smallest item off the heap, maintaining the heap invariant.""" 139 endpos = len(heap) - 1 140 if endpos <= 0: 141 return heap.pop() 142 returnitem = heap[0] 143 item = heap.pop() 144 pos = 0 145 while True: 146 child2pos = (pos + 1) * 2 147 child1pos = child2pos - 1 148 if child2pos < endpos: 149 child1 = heap[child1pos] 150 child2 = heap[child2pos] 151 if item <= child1 and item <= child2: 152 break 153 if child1 < child2: 154 heap[pos] = child1 155 pos = child1pos 156 continue 157 heap[pos] = child2 158 pos = child2pos 159 continue 160 if child1pos < endpos: 161 child1 = heap[child1pos] 162 if child1 < item: 163 heap[pos] = child1 164 pos = child1pos 165 break 166 heap[pos] = item 167 return returnitem 168 169if __name__ == "__main__": 170 # Simple sanity test 171 heap = [] 172 data = [1, 3, 5, 7, 9, 2, 4, 6, 8, 0] 173 for item in data: 174 heappush(heap, item) 175 sort = [] 176 while heap: 177 sort.append(heappop(heap)) 178 print sort 179