I've copied code from the book "grokking algorithms" for finding an item in the list using the binary search algorithm. The code launches great and returns no errors. However, sometimes it just doesn't work for certain numbers, for example if i call it to find "1". What's wrong?
def binary_search(list, item):
low = 0
high = len(list) - 1
while low <= high:
mid = (low + high)/2
guess = list[mid]
if guess == item:
return mid
if guess > item:
high = mid + 1
else:
low = mid + 1
return None
list1 = []
for i in range (1, 101):
list1.append(i)
print(list1)
print(binary_search(list1, 1))
Two issues:
Use integer division (so it will work in Python 3): mid = (low + high)//2
When guess > item you want to exclude the guessed value from the next range, but with high = mid + 1 it still is in the range. Correct to high = mid - 1
So I understand conceptually how binary search works, but I always have problems with implementing it when trying to find an index in an array. For instance, for the Search Insert Position on LC, this is what I wrote:
def searchInsert(self, nums, target):
"""
:type nums: List[int]
:type target: int
:rtype: int
"""
if target > nums[-1]:
return len(nums)
low = 0
high = len(nums) - 1
while high > low:
mid = (low + high) // 2
if nums[mid] == target:
return mid
elif nums[mid] > target:
high = mid
else:
low = mid + 1
return low
It works, but I don't understand why I have to update low as mid + 1 instead of updating low as mid. Similarly, why am I updating high as mid instead of mid - 1. I've tried updating low/high as every combination of mid, mid - 1, and mid + 1 and the above is the only one that works but I have no idea why.
When implementing binary search for these kinds of problems, is there a way to reason through how you update the low/high values?
This is personal favorite:
while high >= low:
mid = (low + high) // 2
if nums[mid] >= target:
high = mid - 1
else:
low = mid + 1
return low
# or return nums[low] == target for boolean
It has difference in the case that has same values.
for example, Let's assume the array is [1,1,2,2,3,3,3,3,4].
with your function, search(arr, 1) returned 1 BUT search(arr, 2) returned 2.
why does it returned most RIGHT index on the interval 1s, and returned most LEFT index on 2s?
As i think, the key is at if nums[mid] >= target:.
when it finds the target exactly same one, the range changes by high = mid - 1. it means high might not be answer because the answer we found is mid. [1]
At the last step of binary search, the range is going to close to zero. and finally loop breaks by they crossed. thus, the answer must be low or high. but we know high is not an answer at [1].
I have made a binary search algorithm, biSearch(A, high, low, key). It takes in a sorted array and a key, and spits out the position of key in the array. High and low are the min and max of the search range.
It almost works, save for one problem:
On the second "iteration" (not sure what the recursive equivalent of that is), a condition is met and the algorithm should stop running and return "index". I commented where this happens. Instead, what ends up happening is that the code continues on to the next condition, even though the preceding condition is true. The correct result, 5, is then overridden and the new result is a nonetype object.
within my code, I have commented in caps the problems at the location in which they occur. Help is much appreciated, and I thank you in advance!
"""
Created on Sat Dec 28 18:40:06 2019
"""
def biSearch(A, key, low = False, high = False):
if low == False:
low = 0
if high == False:
high = len(A)-1
if high == low:
return A[low]
mid = low + int((high -low)/ 2)
# if key == A[mid] : two cases
if key == A[mid] and high - low == 0: #case 1: key is in the last pos. SHOULD STOP RUNNING HERE
index = mid
return index
elif key == A[mid] and (high - low) > 0:
if A[mid] == A[mid + 1] and A[mid]==A[mid -1]: #case 2: key isnt last and might be repeated
i = mid -1
while A[i] == A[i+1]:
i +=1
index = list(range(mid- 1, i+1))
elif A[mid] == A[mid + 1]:
i = mid
while A[i]== A[i+1]:
i += 1
index = list(range(mid, i+1))
elif A[mid] == A[mid -1]:
i = mid -1
while A[i] == A[i +1]:
i += 1
index = list(range(mid, i +1))
elif key > A[mid] and high - low > 0: # BUT CODE EXECTUES THIS LINE EVEN THOUGH PRECEDING IS ALREADY MET
index = biSearch(A, key, mid+1, high)
elif key < A[mid] and high - low > 0:
index = biSearch(A, key, low, mid -1)
return index
elif A[mid] != key: # if key DNE in A
return -1
#biSearch([1,3,5, 4, 7, 7,7,9], 1, 8, 7)
#x = biSearch([1,3,5, 4, 7,9], 1, 6, 9)
x = biSearch([1,3,5, 4, 7,9],9)
print(x)
# x = search([1,3,5, 4, 7,9], 9)
This function is not a binary search. Binary search's time complexity should be O(log(n)) and works on pre-sorted lists, but the complexity of this algorithm is at least O(n log(n)) because it sorts its input parameter list for every recursive call. Even without the sorting, there are linear statements like list(range(mid, i +1)) on each call, making the complexity quadratic. You'd be better off with a linear search using list#index.
The function mutates its input parameter, which no search function should do (we want to search, not search and sort).
Efficiencies and mutation aside, the logic is difficult to parse and is overkill in any circumstance. Not all nested conditionals lead to a return, so it's possible to return None by default.
You can use the builtin bisect module:
>>> from bisect import *
>>> bisect_left([1,2,2,2,2,3,4,4,4,4,5], 2)
1
>>> bisect_left([1,2,2,2,2,3,4,4,4,4,5], 4)
6
>>> bisect_right([1,2,2,2,2,3,4,4,4,4,5], 4)
10
>>> bisect_right([1,2,2,2,2,3,4,4,4,4,5], 2)
5
>>> bisect_right([1,2,2,2,2,3,4,4,4,4,5], 15)
11
>>> bisect_right([1,2,5,6], 3)
2
If you have to write this by hand as an exercise, start by looking at bisect_left's source code:
def bisect_left(a, x, lo=0, hi=None):
"""Return the index where to insert item x in list a, assuming a is sorted.
The return value i is such that all e in a[:i] have e < x, and all e in
a[i:] have e >= x. So if x already appears in the list, a.insert(x) will
insert just before the leftmost x already there.
Optional args lo (default 0) and hi (default len(a)) bound the
slice of a to be searched.
"""
if lo < 0:
raise ValueError('lo must be non-negative')
if hi is None:
hi = len(a)
while lo < hi:
mid = (lo+hi)//2
# Use __lt__ to match the logic in list.sort() and in heapq
if a[mid] < x: lo = mid+1
else: hi = mid
This is easy to implement recursively (if desired) and then test against the builtin:
def bisect_left(a, target, lo=0, hi=None):
if hi is None: hi = len(a)
mid = (hi + lo) // 2
if lo >= hi:
return mid
elif a[mid] < target:
return bisect_left(a, target, mid + 1, hi)
return bisect_left(a, target, lo, mid)
if __name__ == "__main__":
from bisect import bisect_left as builtin_bisect_left
from random import choice, randint
from sys import exit
for _ in range(10000):
a = sorted(randint(0, 100) for _ in range(100))
if any(bisect_left(a, x) != builtin_bisect_left(a, x) for x in range(-1, 101)):
print("fail")
exit(1)
Logically, for any call frame, there's only 3 possibilities:
The lo and hi pointers have crossed, in which case we've either found the element or figured out where it should be if it were in the list; either way, return the midpoint.
The element at the midpoint is less than the target, which guarantees that the target is in the tail half of the search space, if it exists.
The element at the midpoint matches or is less than the target, which guarantees that the target is in the front half of the search space.
Python doesn't overflow integers, so you can use the simplified midpoint test.
I am thinking this particular code is (log n)^2 because each findindex function takes logn depth and we are calling it logn times? Can someone confirm this?
I hope one of you can think of this as a small quiz and help me with it.
Given a sorted array of n integers that has been rotated an unknown
number of times, write code to find an element in the array. You may
assume that the array was originally sorted in increasing order.
# Ex
# input find 5 in {15,16,19,20,25,1,3,4,5,7,10,14}
# output 8
# runtime(log n)
def findrotation(a, tgt):
return findindex(a, 0, len(a)-1, tgt, 0)
def findindex(a, low, high, target, index):
if low>high:
return -1
mid = int((high + low) / 2)
if a[mid] == target:
index = index + mid
return index
else:
b = a[low:mid]
result = findindex(b, 0, len(b)-1, target, index)
if result == -1:
index = index + mid + 1
c = a[mid+1:]
return findindex(c, 0, len(c)-1, target, index)
else:
return result
This algorithm is supposed to be O(logn) but is not from implementation perspectives.
In your algorithm, you're not making decision either to go for left subarray or right subarray only, you're trying with both subarray which is O(N).
You're doing slicing on array a[low:mid] and a[mid + 1:] which is O(n).
Which makes your overall complexity O(n^2) in worst case.
Assuming there is no duplicates in the array, an ideal implementation in Python 3 of O(logn) binary search looks like this -
A=[15,16,19,20,25,1,3,4,5,7,10,14]
low = 0
hi = len(A) - 1
def findindex(A, low, hi, target):
if low > hi:
return -1
mid = round((hi + low) / 2.0)
if A[mid] == target:
return mid
if A[mid] >= A[low]:
if target < A[mid] and target >= A[low]:
return findindex(A, low, mid - 1, target)
else :
return findindex(A, mid + 1, hi, target)
if A[mid] < A[low]:
if target < A[mid] or target >= A[low]:
return findindex(A, low, mid - 1, target)
else :
return findindex(A, mid + 1, hi, target)
return -1
print(findindex(A, low, hi, 3))
Here is the code for binary search, wondering if there is no match value, the question is the upper bound should be positioned at either small or small+1? And lower bound should be located at small or small-1, correct? Thanks.
def binary_search(my_list, key):
assert my_list == sorted(my_list)
large = len(my_list) -1
small = 0
while (small <= large):
mid = (small + large) // 2 )
# print small, mid, large, " - ", my_list[mid], "~", key
if my_list[mid] < key:
small = mid + 1
elif my_list[mid] > key:
large = mid - 1
else:
return mid
raise ValueError
#Paul is close, it should return mid+1 for the upper value. Instead of raising an exception, here is the code in Python that will return the lower and upper values:
value = my_list[mid]
lower = mid if value < key else mid-1
upper = mid if value > key else mid+1
return (lower, upper)
Here is an even simpler answer:
return (small-1, small)
Because small ends up above the key's index when you're looking for it, the lower bound is small-1 and the upper bound is small.
That's not that simple. You'll have to take account of the fact, that the algorithm might either approximate the position of the searched item from the left (lower values) or right (higher values), which can't be determined after exiting the while-loop. Thus you'll have to check whether the value at mid is smaller or greater than the key:
lower = (my_list[mid] < key ? my_list[mid] : my_list[mid - 1]);
upper = (my_list[mid] > key ? my_list[mid] : my_list[mid + 1]);