How do I sort a list with positives coming before negatives with values sorted respectively?
Question:
I have a list that contains a mixture of positive and negative numbers, as the following
lst = [1, -2, 10, -12, -4, -5, 9, 2]
What I am trying to accomplish is to sort the list with the positive numbers coming before the negative numbers, respectively sorted as well.
Desired output:
[1, 2, 9, 10, -12, -5, -4, -2]
I was able to figure out the first part sorting with the positive numbers coming before the and negative numbers, unfortunately this does not respectively sort the positive and negative numbers.
lst = [1, -2, 10, -12, -4, -5, 9, 2]
lst = sorted(lst, key=lambda o: not abs(o) == o)
print(lst)
>>> [1, 10, 2, 9, -2, -12, -4, -5]
How may I achieve my desired sorting with a pythonic solution?
Answers:
You could just use a regular sort, and then bisect the list at 0:
>>> lst
[1, -2, 10, -12, -4, -5, 9, 2]
>>> from bisect import bisect
>>> lst.sort()
>>> i = bisect(lst, 0) # use `bisect_left` instead if you want zeroes first
>>> lst[i:] + lst[:i]
[1, 2, 9, 10, -12, -5, -4, -2]
The last line here takes advantage of a slice invariant lst == lst[:n] + lst[n:]
Another option would be to use a tuple as a sort key, and rely on lexicographical ordering of tuples:
>>> sorted(lst, key=lambda x: (x<0, x)) # use <= instead if you want zeroes last
[1, 2, 9, 10, -12, -5, -4, -2]
Create two lists, one with positive value and another with the negative values and then sort the content of each list the way you like. For example:
my_list = [1, -2, 10, -12, -4, -5, 9, 2]
pos_list, neg_list = [], []
for item in my_list:
if item < 0:
neg_list.append(item)
else:
pos_list.append(item)
final_list = sorted(pos_list) + sorted(neg_list)
import numpy as np
lst = [1, -2, 10, -12, -4, -5, 9, 2]
ar = np.array(lst)
lst = list(np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0], reverse = True)], axis = 0))
print(lst)
And if you don’t have to use a list but are happy with numpy arrays then you don’t have to pay the costs of casting, i.e.
import numpy as np
ar = np.array([1, -2, 10, -12, -4, -5, 9, 2])
ar = np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0])], axis = 0)
print(ar)
Create two separate list. One positive values one with negative values. sort the negative list, then concatenate them together:
>>> lst = [1, -2, 10, -12, -4, -5, 9, 2]
>>> sorted([i for i in lst if i > 0]) + sorted([i for i in lst if i =< 0])
[1, 2, 9, 10, -12, -5, -4, -2]
>>>
Just comparing the different ways.
Results:
> Shuffle cost comparison small
shuffle_lst: 0.001181483967229724
shuffle_ar: 0.014688121969811618
> Shuffle cost comparison medium
shuffle_lst: 0.572294642101042
shuffle_ar: 0.3266364939045161
> Shuffle cost comparison large
shuffle_lst: 26.5786890439922
shuffle_ar: 6.284286553971469
+cost -cost
bisectme: 0.004252934013493359 0.003071450046263635
lexicon: 0.010936842067167163 0.009755358099937439
compreh.: 0.0071560649666935205 0.005974580999463797
arrayme: 0.03787591797299683 0.023187796003185213
nplexicon: 0.022204622975550592 0.007516501005738974
npbisect: 0.023507782025262713 0.008819660055451095
+cost -cost
bisectme: 7.716002315981314 7.143707673880272
lexicon: 22.17862514301669 21.606330500915647
compreh.: 8.690494343056343 8.118199700955302
arrayme: 1.5029839979251847 1.1763475040206686
nplexicon: 2.0811527019832283 1.7545162080787122
npbisect: 1.3076487149810418 0.9810122210765257
+cost -cost
bisectme: 180.77819497592282 154.19950593193062
arrayme: 22.476932613993995 16.192646060022525
nplexicon: 41.74795828794595 35.46367173397448
npbisect: 20.13856932707131 13.85428277309984
Code:
import sys
import numpy as np
from timeit import timeit
from bisect import bisect
from random import shuffle
def shuffle_lst():
np.random.shuffle(lst)
def shuffle_ar():
np.random.shuffle(ar)
def bisectme():
np.random.shuffle(lst)
lst.sort()
i = bisect(lst, 0)
return lst[i:] + lst[:i]
def lexicon():
np.random.shuffle(lst)
return sorted(lst, key=lambda x: (x < 0, x))
def comprehension():
np.random.shuffle(lst)
return sorted([i for i in lst if i > 0]) + sorted([i for i in lst if i < 0])
def arrayme():
np.random.shuffle(ar)
return np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0])], axis=0)
def nplexicon():
np.random.shuffle(ar)
return ar[np.lexsort((ar, ar < 0))]
def numpybisect():
np.random.shuffle(ar)
ar.sort()
i = ar.__abs__().argmin()
return np.concatenate((ar[i:], ar[:i]))
nloops = 1000
lst = list(range(-10**1, 0, 1)) + list(range(10**1, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison small")
cost_shuffle_list_small = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_small)
cost_shuffle_array_small = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_small)
lst = list(range(-10**4, 0, 1)) + list(range(10**4, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison medium")
cost_shuffle_list_medium = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_medium)
cost_shuffle_array_medium = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_medium)
nloops = 100
lst = list(range(-10**6, 0, 1)) + list(range(10**6, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison large")
cost_shuffle_list_large = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_large)
cost_shuffle_array_large = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_large)
print()
nloops = 1000
## With small lists/arrays
lst = list(range(-10**1, 0, 1)) + list(range(10**1, -1, -1))
ar = np.array(lst)
print("tttttw/o pen.ttttw. pen.")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(lexicon, number=nloops)
print("lexicon:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(comprehension, number=nloops)
print("compreh.:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_small)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_small)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_small)
print()
## With medium lists/arrays
lst = list(range(-10**4, 0, 1)) + list(range(10**4, -1, -1))
ar = np.array(lst)
print("tttttw/o costttttw. cost")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(lexicon, number=nloops)
print("lexicon:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(comprehension, number=nloops)
print("compreh.:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_medium)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_medium)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_medium)
print()
## With large lists/arrays
nloops = 100
lst = list(range(-10**6, 0, 1)) + list(range(10**6, -1, -1))
ar = np.array(lst)
print("tttttw/o costttttw. cost")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_large)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_large)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_large)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_large)
print()
import numpy as np
l = np.array([1, -2, 10, -12, -4, -5, 9, 2])
l[np.lexsort((l, l < 0))]
array([ 1, 2, 9, 10, -12, -5, -4, -2])
You could just sort by the negative of the element’s inverse:
from __future__ import division
sorted(lst, key=lambda i: 0 if i == 0 else -1 / i)
Taking the inverse switches the order of the magnitudes (larger numbers in the middle, smaller on the outside). Taking the negative reverses the order (positives first, negatives last).
Be aware of the size of your numbers of course and if they’ll cause any over- or underflow issues.
Kudos to wmin for the logic of his solution (accepted answer) which is great. So for completeness, a similar answer to this but based on numpy, which is significantly faster for anything else other than small lists/arrays, is as follows:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
ar = np.array(lst)
ar.sort()
i = ar.__abs__().argmin()
np.concatenate((ar[i:], ar[:i]))
I don’t know if it’s the most Pythonic, and it certainly does not have any bells-and-whistles, but IMO it’s a clear and understandable code:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
pos = list()
neg = list()
for i in lst:
neg.append(i) if i < 0 else pos.append(i)
print(sorted(pos) + sorted(neg))
Sort the list in place twice like so:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
lst.sort()
lst.sort(key=int(0).__gt__) # key is True for items <= 0
This takes advantage of the fact the python sort function/method is stable. That means that items with the same value or key stay in the same order. The first sort puts all the items in order from smallest to largest. For the second sort, all the items < 0 get a key of True, all the items >= 0 get a key of False. Because True (1) > False (0), the second sort moves all the negative items to the end, without changing the order of the negative items.
*Here is another solution: *
lst = [1, -2, 10, -12, -4, -5, 9, 2] # list of values.
x = sorted(lst) # x : [-12, -5, -4, -2, 1, 2, 9, 10]
k, m = [], [] # k : [1, 2, 9, 10] # M : [-12, -5, -4, -2]
for i in x:
if i > 0:
k.append(i)
else:
m.append(i)
w = k + m # w : [1, 2, 9, 10, -12, -5, -4, -2]
print(w)
I have a list that contains a mixture of positive and negative numbers, as the following
lst = [1, -2, 10, -12, -4, -5, 9, 2]
What I am trying to accomplish is to sort the list with the positive numbers coming before the negative numbers, respectively sorted as well.
Desired output:
[1, 2, 9, 10, -12, -5, -4, -2]
I was able to figure out the first part sorting with the positive numbers coming before the and negative numbers, unfortunately this does not respectively sort the positive and negative numbers.
lst = [1, -2, 10, -12, -4, -5, 9, 2]
lst = sorted(lst, key=lambda o: not abs(o) == o)
print(lst)
>>> [1, 10, 2, 9, -2, -12, -4, -5]
How may I achieve my desired sorting with a pythonic solution?
You could just use a regular sort, and then bisect the list at 0:
>>> lst
[1, -2, 10, -12, -4, -5, 9, 2]
>>> from bisect import bisect
>>> lst.sort()
>>> i = bisect(lst, 0) # use `bisect_left` instead if you want zeroes first
>>> lst[i:] + lst[:i]
[1, 2, 9, 10, -12, -5, -4, -2]
The last line here takes advantage of a slice invariant lst == lst[:n] + lst[n:]
Another option would be to use a tuple as a sort key, and rely on lexicographical ordering of tuples:
>>> sorted(lst, key=lambda x: (x<0, x)) # use <= instead if you want zeroes last
[1, 2, 9, 10, -12, -5, -4, -2]
Create two lists, one with positive value and another with the negative values and then sort the content of each list the way you like. For example:
my_list = [1, -2, 10, -12, -4, -5, 9, 2]
pos_list, neg_list = [], []
for item in my_list:
if item < 0:
neg_list.append(item)
else:
pos_list.append(item)
final_list = sorted(pos_list) + sorted(neg_list)
import numpy as np
lst = [1, -2, 10, -12, -4, -5, 9, 2]
ar = np.array(lst)
lst = list(np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0], reverse = True)], axis = 0))
print(lst)
And if you don’t have to use a list but are happy with numpy arrays then you don’t have to pay the costs of casting, i.e.
import numpy as np
ar = np.array([1, -2, 10, -12, -4, -5, 9, 2])
ar = np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0])], axis = 0)
print(ar)
Create two separate list. One positive values one with negative values. sort the negative list, then concatenate them together:
>>> lst = [1, -2, 10, -12, -4, -5, 9, 2]
>>> sorted([i for i in lst if i > 0]) + sorted([i for i in lst if i =< 0])
[1, 2, 9, 10, -12, -5, -4, -2]
>>>
Just comparing the different ways.
Results:
> Shuffle cost comparison small
shuffle_lst: 0.001181483967229724
shuffle_ar: 0.014688121969811618
> Shuffle cost comparison medium
shuffle_lst: 0.572294642101042
shuffle_ar: 0.3266364939045161
> Shuffle cost comparison large
shuffle_lst: 26.5786890439922
shuffle_ar: 6.284286553971469
+cost -cost
bisectme: 0.004252934013493359 0.003071450046263635
lexicon: 0.010936842067167163 0.009755358099937439
compreh.: 0.0071560649666935205 0.005974580999463797
arrayme: 0.03787591797299683 0.023187796003185213
nplexicon: 0.022204622975550592 0.007516501005738974
npbisect: 0.023507782025262713 0.008819660055451095
+cost -cost
bisectme: 7.716002315981314 7.143707673880272
lexicon: 22.17862514301669 21.606330500915647
compreh.: 8.690494343056343 8.118199700955302
arrayme: 1.5029839979251847 1.1763475040206686
nplexicon: 2.0811527019832283 1.7545162080787122
npbisect: 1.3076487149810418 0.9810122210765257
+cost -cost
bisectme: 180.77819497592282 154.19950593193062
arrayme: 22.476932613993995 16.192646060022525
nplexicon: 41.74795828794595 35.46367173397448
npbisect: 20.13856932707131 13.85428277309984
Code:
import sys
import numpy as np
from timeit import timeit
from bisect import bisect
from random import shuffle
def shuffle_lst():
np.random.shuffle(lst)
def shuffle_ar():
np.random.shuffle(ar)
def bisectme():
np.random.shuffle(lst)
lst.sort()
i = bisect(lst, 0)
return lst[i:] + lst[:i]
def lexicon():
np.random.shuffle(lst)
return sorted(lst, key=lambda x: (x < 0, x))
def comprehension():
np.random.shuffle(lst)
return sorted([i for i in lst if i > 0]) + sorted([i for i in lst if i < 0])
def arrayme():
np.random.shuffle(ar)
return np.concatenate([np.sort(ar[ar >= 0]), np.sort(ar[ar < 0])], axis=0)
def nplexicon():
np.random.shuffle(ar)
return ar[np.lexsort((ar, ar < 0))]
def numpybisect():
np.random.shuffle(ar)
ar.sort()
i = ar.__abs__().argmin()
return np.concatenate((ar[i:], ar[:i]))
nloops = 1000
lst = list(range(-10**1, 0, 1)) + list(range(10**1, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison small")
cost_shuffle_list_small = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_small)
cost_shuffle_array_small = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_small)
lst = list(range(-10**4, 0, 1)) + list(range(10**4, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison medium")
cost_shuffle_list_medium = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_medium)
cost_shuffle_array_medium = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_medium)
nloops = 100
lst = list(range(-10**6, 0, 1)) + list(range(10**6, -1, -1))
ar = np.array(lst)
print("> Shuffle cost comparison large")
cost_shuffle_list_large = timeit(shuffle_lst, number=nloops)
print("shuffle_lst:", cost_shuffle_list_large)
cost_shuffle_array_large = timeit(shuffle_ar, number=nloops)
print("shuffle_ar:", cost_shuffle_array_large)
print()
nloops = 1000
## With small lists/arrays
lst = list(range(-10**1, 0, 1)) + list(range(10**1, -1, -1))
ar = np.array(lst)
print("tttttw/o pen.ttttw. pen.")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(lexicon, number=nloops)
print("lexicon:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(comprehension, number=nloops)
print("compreh.:t", foo, "t", foo - cost_shuffle_list_small)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_small)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_small)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_small)
print()
## With medium lists/arrays
lst = list(range(-10**4, 0, 1)) + list(range(10**4, -1, -1))
ar = np.array(lst)
print("tttttw/o costttttw. cost")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(lexicon, number=nloops)
print("lexicon:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(comprehension, number=nloops)
print("compreh.:t", foo, "t", foo - cost_shuffle_list_medium)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_medium)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_medium)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_medium)
print()
## With large lists/arrays
nloops = 100
lst = list(range(-10**6, 0, 1)) + list(range(10**6, -1, -1))
ar = np.array(lst)
print("tttttw/o costttttw. cost")
foo = timeit(bisectme, number=nloops)
print("bisectme:t", foo, "t", foo - cost_shuffle_list_large)
foo = timeit(arrayme, number=nloops)
print("arrayme:t", foo, "t", foo - cost_shuffle_array_large)
foo = timeit(nplexicon, number=nloops)
print("nplexicon:t", foo, "t", foo - cost_shuffle_array_large)
foo = timeit(numpybisect, number=nloops)
print("npbisect:t", foo, "t", foo - cost_shuffle_array_large)
print()
import numpy as np
l = np.array([1, -2, 10, -12, -4, -5, 9, 2])
l[np.lexsort((l, l < 0))]
array([ 1, 2, 9, 10, -12, -5, -4, -2])
You could just sort by the negative of the element’s inverse:
from __future__ import division
sorted(lst, key=lambda i: 0 if i == 0 else -1 / i)
Taking the inverse switches the order of the magnitudes (larger numbers in the middle, smaller on the outside). Taking the negative reverses the order (positives first, negatives last).
Be aware of the size of your numbers of course and if they’ll cause any over- or underflow issues.
Kudos to wmin for the logic of his solution (accepted answer) which is great. So for completeness, a similar answer to this but based on numpy, which is significantly faster for anything else other than small lists/arrays, is as follows:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
ar = np.array(lst)
ar.sort()
i = ar.__abs__().argmin()
np.concatenate((ar[i:], ar[:i]))
I don’t know if it’s the most Pythonic, and it certainly does not have any bells-and-whistles, but IMO it’s a clear and understandable code:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
pos = list()
neg = list()
for i in lst:
neg.append(i) if i < 0 else pos.append(i)
print(sorted(pos) + sorted(neg))
Sort the list in place twice like so:
lst = [1, -2, 10, -12, -4, -5, 9, 2]
lst.sort()
lst.sort(key=int(0).__gt__) # key is True for items <= 0
This takes advantage of the fact the python sort function/method is stable. That means that items with the same value or key stay in the same order. The first sort puts all the items in order from smallest to largest. For the second sort, all the items < 0 get a key of True, all the items >= 0 get a key of False. Because True (1) > False (0), the second sort moves all the negative items to the end, without changing the order of the negative items.
*Here is another solution: *
lst = [1, -2, 10, -12, -4, -5, 9, 2] # list of values.
x = sorted(lst) # x : [-12, -5, -4, -2, 1, 2, 9, 10]
k, m = [], [] # k : [1, 2, 9, 10] # M : [-12, -5, -4, -2]
for i in x:
if i > 0:
k.append(i)
else:
m.append(i)
w = k + m # w : [1, 2, 9, 10, -12, -5, -4, -2]
print(w)