Best way to handle large lists?

[Chaz Ginger]

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than

[ i for i in bigList if i not in smallList ]

Thanks.
Chaz

[Duncan Booth]

Re: Best way to handle large lists?

Chaz Ginger <cginboston@hot mail.comwrote:

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than
>
[ i for i in bigList if i not in smallList ]

How about:

smallSet = set(smallList)
something = [ i for i in bigList if i not in smallSet ]

Use timeit.py on some representative data to see what difference that
makes.

[Paul Rubin]

Re: Best way to handle large lists?

Chaz Ginger <cginboston@hot mail.comwrites:

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than
>
[ i for i in bigList if i not in smallList ]

diff = list(set(bigLis t) - set(smallList))

Note that doesn't get you the elements in any particular order.

[durumdara]

Re: Best way to handle large lists?

Chaz Ginger írta:

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than
>
[ i for i in bigList if i not in smallList ]
>
Thanks.
Chaz
>

Hi !

If you have big list, you can use dbm like databases.
They are very quick. BSDDB, flashdb, etc. See SleepyCat, or see python help.

In is very slow in large datasets, but bsddb is use hash values, so it
is very quick.
The SleepyCat database have many extras, you can set the cache size and
many other parameters.

Or if you don't like dbm style databases, you can use SQLite. Also
quick, you can use SQL commands.
A little slower than bsddb, but it is like SQL server. You can improve
the speed with special parameters.

dd

[Bill Williams]

Re: Best way to handle large lists?

I don't know enough about Python internals, but the suggested solutions
all seem to involve scanning bigList. Can this presumably linear
operation be avoided by using dict or similar to find all occurrences of
smallist items in biglist and then deleting those occurrences?

Bill Williams



In article <prrUg.1662$We. 477@trndny08>,
Chaz Ginger <cginboston@hot mail.comwrote:

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than
>
[ i for i in bigList if i not in smallList ]
>
Thanks.
Chaz

[Hari Sekhon]

Re: Best way to handle large lists?

I don't know much about the python internals either, so this may be the
blind leading the blind, but aren't dicts much slower to work with than
lists and therefore wouldn't your suggestion to use dicts be much
slower? I think it's something to do with the comparative overhead of
using keys in dicts rather than using positional indexes in lists/arrays...

At least that is what I thought.

Can anyone confirm this?

-h

Hari Sekhon



Bill Williams wrote:

I don't know enough about Python internals, but the suggested solutions
all seem to involve scanning bigList. Can this presumably linear
operation be avoided by using dict or similar to find all occurrences of
smallist items in biglist and then deleting those occurrences?
>
Bill Williams
>
>
>
In article <prrUg.1662$We. 477@trndny08>,
Chaz Ginger <cginboston@hot mail.comwrote:
>
>

>I have a system that has a few lists that are very large (thousands or
>tens of thousands of entries) and some that are rather small. Many times
>I have to produce the difference between a large list and a small one,
>without destroying the integrity of either list. I was wondering if
>anyone has any recommendations on how to do this and keep performance
>high? Is there a better way than
>>
>[ i for i in bigList if i not in smallList ]
>>
>Thanks.
>Chaz
>>

[Sybren Stuvel]

Re: Best way to handle large lists?

Bill Williams enlightened us with:

I don't know enough about Python internals, but the suggested
solutions all seem to involve scanning bigList. Can this presumably
linear operation be avoided by using dict or similar to find all
occurrences of smallist items in biglist and then deleting those
occurrences?

And how would that beat O(n)? Every element of bigList has to be
scanned at one point, either to compare it to every earlier element in
bigList and eliminate it, or to compare it to every element in
smallList.

Run benchmarks on the suggestions, and see which is fastest for
yourself.

Sybren
--
Sybren Stüvel
Stüvel IT - http://www.stuvel.eu/

[Chaz Ginger]

Re: Best way to handle large lists?

I've done that and decided that Python's 'list comprehension' isn't a
way to go. I was hoping that perhaps someone had some experience with
some C or C++ library that has a Python interface that would make a
difference.

Chaz

Sybren Stuvel wrote:

Bill Williams enlightened us with:

>I don't know enough about Python internals, but the suggested
>solutions all seem to involve scanning bigList. Can this presumably
>linear operation be avoided by using dict or similar to find all
>occurrences of smallist items in biglist and then deleting those
>occurrences?

>
And how would that beat O(n)? Every element of bigList has to be
scanned at one point, either to compare it to every earlier element in
bigList and eliminate it, or to compare it to every element in
smallList.
>
Run benchmarks on the suggestions, and see which is fastest for
yourself.
>
Sybren

[Paul Rubin]

Re: Best way to handle large lists?

Sybren Stuvel <sybrenUSE@YOUR thirdtower.com. imaginationwrit es:

I don't know enough about Python internals, but the suggested
solutions all seem to involve scanning bigList. Can this presumably
linear operation be avoided by using dict or similar to find all
occurrences of smallist items in biglist and then deleting those
occurrences?

>
And how would that beat O(n)? Every element of bigList has to be
scanned at one point, either to compare it to every earlier element in
bigList and eliminate it, or to compare it to every element in
smallList.

Maybe the application should use sets instead of lists for these
collections.

[Chaz Ginger]

Re: Best way to handle large lists?

Paul Rubin wrote:

Sybren Stuvel <sybrenUSE@YOUR thirdtower.com. imaginationwrit es:

>>I don't know enough about Python internals, but the suggested
>>solutions all seem to involve scanning bigList. Can this presumably
>>linear operation be avoided by using dict or similar to find all
>>occurrences of smallist items in biglist and then deleting those
>>occurrences ?

>And how would that beat O(n)? Every element of bigList has to be
>scanned at one point, either to compare it to every earlier element in
>bigList and eliminate it, or to compare it to every element in
>smallList.

>
Maybe the application should use sets instead of lists for these
collections.

What would sets do for me over lists?

Chaz

[Larry Bates]

Re: Best way to handle large lists?

Chaz Ginger wrote:

I have a system that has a few lists that are very large (thousands or
tens of thousands of entries) and some that are rather small. Many times
I have to produce the difference between a large list and a small one,
without destroying the integrity of either list. I was wondering if
anyone has any recommendations on how to do this and keep performance
high? Is there a better way than
>
[ i for i in bigList if i not in smallList ]
>
Thanks.
Chaz

IMHO the only way to speed things up is to know more about the
actual data in the lists (e.g are the elements unique, can they
be sorted, etc) and take advantage of all that information to
come up with a "faster" algorithm. If they are unique, sets
might be a good choice. If they are sorted, bisect module
might help. The specifics about the list(s) may yield a faster
method.

-Larry

[Jeremy Sanders]

Re: Best way to handle large lists?

Chaz Ginger wrote:

What would sets do for me over lists?

It's faster to tell whether something is in a set or dict than in a list
(for some minimum size).

Jeremy

--
Jeremy Sanders

http://www.jeremysanders.net/

[Chaz Ginger]

Re: Best way to handle large lists?

Larry Bates wrote:

Chaz Ginger wrote:

>I have a system that has a few lists that are very large (thousands or
>tens of thousands of entries) and some that are rather small. Many times
>I have to produce the difference between a large list and a small one,
>without destroying the integrity of either list. I was wondering if
>anyone has any recommendations on how to do this and keep performance
>high? Is there a better way than
>>
>[ i for i in bigList if i not in smallList ]
>>
>Thanks.
>Chaz

>
>
IMHO the only way to speed things up is to know more about the
actual data in the lists (e.g are the elements unique, can they
be sorted, etc) and take advantage of all that information to
come up with a "faster" algorithm. If they are unique, sets
might be a good choice. If they are sorted, bisect module
might help. The specifics about the list(s) may yield a faster
method.
>
-Larry

Each item in the list is a fully qualified domain name, e.g.
foo.bar.com. The order in the list has no importance. That is about all
there is to the list other than to say the number of items in a list can
top out about 10,000.

Chaz

[Richard Brodie]

Re: Best way to handle large lists?


"Chaz Ginger" <cginboston@hot mail.comwrote in message
news:45227BA1.6 080107@hotmail. com...

Each item in the list is a fully qualified domain name, e.g.
foo.bar.com. The order in the list has no importance.

So you don't actually need to use lists at all, then.
You can just use sets and write:

newSet = bigSet - littleSet

[Jeremy Sanders]

Re: Best way to handle large lists?

Jeremy Sanders wrote:

Chaz Ginger wrote:
>

>What would sets do for me over lists?

>
It's faster to tell whether something is in a set or dict than in a list
(for some minimum size).

As a footnote, this program

import random
num = 100000

a = set( range(num) )
for i in range(100000):
x = random.randint( 0, num-1) in a

completes in less than a second, whereas

import random
num = 100000

a = range(num)
for i in range(100000):
x = random.randint( 0, num-1) in a

takes a long time on my computer.

Jeremy

--
Jeremy Sanders

http://www.jeremysanders.net/