Fast Data Comparison (dict v. list. v string)

[Scott Brady Drummonds]

Hi, everyone,

I'm a relative novice at Python and am working on some optimizations in my
first Python project. At its core, this tool I'm writing needs to perform
many comparisons to fixed-size lists of fixed-length strings.

Currently, my implementation uses dictionaries to store each string. I
maintain a separate "key-mapping" dictionary that maps keys from one
dictionary to the other. Then, all I have to do to compare the two
dictionaries is as follows:
for metaKey in keyMap.keys():
if dict1[metaKey] != dict2[keyMap[metaKey]]:
# Do some processing

Since the sizes of the dictionaries never change, I tried implementing this
using lists. The solution looks something like this (and assumes that a
pre-processing phase has sorted the contents of each list so their indexes
are the same):
for i in len(list1):
if list1 != list2:
# Do some processing

As it turns out, this implementation appears to be about 33% faster than the
dictionary-based one. Now, assuming that the datum being stored at each
index can fit into one character, I could do a string-based implementation
like this:
for i in len(string1):
if string1 != string:
# Do some processing

This last solution actually runs about the same as the dictionary, which
takes 50% longer than the list implementation.

Now, my questions are:
1) Does anyone have another suggestion as to how I can organize these data
so that I can compare many elements many times?
2) Is there a string comparison operator that will return which indexes
have different values? Maybe it would be faster than the iterative
comparison approach for the third implementation.
3) Since my data are changing between the successive executions of the code
snippets above, I need a way of having the other parts of the program update
it. But, it appears that strings are constant as I can't assign individual
characters with "string1 = '0'". Is there any way around this?

Thanks!
Scott

 

[Larry Bates]

Scott,

You should take a close look at list comprehensions
and possibly sets (new in V2.3). I'm not all that
familiar with sets, but your problem sounds like
they might come in handy.

Example list comprehension:

list1notinlist2=[x for x in list1 if x not in list2]
for item in list1notinlist:
# Do some processing

Note: In Python if you find yourself using
indexes into lists, there is "almost" always a
better way (at least that is my experience).

Larry Bates
Syscon, Inc.

 

[Scott Brady Drummonds]

Scott,

You should take a close look at list comprehensions
and possibly sets (new in V2.3). I'm not all that
familiar with sets, but your problem sounds like
they might come in handy.

Excellent suggestion! Thanks so much for the idea.

Scott

 

[george young]

Hi, everyone,

I'm a relative novice at Python and am working on some optimizations in my
first Python project. At its core, this tool I'm writing needs to perform
many comparisons to fixed-size lists of fixed-length strings.

Currently, my implementation uses dictionaries to store each string. I
maintain a separate "key-mapping" dictionary that maps keys from one
dictionary to the other. Then, all I have to do to compare the two
dictionaries is as follows:
for metaKey in keyMap.keys():
if dict1[metaKey] != dict2[keyMap[metaKey]]:
# Do some processing

Since the sizes of the dictionaries never change, I tried implementing this
using lists. The solution looks something like this (and assumes that a
pre-processing phase has sorted the contents of each list so their indexes
are the same):
for i in len(list1):
if list1 != list2:
# Do some processing

As it turns out, this implementation appears to be about 33% faster than the
dictionary-based one. Now, assuming that the datum being stored at each
index can fit into one character, I could do a string-based implementation
like this:
for i in len(string1):
if string1 != string:
# Do some processing

This last solution actually runs about the same as the dictionary, which
takes 50% longer than the list implementation.

Now, my questions are:
1) Does anyone have another suggestion as to how I can organize these data
so that I can compare many elements many times?
2) Is there a string comparison operator that will return which indexes
have different values? Maybe it would be faster than the iterative
comparison approach for the third implementation.
3) Since my data are changing between the successive executions of the code
snippets above, I need a way of having the other parts of the program update
it. But, it appears that strings are constant as I can't assign individual
characters with "string1 = '0'". Is there any way around this?

You might also take a look at psyco: http://psyco.sourceforge.net. It is
very easy to use, non-intrusive, and claims large performance improvements for
code like this. I have hopes that things like psyco can let us keep code
in the simplest and *clearest* form, and let external optimization mechanisms
make it fast enough when needed.

-- George Young