"groupby" is brilliant!

[Frank Millman]

Hi all

This is probably old hat to most of you, but for me it was a
revelation, so I thought I would share it in case someone has a similar
requirement.

I had to convert an old program that does a traditional pass through a
sorted data file, breaking on a change of certain fields, processing
each row, accumulating various totals, and doing additional processing
at each break. I am not using a database for this one, as the file
sizes are not large - a few thousand rows at most. I am using csv
files, and using the csv module so that each row is nicely formatted
into a list.

The traditional approach is quite fiddly, saving the values of the
various break fields, comparing the values on each row with the saved
values, and taking action if the values differ. The more break fields
there are, the fiddlier it gets.

I was going to do the same in python, but then I vaguely remembered
reading about 'groupby'. It took a little while to figure it out, but
once I had cracked it, it transformed the task into one of utter
simplicity.

Here is an example. Imagine a transaction file sorted by branch,
account number, and date, and you want to break on all three.

-----------------------------
import csv
from itertools import groupby
from operator import itemgetter

BRN = 0
ACC = 1
DATE = 2

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

for brn,brnList in groupby(rows,it emgetter(BRN)):
for acc,accList in groupby(brnList ,itemgetter(ACC )):
for date,dateList in groupby(accList ,itemgetter(DAT E)):
for row in dateList:
[do something with row]
[do something on change of date]
[do something on change of acc]
[do something on change of brn]
-----------------------------

Hope someone finds this of interest.

Frank Millman

[vpr]

Hi Frank

This is one of the reasons why I love Python, you can write readable
code.
I strive to write clean code but I find that exception handling code
e.g. try:
makes my code ugly and significantly harder to read. Does anyone have
any good
pointers for a former C++ / Perl coder.

/vpr
Frank Millman wrote:

Hi all

This is probably old hat to most of you, but for me it was a
revelation, so I thought I would share it in case someone has a similar
requirement.

I had to convert an old program that does a traditional pass through a
sorted data file, breaking on a change of certain fields, processing
each row, accumulating various totals, and doing additional processing
at each break. I am not using a database for this one, as the file
sizes are not large - a few thousand rows at most. I am using csv
files, and using the csv module so that each row is nicely formatted
into a list.

The traditional approach is quite fiddly, saving the values of the
various break fields, comparing the values on each row with the saved
values, and taking action if the values differ. The more break fields
there are, the fiddlier it gets.

I was going to do the same in python, but then I vaguely remembered
reading about 'groupby'. It took a little while to figure it out, but
once I had cracked it, it transformed the task into one of utter
simplicity.

Here is an example. Imagine a transaction file sorted by branch,
account number, and date, and you want to break on all three.

-----------------------------
import csv
from itertools import groupby
from operator import itemgetter

BRN = 0
ACC = 1
DATE = 2

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

for brn,brnList in groupby(rows,it emgetter(BRN)):
for acc,accList in groupby(brnList ,itemgetter(ACC )):
for date,dateList in groupby(accList ,itemgetter(DAT E)):
for row in dateList:
[do something with row]
[do something on change of date]
[do something on change of acc]
[do something on change of brn]
-----------------------------

Hope someone finds this of interest.

Frank Millman

[Paul McGuire]

>

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

This is untested, but you might think about converting your explicit "for...
append" loop into either a list comp,

rows = [row for row in reader]

or just a plain list constructor:

rows = list(reader)

Neh?

-- Paul
(Oh, and I like groupby too! Combine it with sort to quickly create
histograms.)

# tally a histogram of a list of values from 1-10
dataValueRange = range(1,11)
data = [random.choice(d ataValueRange) for i in xrange(10000)]

hist = [ (k,len(list(g)) ) for k,g in itertools.group by(sorted(data) ) ]
print hist

histAsDict = dict((k,len(lis t(g))) for k,g in
itertools.group by(sorted(data) ))
print histAsDict

Gives:

[(1, 979), (2, 1034), (3, 985), (4, 969), (5, 1020), (6, 975), (7, 981), (8,
1070), (9, 1003), (10, 984)]
{1: 979, 2: 1034, 3: 985, 4: 969, 5: 1020, 6: 975, 7: 981, 8: 1070, 9: 1003,
10: 984}

[Frank Millman]

Paul McGuire wrote:

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

This is untested, but you might think about converting your explicit "for...
append" loop into either a list comp,

rows = [row for row in reader]

or just a plain list constructor:

rows = list(reader)

Neh?

-- Paul

Yup, they both work fine.

There may be times when you want to massage the data before appending
it, in which case you obviously have to do it the long way. Otherwise
these are definitely neater, the last one especially.

You could even do it as a one-liner -
rows = list(csv.reader (open('trans.cs v', 'rb')))

It still looks perfectly readable to me.

Thanks

Frank

[geskerrett]

Frank;
I would just like to thank-you for this timely post.
I am working on a reporting project that needed "groupby" functionality
and I was going to sit down this morning to rework some "very ugly
code" into some "not quite so ugly code".

Your post got me pointed to in the "right" direction and the end
results will be much more flexible and ALOT more maintainable.

Thanks.

[Benji York]

Frank Millman wrote:

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

Why do you create a list of rows instead of just iterating over the
reader directly?
--
Benji York

[James Stroud]

Frank Millman wrote:

Hi all

This is probably old hat to most of you, but for me it was a
revelation, so I thought I would share it in case someone has a similar
requirement.

I had to convert an old program that does a traditional pass through a
sorted data file, breaking on a change of certain fields, processing
each row, accumulating various totals, and doing additional processing
at each break. I am not using a database for this one, as the file
sizes are not large - a few thousand rows at most. I am using csv
files, and using the csv module so that each row is nicely formatted
into a list.

The traditional approach is quite fiddly, saving the values of the
various break fields, comparing the values on each row with the saved
values, and taking action if the values differ. The more break fields
there are, the fiddlier it gets.

I was going to do the same in python, but then I vaguely remembered
reading about 'groupby'. It took a little while to figure it out, but
once I had cracked it, it transformed the task into one of utter
simplicity.

Here is an example. Imagine a transaction file sorted by branch,
account number, and date, and you want to break on all three.

-----------------------------
import csv
from itertools import groupby
from operator import itemgetter

BRN = 0
ACC = 1
DATE = 2

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

for brn,brnList in groupby(rows,it emgetter(BRN)):
for acc,accList in groupby(brnList ,itemgetter(ACC )):
for date,dateList in groupby(accList ,itemgetter(DAT E)):
for row in dateList:
[do something with row]
[do something on change of date]
[do something on change of acc]
[do something on change of brn]
-----------------------------

Hope someone finds this of interest.

Frank Millman

I'm sure I'm going to get a lot of flac on this list for proposing to
turn nested for-loops into a recursive function, but I couldn't help
myself. This seems more simple to me, but for others it may be difficult
to look at, and these people will undoubtedly complain.
import csv
from itertools import groupby
from operator import itemgetter

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

def brn_doer(row):
[doing something with brn here]

def acc_doer(date):
[you get the idea]

[etc.]

doers = [brn_doer, acc_doer, date_doer, row_doer]

def doit(rows, doers, i=0):
for r, alist in groupby(rows, itemgetter(i)):
doit(alist, doers[1:], i+1)
doers[0](r)

doit(rows, doers, 0)

Now all of those ugly for loops become one recursive function. Bear in
mind, its not all that 'elegant', but it looks nicer, is more succinct,
abstracts the process, and scales to arbitrary depth. Tragically,
however, it has been generalized, which is likely to raise some hackles
here. And, oh yes, it didn't answer exactly your question (which you
didn't really have). I'm sure I will regret this becuase, as you will
find, suggesting code on this list with additional utility is somewhat
discouraged by the vociferous few who make a religion out of 'import this'.

Also, I still have no idea what 'groupby' does. It looks interesting
thgough, thanks for pointing it out.

James

--
James Stroud
UCLA-DOE Institute for Genomics and Proteomics
Box 951570
Los Angeles, CA 90095

http://www.jamesstroud.com/

[James Stroud]

James Stroud wrote:

Frank Millman wrote:

Hi all

This is probably old hat to most of you, but for me it was a
revelation, so I thought I would share it in case someone has a similar
requirement.

I had to convert an old program that does a traditional pass through a
sorted data file, breaking on a change of certain fields, processing
each row, accumulating various totals, and doing additional processing
at each break. I am not using a database for this one, as the file
sizes are not large - a few thousand rows at most. I am using csv
files, and using the csv module so that each row is nicely formatted
into a list.

The traditional approach is quite fiddly, saving the values of the
various break fields, comparing the values on each row with the saved
values, and taking action if the values differ. The more break fields
there are, the fiddlier it gets.

I was going to do the same in python, but then I vaguely remembered
reading about 'groupby'. It took a little while to figure it out, but
once I had cracked it, it transformed the task into one of utter
simplicity.

Here is an example. Imagine a transaction file sorted by branch,
account number, and date, and you want to break on all three.

-----------------------------
import csv
from itertools import groupby
from operator import itemgetter

BRN = 0
ACC = 1
DATE = 2

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

for brn,brnList in groupby(rows,it emgetter(BRN)):
for acc,accList in groupby(brnList ,itemgetter(ACC )):
for date,dateList in groupby(accList ,itemgetter(DAT E)):
for row in dateList:
[do something with row]
[do something on change of date]
[do something on change of acc]
[do something on change of brn]
-----------------------------

Hope someone finds this of interest.

Frank Millman

I'm sure I'm going to get a lot of flac on this list for proposing to
turn nested for-loops into a recursive function, but I couldn't help
myself. This seems more simple to me, but for others it may be difficult
to look at, and these people will undoubtedly complain.
import csv
from itertools import groupby
from operator import itemgetter

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

def brn_doer(row):
[doing something with brn here]

def acc_doer(date):
[you get the idea]

[etc.]

doers = [brn_doer, acc_doer, date_doer, row_doer]

def doit(rows, doers, i=0):
for r, alist in groupby(rows, itemgetter(i)):
doit(alist, doers[1:], i+1)
doers[0](r)

doit(rows, doers, 0)

Now all of those ugly for loops become one recursive function. Bear in
mind, its not all that 'elegant', but it looks nicer, is more succinct,
abstracts the process, and scales to arbitrary depth. Tragically,
however, it has been generalized, which is likely to raise some hackles
here. And, oh yes, it didn't answer exactly your question (which you
didn't really have). I'm sure I will regret this becuase, as you will
find, suggesting code on this list with additional utility is somewhat
discouraged by the vociferous few who make a religion out of 'import this'.

Also, I still have no idea what 'groupby' does. It looks interesting
thgough, thanks for pointing it out.

James

Forgot to test for stopping condition:
def doit(rows, doers, i=0):
for r, alist in groupby(rows, itemgetter(i)):
if len(doers) > 1:
doit(alist, doers[1:], i+1)
doers[0](r)

--
James Stroud
UCLA-DOE Institute for Genomics and Proteomics
Box 951570
Los Angeles, CA 90095

http://www.jamesstroud.com/

[Jon Clements]

Not related to itertools.group by, but the csv.reader object...

If for some reason you have malformed CSV files, with embedded newlines
or something of that effect, it will raise an exception. To skip those,
you will need a construct of something like this:

raw_csv_in = file('filenameh ere.csv')
for raw_line in raw_csv_in:
try:
# Do something to rawline here maybe if necessary to "clean it
up"
row = csv.reader( [raw_line] ).next()
# Do your stuff here
except csv.Error:
pass # or do something more appropriate if the record is
important

May not be applicable in your case, but has stung me a few times...

All the best,

Jon.
Frank Millman wrote:

Paul McGuire wrote:

reader = csv.reader(open ('trans.csv', 'rb'))
rows = []
for row in reader:
rows.append(row )

This is untested, but you might think about converting your explicit "for...
append" loop into either a list comp,

rows = [row for row in reader]

or just a plain list constructor:

rows = list(reader)

Neh?

-- Paul

Yup, they both work fine.

There may be times when you want to massage the data before appending
it, in which case you obviously have to do it the long way. Otherwise
these are definitely neater, the last one especially.

You could even do it as a one-liner -
rows = list(csv.reader (open('trans.cs v', 'rb')))

It still looks perfectly readable to me.

Thanks

Frank

[John Machin]

On 13/06/2006 6:28 PM, Paul McGuire wrote:

(Oh, and I like groupby too! Combine it with sort to quickly create
histograms.)

# tally a histogram of a list of values from 1-10
dataValueRange = range(1,11)
data = [random.choice(d ataValueRange) for i in xrange(10000)]

hist = [ (k,len(list(g)) ) for k,g in itertools.group by(sorted(data) ) ]

That len(list(g)) looks like it uses O(N) memory just to find out what N
is :-(

The best I could come up with is sum(itertools.i map(lambda x: 1, g)) --
but that does look a bit ugly ...