Feature Request: skipcols in .read_csv

[pylang]

I'd like to read a set of csv files but exclude specific columns. read_csv currently has a usecols keyword, but it requires writing a list of all the columns present. This is a bit tedious and more importantly, not all files have the same columns, so usecols would not work in general cases, whereas a complimentary function would work. Can a skipcols keyword be added to 0.17 that accepts a list of column names and reads all but those columns into a DataFrame? Thanks.

xref #4749
xref #8985
xref #6710

[terrytangyuan]

So read_csv() is defined by calling _make_parser_function() which calls _read(). Any instructions would be appreciated. It's a bit confusing to me. @jreback

[jreback]

parser is a bit complicated. see how usecols is used.

[terrytangyuan]

It looks like the code related to usecols needs to be re-factored before I add things on top of it. I won't be able to re-factor coz I might break a lot of internal things. @jreback

[TomAugspurger]

Is there a spot in the code where when know all the columns before starting to parse the rows? If so you can assign usecols=set(all_cols) - set(skipcols) (would need to fixup the ordering afterwards) and go from there.

[terrytangyuan]

Yeah I did something similar but stopped due to other code related to
usecols. I'll look into it again. Thanks.
On Sep 2, 2015 8:38 AM, "Tom Augspurger" [email protected] wrote:

Is there a spot in the code where when know all the columns before
starting to parse the rows? If so you can take set usecols=set(all_cols)

• set(skipcols) (would need to fixup the ordering afterwards) and go from
  there.

—
Reply to this email directly or view it on GitHub
#10882 (comment).

[pylang]

Any progress on this addition? Thanks.

[pylang]

ping @jreback

[jreback]

if you submit a PR there will be progress
we have 1700 open issues

[pylang]

many thanks

[pylang]

In a similar vain, is there a way to read in a subset of rows? In other words, is there a counterpart to the skiprows keyword? For example, this feature is desired:

df = pd.read_csv("bigdata.csv")
df
# Output: Millions of rows

selection = [i for i in range(0, 1000000) if i % 2 == 0]
subset = pd.read_csv("bigdata.csv", use_rows=selection)    # skip all rows except those listed
subset
# Output: only even rows for the first million

[gfyoung]

@pylang : We now accept callable for usecols. Does that help to resolve this issue?

[jreback]

sure, maybe an example of doing that in io.rst would be helpful?

[pylang]

@gfyoung I'm not sure what you have in mind. I am interested in selecting rows. An example would be helpful, thank you.

[gfyoung]

@pylang :

1. Your original issue was for skipcols though?
2. skiprows is currently not supported by the C engine. However, we could by all means allow skiprows be a callable like usecols is? How does that sound? Something like:

>>> data = 'a,b,c\n1,2,3\n2,3,4'
>>> read_csv(StringIO(data), skiprows=lambda x: x%2 == 0, engine='python')
a b c
2 3 4

where x is the row number (starting at 0)

[gfyoung]

@jreback : There are examples in the docs to illustrate usecols, but we can also mention that we can use the callable to exclude columns as well. How does that sound?

[jreback]

yes that's what i mean, to show using s callable to skipcols

[pylang]

@gfyoung I think your example for skiprows would suffice. And yes you are correct re: skipcols. A similar callable option to filter usecols with an example in the docs would be sufficient imo.

[gfyoung]

@pylang : #15059 is up to address skiprows. I've hit a roadblock at this point implementing it for the C engine, so any input on that would be appreciated!