Subsetting data in Python


Question

I want to use the equivalent of the subset command in R for some Python code I am writing.

Here is my data:

col1    col2    col3    col4    col5
100002  2006    1.1 0.01    6352
100002  2006    1.2 0.84    304518
100002  2006    2   1.52    148219
100002  2007    1.1 0.01    6292
10002   2006    1.1 0.01    5968
10002   2006    1.2 0.25    104318
10002   2007    1.1 0.01    6800
10002   2007    4   2.03    25446
10002   2008    1.1 0.01    6408

I want to subset the data based on contents of col1 and col2. (The unique values in col1 are 100002 and 10002, and in col2 are 2006,2007 and 2008.)

This can be done in R using the subset command, is there anything similar in Python?

1
17
5/11/2017 6:50:23 PM

Accepted Answer

While the iterator-based answers are perfectly fine, if you're working with numpy arrays (as you mention that you are) there are better and faster ways of selecting things:

import numpy as np
data = np.array([
        [100002, 2006, 1.1, 0.01, 6352],
        [100002, 2006, 1.2, 0.84, 304518],
        [100002, 2006, 2,   1.52, 148219],
        [100002, 2007, 1.1, 0.01, 6292],
        [10002,  2006, 1.1, 0.01, 5968],
        [10002,  2006, 1.2, 0.25, 104318],
        [10002,  2007, 1.1, 0.01, 6800],
        [10002,  2007, 4,   2.03, 25446],
        [10002,  2008, 1.1, 0.01, 6408]    ])

subset1 = data[data[:,0] == 100002]
subset2 = data[data[:,0] == 10002]

This yields

subset1:

array([[  1.00002e+05,   2.006e+03,   1.10e+00, 1.00e-02,   6.352e+03],
       [  1.00002e+05,   2.006e+03,   1.20e+00, 8.40e-01,   3.04518e+05],
       [  1.00002e+05,   2.006e+03,   2.00e+00, 1.52e+00,   1.48219e+05],
       [  1.00002e+05,   2.007e+03,   1.10e+00, 1.00e-02,   6.292e+03]])

subset2:

array([[  1.0002e+04,   2.006e+03,   1.10e+00, 1.00e-02,   5.968e+03],
       [  1.0002e+04,   2.006e+03,   1.20e+00, 2.50e-01,   1.04318e+05],
       [  1.0002e+04,   2.007e+03,   1.10e+00, 1.00e-02,   6.800e+03],
       [  1.0002e+04,   2.007e+03,   4.00e+00, 2.03e+00,   2.5446e+04],
       [  1.0002e+04,   2.008e+03,   1.10e+00, 1.00e-02,   6.408e+03]])

If you didn't know the unique values in the first column beforehand, you can use either numpy.unique1d or the builtin function set to find them.

Edit: I just realized that you wanted to select data where you have unique combinations of two columns... In that case, you might do something like this:

col1 = data[:,0]
col2 = data[:,1]

subsets = {}
for val1, val2 in itertools.product(np.unique(col1), np.unique(col2)):
    subset = data[(col1 == val1) & (col2 == val2)]
    if np.any(subset):
        subsets[(val1, val2)] = subset

(I'm storing the subsets as a dict, with the key being a tuple of the combination... There are certainly other (and better, depending on what you're doing) ways to do this!)

20
9/27/2010 7:46:56 PM

subset() in R is pretty much analogous to filter() in Python. As the reference notes, this will be used implicitly by list comprehensions, so the most concise and clear way to write the code might be

[ item for item in items if item.col2 == 2006 ] 

if, for example, your data rows were in an iterable called items.


Licensed under: CC-BY-SA with attribution
Not affiliated with: Stack Overflow
Icon