How to use the method loadtxt() of numpy neatly?

[Peter Otten]

chao dong wrote:

>     HI, everybody. When I try to use numpy to deal with my dataset in the
>     style of csv, I face a little problem.
>     
>     In my dataset of the csv file, some columns are string that can not
>     convert to float easily. Some of them can ignore, but other columns I
>     need to change the data to a enum style.
> 
>     for example, one column just contain three kinds : S,Q,C. Each of them
>     can declare one meaning, so I must convert them to a dict just like
>     {1,2,3}
>     
>     Now the question is, when I use numpy.loadtxt, I must do all things
>     above in just one line and one fuction. So as a new user in numpy, I
>     don't know how to solve it.
> 
>     Thank you.

Here's a standalone demo:

import numpy

_lookup={"A": 1, "B": 2}
def convert(x):
    return _lookup.get(x, -1)

converters = {
    0: convert, # in column 0 convert "A" --> 1, "B" --> 2, 
                # anything else to -1
}


if __name__ == "__main__":
    # generate csv
    with open("tmp_sample.csv", "wb") as f:
        f.write("""\
A,1,this,67.8
B,2,should,56.7
C,3,be,34.5
A,4,skipped,12.3
""")

    # load csv
    a = numpy.loadtxt(
        "tmp_sample.csv",
        converters=converters,
        delimiter=",",
        usecols=(0, 1, 3) # skip third column
        )
    print a

Does that help?