Filling missing values using numpy.genfromtxt

问题:

Despite the advice from the previous questions:

-9999 as missing value with numpy.genfromtxt()

Using genfromtxt to import csv data with missing values in numpy

I still am unable to process a text file that ends with a missing value,

a.txt:

1 2 3
4 5 6
7 8

I've tried multiple arrangements of options of missing_values, filling_values and can not get this to work:

import numpy as np

sol = np.genfromtxt("a.txt", 
                    dtype=float,
                    invalid_raise=False, 
                    missing_values=None,
                    usemask=True,
                    filling_values=0.0)
print sol

What I would like to get is:

[[1.0 2.0 3.0]
 [4.0 5.0 6.0]
 [7.0 8.0 0.0]]

but instead I get:

/usr/local/lib/python2.7/dist-packages/numpy/lib/npyio.py:1641: ConversionWarning: Some errors were detected !
    Line #3 (got 2 columns instead of 3)
  warnings.warn(errmsg, ConversionWarning)
[[1.0 2.0 3.0]
 [4.0 5.0 6.0]]

回答1:

The issue is that numpy doesn't like ragged arrays. Since there is no character in the third position of the last row of the file, so genfromtxt doesn't even know it's something to parse, let alone what to do with it. If the missing value had a filler (any filler) such as:

1 2 3
4 5 6
7 8 ''

Then you'd be able to:

sol = np.genfromtxt("a.txt",
                dtype=float,
                invalid_raise=False,
                missing_values='',
                usemask=False,
                filling_values=0.0)

and: sol

array([[  1.,   2.,   3.],
       [  4.,   5.,   6.],
       [  7.,   8.,  nan]])

Unfortunately, if making the columns of the file uniform isn't an option, you might be stuck with line-by-line parsing.

One other possibility would be IF all the "short" rows are at the end... in which case you might be able to utilize the 'usecols' flag to parse all columns that are uniform, and then the skip_footer flag to do the same for the remaining columns while skipping those that aren't available:

sol = np.genfromtxt("a.txt",
                dtype=float,
                invalid_raise=False,
                usemask=False,
                filling_values=0.0,
                usecols=(0,1))
sol
array([[ 1.,  2.],
   [ 4.,  5.],
   [ 7.,  8.]])

sol2 = np.genfromtxt("a.txt",
                dtype=float,
                invalid_raise=False,
                usemask=False,
                filling_values=0.0,
                usecols=(2,),
                skip_footer=1)
sol2
array([ 3.,  6.])

And then combine the arrays from there adding the fill value:

sol2=np.append(sol2, 0.0)
sol2=sol2.reshape(3,1)
sol=np.hstack([sol,sol2])
sol
array([[ 1.,  2.,  3.],
   [ 4.,  5.,  6.],
   [ 7.,  8.,  0.]])

回答2:

Using pandas:

import pandas as pd

df = pd.read_table('data', sep='\s+', header=None)
df.fillna(0, inplace=True)
print(df)
#    0  1  2
# 0  1  2  3
# 1  4  5  6
# 2  7  8  0

pandas.read_table replaces missing data with NaNs. You can replace those NaNs with some other value using df.fillna.

df is a pandas.DataFrame. You can access the underlying NumPy array with df.values:

print(df.values)
# [[ 1.  2.  3.]
#  [ 4.  5.  6.]
#  [ 7.  8.  0.]]

回答3:

In my experience the best is to just parse manually, this function works for me, it might be slow but generally fast enough.

def manual_parsing(filename,delim,dtype):
    out = list()
    lengths = list()
    with open(filename,'r') as ins:
        for line in ins:
            l = line.split(delim)
            out.append(l)
            lengths.append(len(l))
    lim = np.max(lengths)
    for l in out:
        while len(l)<lim:
            l.append("nan")
    return np.array(out,dtype=dtype)