当name = True for Python 3时，numpy genfromtxt似乎不起作用

Question

I am using the Google Colab enviroment.

The file I am using can be found here. It is a csv file

https://drive.google.com/open?id=1v7Mm6S8BVtou1iIfobY43LRF8MgGdjfU

Warning: it has several million rows.

This code runs within a minute in Google Colab Python 3 notebook. I tried this several times with no problem.

from numpy import genfromtxt
my_data = genfromtxt('DlRefinedRatings.csv', delimiter=',' ,  dtype=int)

print(my_data[0:50])

The code below, on the other hand, runs for several minutes before disconnecting from Google Colab's server. I tried multiple times. Eventually colab gives me a 'running out of memory' warning.

from numpy import genfromtxt
my_data = genfromtxt('DlRefinedRatings.csv', delimiter=',' ,  dtype=int,  names=True)

print(my_data[0:50])

It seems that there used to be an issue with names=True in Python 3 but that issue was fixed https://github.com/numpy/numpy/issues/5411

I check which version I was using in Colab and it was up to date

import numpy as np

print(np.version.version)

>1.14.3

Answer 1

With

my_data = genfromtxt('DlRefinedRatings.csv', delimiter=',' ,  dtype=int, max_rows=100)

I got a (100,4) int array.

With names=True it took long, and then issued an long list of errors, all the same except for line number (even with the max_rows):

Line #4121986 (got 4 columns instead of 3)

The header line is screwy - with an initial blank name:

In [753]: !head ../Downloads/refinedRatings.csv
,user_id,book_id,rating
0,1,258,5
1,2,4081,4
2,2,260,5
3,2,9296,5
5,2,26,4
7,2,33,4
8,2,301,5
9,2,2686,5
10,2,3753,5

So based on names it thinks there are 3 columns, but all data lines have 4. Hence the error. I don't know why it ignores the max_rows in this case.

But with my own names

In [755]: np.genfromtxt('../Downloads/refinedRatings.csv',delimiter=',',dtype=in
     ...: t, max_rows=10, names='foo,bar,dat,me')
Out[755]: 
array([(-1, -1,   -1, -1), ( 0,  1,  258,  5), ( 1,  2, 4081,  4),
       ( 2,  2,  260,  5), ( 3,  2, 9296,  5), ( 5,  2,   26,  4),
       ( 7,  2,   33,  4), ( 8,  2,  301,  5), ( 9,  2, 2686,  5),
       (10,  2, 3753,  5)],
      dtype=[('foo', '<i8'), ('bar', '<i8'), ('dat', '<i8'), ('me', '<i8')])

The first record (-1,-1,-1,-1) is the initial bad header line, with -1 inplace of strings it couldn't turn into ints. So we should skip_header as done below.

alternatively:

In [756]: np.genfromtxt('../Downloads/refinedRatings.csv',delimiter=',',dtype=in
     ...: t, max_rows=10, skip_header=1)
Out[756]: 
array([[   0,    1,  258,    5],
       [   1,    2, 4081,    4],
       [   2,    2,  260,    5],
       [   3,    2, 9296,    5],
       [   5,    2,   26,    4],
       [   7,    2,   33,    4],
       [   8,    2,  301,    5],
       [   9,    2, 2686,    5],
       [  10,    2, 3753,    5],
       [  11,    2, 8519,    5]])

In sum, skip the header, and use your own names if you want a structured array.