Converting list of lists to numpy array with multiple data types

Question

I have a list of lists I\'ve read from a file. Each of the inner lists is six elements in length, and has 3 strings and 5 floats. How do I convert this list of lists into

Answer 1

I had the same problem, but tuples are no solution. So I found (python 3.7.1):

ll = [['one','two',1,1.23],['four','five',4,34.3],['six','seven',4,34.3]]

np.array(ll, dtype = 'object')

result:

array([['one', 'two', 1, 1.23],
   ['four', 'five', 4, 34.3],
   ['six', 'seven', 4, 34.3]], dtype=object)

Answer 2

You want a structured array, one that has a compound dtype:

A sample list of lists:

In [4]: ll = [['one','two',1,1.23],['four','five',4,34.3],['six','seven',4,34.3]]

trying to make a regular array, produces an array of strings:

In [5]: np.array(ll)
Out[5]: 
array([['one', 'two', '1', '1.23'],
       ['four', 'five', '4', '34.3'],
       ['six', 'seven', '4', '34.3']], 
       dtype='|S5')

But if I specify a dtype that contains 2 strings, and int and a float, I get a 1d structured array:

In [8]: np.array([tuple(x) for x in ll],dtype='S5,S5,i,f')
Out[8]: 
array([('one', 'two', 1, 1.2300000190734863),
       ('four', 'five', 4, 34.29999923706055),
       ('six', 'seven', 4, 34.29999923706055)], 
      dtype=[('f0', 'S5'), ('f1', 'S5'), ('f2', '<i4'), ('f3', '<f4')])

Note that I had to convert the inner lists to tuples. That's how a structured array takes its input, and also how it displays it. It helps distinguish the structured 'row' from the uniform 'row' of a regular (2d) array.

This the same sort of structured array that genfromtxt or loadtxt produces when reading from a csv file.

There are other ways of specifying the dtype, and a couple of other ways of loading the data into such an array. But this is a start.

---

Further testing, https://stackoverflow.com/a/47774915/901925, shows that this tuple conversion is not that time consuming. Simply creating the array takes more time.