0

Following on from this question:

Unexpectedly large array created with numpy.ones when setting names

When I multiply 

a = np.ones([len(sectors),len(columns)])
a[0,:] *= [1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8]

It works fine.

When I try

columns = ["Port Wt", "Bench Wt", "Port Retn", "Bench Retn", 
            "Attrib", "Select", "Inter", "Total"]
a = np.ones((10,), dtype={"names":columns, "formats":["f8"]*len(columns)})
a[0] *= [1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8]

I get the error

TypeError: cannot convert to an int; scalar object is not a number

I would like to use field-names if possible. What am I doing wrong here?

Many thanks.

Improve this question
2
  • 1
    Shouldn't the first parameter for ones, shape, have two elements? You are just creating a 1D array here, and a[0] will be a number. Commented Nov 23, 2014 at 9:23
  • Hi @FerdinandBeyer - using two elements caused unusual array results per my previous question, hence the approach I've taken is based on the response to that question. Commented Nov 23, 2014 at 10:34

2 Answers 2

Reset to default
2

A element (row) of this a can be modified by assigning it a tuple. We can take advantage of the fact that lists easily convert to and from arrays, to write:

In [162]: a = np.ones((10,), dtype={"names":columns, "formats":["f8"]*len(columns)})

In [163]: x=[1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8]

In [164]: a[0]=tuple(np.array(x)*list(a[0]))

In [165]: a
Out[165]: 
array([(1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8),
       ...], dtype=[('Port Wt', '<f8'), ('Bench Wt', '<f8'),...

More generally you could write

a[i] = tuple(foo(list(a[i]))

Multiple values ('rows') of a can be set with a list of tuples.

An earlier SO structure array question (https://stackoverflow.com/a/26183332/901925) suggests another solution - create a partner 2d array that shares the same data buffer.

In [311]: a1 = np.empty((10,8))  # conventional 2d array

In [312]: a1.data = a.data   # share the a's data buffer

In [313]: a1[0] *= x   # do math on a1

In [314]: a1
Out[314]: 
array([[ 1.1,  1.2,  1.3,  1.4,  1.5,  1.6,  1.7,  1.8],
       ...
       [ 1. ,  1. ,  1. ,  1. ,  1. ,  1. ,  1. ,  1. ]])

In [315]: a
Out[315]: 
array([(1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8),
       ...
       (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)], 
      dtype=[('Port Wt', '<f8'), ('Bench Wt', '<f8'), ('Port Retn', '<f8'), ('Bench Retn', '<f8'), ('Attrib', '<f8'), ('Select', '<f8'), ('Inter', '<f8'), ('Total', '<f8')])

By sharing the data buffer, changes made to a1 affect a as well.

It might be better to view 2d a1 as the primary array, and a as a structured view. a could be constructed on the fly, as needed to display the data, access columns by name, or write to a csv file.

Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

1

The rows of your array a are not numpy's arrays, the closest things to them are possibly tuples

>>> import numpy as np
>>> columns = ["Port Wt", "Bench Wt", "Port Retn", "Bench Retn", 
...            "Attrib", "Select", "Inter", "Total"]
>>> a = np.ones((10,), dtype={"names":columns, "formats":["f8"]*len(columns)})
>>> type(a[0,0])
IndexError: too many indices
>>> type(a[0][0])
numpy.float64
>>> type(a[0])
numpy.void
>>>

on the contrary the columns of a are ndarray's and you can multiply them by a list of floats of the correct length (not the nuber of columns but the number of rows)

>>> type(a['Select'])
numpy.ndarray
>>> a['Select']*[1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8]
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-27-fc8dc4596098> in <module>()
----> 1 a['Select']*[1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8]

ValueError: operands could not be broadcast together with shapes (10,) (8,) 
>>> a['Select']*[1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,       0,0]
array([ 1.1,  1.2,  1.3,  1.4,  1.5,  1.6,  1.7,  1.8,  0. ,  0. ])
>>>

Edit

In response to a comment from OP: «is it not possible to apply a function to a row in a named array of fields (or tuple) in numpy?»

The only way that I know of is

>>> a[0] = tuple(b*a[c][0] for b, c in zip([1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8],columns))
>>> print a
[(1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)
 (1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)]
>>>

but I'm not the most skilled numpy expert around... maybe one of the least skilled indeed

Improve this answer

2 Comments

thanks @gboffi, does this mean that its not possible to apply a function to a row in a named array of fields (or tuple) in numpy? Or is there another way to do what I'm trying to do?
@TahnoonPasha I've edited my A to tell you what I know is possible. I cannot exclude that other methods, both more efficient enad more elegant, are possible but what I have added to my answer is the best that I know of. IMHO when you have a named ndarray you work with named columns, which contains homogeneous values. If you want to modify a row of your data (typically an event) you should do it before creating the named array.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.