Define dtypes in NumPy using a list?

Question

I just am having a problem with NumPy dtypes. Essentially I'm trying to create a table that looks like the following (and then save it using rec2csv):

      name1   name2   name3 . . . 
name1  #       #      #
name2  #       #      #
name2  #       #      #
.
.
.

The matrix (numerical array in the center), is already computed before I attempt to add the name tags. I've tried to use the following code:

    dt = dtype({'names' : tuple(blah), 'formats' : tuple(fmt)}) 
    ReadArray = array(tuplelist, dtype=dt)

where tuplelist is a list of rows (i.e. the row [name1, #, #, #...]), blah is a list of strings (i.e. the names, blah = ['name1', 'name2', ...]) and fmt is the list of format,s (i.e. fmt = [str, float, float, ...]).

The error I'm getting is the following:

Traceback (most recent call last):

  File "<stdin>", line 1, in <module>
  File "table_calc_try2.py", line 152, in table_calc_try2
    dt = dtype({'names' : tuple(blah), 'formats' : tuple(fmt)}) 
TypeError: data type not understood

Can anyone help?

Thanks!

unutbu · Accepted Answer · 2010-08-05 00:38:40Z

12

The following code might help:

import numpy as np

dt = np.dtype([('name1', '|S10'), ('name2', '<f8')])
tuplelist=[
    ('n1', 1.2),
    ('n2', 3.4),    
     ]
arr = np.array(tuplelist, dtype=dt)

print(arr['name1'])
# ['n1' 'n2']
print(arr['name2'])
# [ 1.2  3.4]

Your immediate problem was that np.dtype expects the format specifiers to be numpy types, such as '|S10' or '<f8' and not Python types, such as str or float. If you type help(np.dtype) you'll see many examples of how np.dtypes can be specified. (I've only mentioned a few.)

Note that np.array expects a list of tuples. It's rather particular about that.

A list of lists raises TypeError: expected a readable buffer object.

A (tuple of tuples) or a (tuple of lists) raises ValueError: setting an array element with a sequence.

answered Aug 5, 2010 at 0:38

unutbu

886k197 gold badges1.9k silver badges1.7k bronze badges

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

3 Comments

Joe Kington Over a year ago

Just as a note: a dict of the form the OP gave is a perfectly valid dtype (other than not specifying numpy types, as you mentioned (e.g. np.float rather than float)). It doesn't have to be a list of tuples, and specifying a dict of {'names':['f0', 'f1' ...], 'formats':[np.float, np.int, ...]} as a dtype is often a lot more convenient.

Merlin Over a year ago

@Joe, Can 'f0', 'f1' be index? I typically have one col date or string, the rest are floats. Ex: 'foo', 1,2,2,44,3 or 22,2,2,2,2,'3/2/2001' whats the best dtype solutions?

user707650 Over a year ago

+1 for noting the exceptions raised when the data input is different than a list of tuples, having bumped myself into this too often.

Collectives™ on Stack Overflow

Define dtypes in NumPy using a list?

1 Answer 1

3 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related