CODEWITHSUNDEEP
pythonnumpy
Wallace
Wallace

Reputation: 580

Keep getting ValueError with numpy while trying to create array

I am trying to create a new array with an existing array's elements. I keep getting:ValueError: Setting void-array with object members using buffer. 

import numpy as np
import datetime

date = datetime.date(2013, 4, 5)
results = [(date,0,1,2,3), (date,5,1,5,6), (date,3,4,4,7)] 
stock_dt = np.dtype([('date', object),
                     ('open', np.int8),
                     ('high', np.int8),
                     ('low', np.int8),
                     ('close', np.int8)])

d = np.array(results, dtype=stock_dt)
matches = []
for item in d:
    if item['high'] == 1:
        matches.append(item)

rec = np.array(matches, dtype=stock_dt)

print rec

Upvotes: 1

Views: 2227

Answers (2)

Brendan Dolan-Gavitt
Brendan Dolan-Gavitt

Reputation: 667

The problem is that matches is not a list of tuples, so you cant make a structured array out of it. Instead it's a list of structured arrays, which need to be merged back into a single structured array. You can use numpy.lib.recfunctions.stack_arrays for this:

In [21]: import numpy.lib.recfunctions as rfn

In [22]: rfn.stack_arrays(matches,usemask=False)
Out[22]: 
array([(datetime.date(2013, 4, 5), 0, 1, 2, 3),
       (datetime.date(2013, 4, 5), 5, 1, 5, 6)], 
      dtype=[('date', 'O'), ('open', 'i1'), ('high', 'i1'), ('low', 'i1'), ('close', 'i1')])

You could also consider doing away with the loop entirely:

In [23]: d[d['high'] == 1]
Out[23]: 
array([(datetime.date(2013, 4, 5), 0, 1, 2, 3),
       (datetime.date(2013, 4, 5), 5, 1, 5, 6)], 
      dtype=[('date', 'O'), ('open', 'i1'), ('high', 'i1'), ('low', 'i1'), ('close', 'i1')])

Which should be faster, to boot.

Upvotes: 4

Jared
Jared

Reputation: 26397

Change

rec = np.array(matches, dtype=stock_dt)

to

rec = np.array(matches)

When you're iterating over matches you aren't dealing with tuples anymore so you shouldn't pass dtype=stock_dt to np.array again.

Upvotes: 1

Related Questions