record array is made up of numpy.void, not numpy.record #271

Activity

1. scopatz commented on Aug 5, 2013

   

   scopatz

   on Aug 5, 2013

   Member

   More actions

   Don't you need to do .view(type=np.recarry)?
2. potyt commented on Aug 6, 2013

   

   potyt

   on Aug 6, 2013

   Author

   More actions

   Exactly, that's what I am asking about:

   >>> import numpy as np
   >>> a = (1, 2, 3)
   >>> ra = np.rec.fromrecords([a])
   >>> na = np.array([a], dtype='i8,i8,i8').view(type=np.recarray)
   >>> na
   rec.array([(1, 2, 3)], 
         dtype=[('f0', '<i8'), ('f1', '<i8'), ('f2', '<i8')])
   >>> ra
   rec.array([(1, 2, 3)], 
         dtype=[('f0', '<i8'), ('f1', '<i8'), ('f2', '<i8')])
   >>> type(ra)
   <class 'numpy.core.records.recarray'>
   >>> type(na)
   <class 'numpy.core.records.recarray'>
   >>> ra[0]
   (1, 2, 3)
   >>> na[0]
   (1, 2, 3)
   >>> type(ra[0])
   <class 'numpy.core.records.record'>
   >>> type(na[0])
   <class 'numpy.void'>
   

   Now, I realise the example above is NumPy only, but the same type of object, a sort of 'fake recarray', gets returned from table.read, so wanted to see if you guys were aware of this
3. scopatz commented on Aug 6, 2013

   

   scopatz

   on Aug 6, 2013

   Member

   More actions

   Hi @potyt, I was not aware of this behavior. (I almost never use recarrays _.) But this is not surprising. Reading data out of Tables returns numpy.ndarrays. After reading, I would expect these to perform exactly the same way as the pure numpy example above. I think that you should report this to numpy. Since it is outside the scope of PyTables, would you mind if I closed this as invalid?
4. potyt commented on Aug 6, 2013

   

   potyt

   on Aug 6, 2013

   Author

   More actions

   Sure, I'll raise this on a NumPy list, ah, it's on github too...
5. avalentino commented on Aug 6, 2013

   

   avalentino

   on Aug 6, 2013

   Member

   More actions

   Just for the record, a simple workaround could be to swap the order of operations: arr.price[0] instead of arr[0].price

   In [14]: na[0].f0
   ---------------------------------------------------------------------------
   AttributeError                            Traceback (most recent call last)
   <ipython-input-14-cca549dcc2ad> in <module>()
   ----> 1 na[0].f0
   
   AttributeError: 'numpy.void' object has no attribute 'f0'
   
   In [15]: na.f0
   Out[15]: array([1])
   
   In [16]: na.f0[0]
   Out[16]: 1
   
   In [17]: ra.f0[0]
   Out[17]: 1
   
   
6. potyt commented on Aug 6, 2013

   

   potyt

   on Aug 6, 2013

   Author

   More actions

   Indeed, been making these sorts of compromises for a while, but on the occasion that you need to row-by-row iterator through an array, rather than access the data column-wise, this isn't so easy (or efficient I would think, as you would index into the overall array twice with something like arr.price[0] > arr.limit[0] rather than row.price > row.limit

   Not a big deal for one simple comparison, but if you are picking out a lot of these columns, and acting on their values in a loop that runs a lot, it could add up!?

   More importantly, it also means keeping track of your own row index(es), which could introduce the odd bug, reduces the readability of the code and takes away some of the elegance of the row-based approach.

   Thanks for the v.quick responses btw.
7. FrancescAlted commented on Apr 27, 2015

   

   FrancescAlted

   on Apr 27, 2015

   Member

   More actions

   Out of scope of PyTables.
8. FrancescAlted

   added

   invalid

   on Apr 27, 2015
9. FrancescAlted

   closed this as completedon Apr 27, 2015