Using numpy functions when dtype is object

Question

I have a a column in my numpy array which has a bunch of longs and a few NaNs. There are no other floats in it. If I convert the entire column to float I will lose precision.

If I don't convert the entire column to float, I cannot use numpy functions like isnan or isfinite because the array is of type object despite the elements being valid types.

Is there anyway of preserving precision while being able to use numpy functions?

a = np.array([10**50,19**50,float('NaN')])

a
#outputs:
array([100000000000000000000000000000000000000000000000000L,
   8663234049605954426644038200675212212900743262211018069459689001L,
   nan], dtype=object)

np.isnan(a)

#outputs error:

ufunc 'isnan' not supported for the input types, and the inputs could not be safely 
coerced to any supported types according to the casting rule ''safe''

You could also consider using a masked array instead of using NaN's. That may or may not fit your problem, but it avoids the need to directly call np.isnan, etc. — Joe Kington
– Joe Kington, Commented Feb 8, 2014 at 1:13
@JoeKington actually I need np.isnan to mask my arrays...I don't actually know where the Nans are...I will accept Zhangxaochen's answer as it does what I need currently... — Terence Chow
– Terence Chow, Commented Feb 8, 2014 at 1:16

zhangxaochen · Accepted Answer · 2014-02-08 01:27:03Z

4

In [1760]: np.isnan(a.astype(float))
Out[1760]: array([False, False,  True], dtype=bool)

So if precision is not that important, just a=a.astype(float).

edited Feb 8, 2014 at 1:27

answered Feb 8, 2014 at 1:11

zhangxaochen

34.2k15 gold badges82 silver badges115 bronze badges

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

1 Comment

Terence Chow Over a year ago

I like this solution. I will need to use other numpy functions but for this situation is a good work around.

Dietrich · Accepted Answer · 2014-02-08 01:14:55Z

0

A possibility could just ignoring the nans:

ii = np.where([type(a) is long for a in A if type(a)]
B = 5*A[ii]

answered Feb 8, 2014 at 1:14

Dietrich

5,6014 gold badges28 silver badges42 bronze badges

Collectives™ on Stack Overflow

Using numpy functions when dtype is object

2 Answers 2

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related