Lost information when converting from fits rec to ndaray

Question

I loaded a fits file and converted the fitsrec data in a numpy ndarray:

import pyfits
import os, numpy as np
dataPath ='irac1_dataset.fits'

hduTab=pyfits.open(dataPath)
data_rec = np.array(hduTab[1].data)
data=data_rec.view(np.float64).reshape(data_rec.shape + (-1,))

I found that in data there are some nan that don't exist in the rec:

data_rec[3664]
(2.52953742092, 3.636058484, -3.0, 1.16584000133, 0.13033115092, 0.0545114121049, 0.0977915267677, 0.0861630982921, 0.0935291710016)
data[3664]
array([  8.01676073e+230,  -1.68253090e-183,   1.10670705e-320,
        -5.38247269e-235,               nan,   3.19504591e+186,
        -6.19704421e+125,  -1.40287783e+079,   1.94744862e+094])

and, as you can see, the values are changed in significant way, how it is possible?

About hduTab[1].data:

data_rec = hduTab[1].data
>>> data_rec.dtype
dtype((numpy.record, [('entr_35_1', '>f8'), ('kurt_5_1', '>f8'), ('skew_23_1', '>f8'), ('skew_35_1', '>f8'), ('mean_23_2', '>f8'), ('mean_35_2', '>f8'), ('stdDev_23_1', '>f8'), ('stdDev_35_1', '>f8'), ('pixVal', '>f8')]))

is a numpy record

What's the `dtype` of `data_rec`? Or better yet, give more information about `hduTab[1].data`. I haven't worked with `pyfits`. — hpaulj, Jun 29 '17 at 20:08
What are you actually trying to do? Because I'm not sure why you're making these manipulations of the array type. — Iguananaut, Jun 30 '17 at 11:43
I have to convert the contents of the fits file (whole table) into a numpy array — Giuseppe Angora, Jun 30 '17 at 12:29

hpaulj · Accepted Answer · 2017-06-29T23:02:02.933

It's the `>f8' that messing you up.

In [380]: dt= [('entr_35_1', '>f8'), ('kurt_5_1', '>f8'), ('skew_23_1', '>f8'), 
     ...: ('skew_35_1', '>f8'), ('mean_23_2', '>f8'), ('mean_35_2', '>f8'), ('st
     ...: dDev_23_1', '>f8'), ('stdDev_35_1', '>f8'), ('pixVal', '>f8')]

In [382]: np.dtype(dt)
Out[382]: dtype([('entr_35_1', '>f8'),....('pixVal', '>f8')])

In [383]: np.array([(2.52953742092, 3.636058484, -3.0, 1.16584000133, 0.13033115
     ...: 092, 0.0545114121049, 0.0977915267677, 0.0861630982921, 0.093529171001
     ...: 6)],dtype=dt)
Out[383]: 
array([ ( 2.52953742,  3.63605848, -3.,  1.16584,  0.13033115,  0.05451141,  0.09779153,  0.0861631,  0.09352917)], 
      dtype=[('entr_35_1', '>f8'), ('kurt_5_1', '>f8'), ('skew_23_1', '>f8'), ('skew_35_1', '>f8'), ('mean_23_2', '>f8'), ('mean_35_2', '>f8'), ('stdDev_23_1', '>f8'), ('stdDev_35_1', '>f8'), ('pixVal', '>f8')])
In [384]: x=_

A float view has the nan and unrecognizable values:

In [385]: x.view(float)
Out[385]: 
array([  8.01676073e+230,  -1.68253090e-183,   1.10670705e-320,
        -5.38247269e-235,               nan,   3.19504591e+186,
        -6.19704421e+125,  -1.40287783e+079,   1.94744862e+094])

But view with >f8 matches the input:

In [386]: x.view('>f8')
Out[386]: 
array([ 2.52953742,  3.63605848, -3.        ,  1.16584   ,  0.13033115,
        0.05451141,  0.09779153,  0.0861631 ,  0.09352917])

I can then use astype to convert to float, (which evidently is <f8):

In [387]: _.astype(float)
Out[387]: 
array([ 2.52953742,  3.63605848, -3.        ,  1.16584   ,  0.13033115,
        0.05451141,  0.09779153,  0.0861631 ,  0.09352917])

In [389]: np.dtype('<f8')
Out[389]: dtype('float64')
In [390]: np.dtype('>f8')
Out[390]: dtype('>f8')

Using astype can be tricky, but it appears that if I keep the field layout the same I can use it directly. So I can use it to change '>f8' to

In [407]: dt1= [('entr_35_1', '<f8'), ('kurt_5_1', '<f8'), ('skew_23_1', '<f8'),
     ...:  ('skew_35_1', '<f8'), ('mean_23_2', '<f8'), ('mean_35_2', '<f8'), ('s
     ...: tdDev_23_1', '<f8'), ('stdDev_35_1', '<f8'), ('pixVal', '<f8')]
In [408]: x.astype(dt1)
Out[408]: 
array([ ( 2.52953742,  3.63605848, -3.,  1.16584,  0.13033115,  0.05451141,  0.09779153,  0.0861631,  0.09352917)], 
      dtype=[('entr_35_1', '<f8'), ('kurt_5_1', '<f8'), ('skew_23_1', '<f8'), ('skew_35_1', '<f8'), ('mean_23_2', '<f8'), ('mean_35_2', '<f8'), ('stdDev_23_1', '<f8'), ('stdDev_35_1', '<f8'), ('pixVal', '<f8')])

I still need to use view to change the number of fields:

In [409]: x.astype(dt1).view(float)
Out[409]: 
array([ 2.52953742,  3.63605848, -3.        ,  1.16584   ,  0.13033115,
        0.05451141,  0.09779153,  0.0861631 ,  0.09352917])

Lost information when converting from fits rec to ndaray

1 Answers1