CODEWITHSUNDEEP
pythonnumpyscipy
Sun Bear
Sun Bear

Reputation: 8234

Unable to change the dtype from float64 to int8 in NumPy/SciPy. How to overcome?

I wrote the following script. I am expecting np_Xs to be a np.int8 dtype. However, it is still np.float64 despite having declared np_Xs.astype( np.int8, copy=False ).

from scipy.stats import norm

import numpy as np
from numpy.random import default_rng

np_samples = np.random.randint(0, 10, size=2)
print( 'np_samples' )
print( np_samples, np_samples.size, '\n' )

rg = default_rng()
np_p_all = rg.random( size=np_samples.sum() )
print( 'np_p_all' )
print( np_p_all, np_p_all.size, '\n' )

np_Xs = np.floor( norm.ppf( np_p_all, loc=15, scale=5.25 ) )
np_Xs.astype( np.int8, copy=False )
print( 'np_Xs' )
print( np_Xs, np_Xs.size, np_Xs.dtype, '\n' )

Output:

np_samples
[7 3] 2 

np_p_all
[0.40001298 0.86285348 0.78885307 0.678035   0.08950794 0.15995076
 0.19581122 0.29012991 0.30003468 0.89203236] 10 

np_Xs
[13. 20. 19. 17.  7.  9. 10. 12. 12. 21.] 10 float64  #Why is np_Xs not a np.int8?

When I submitted the same commands via the idle terminal, np_Xs does change to a np.int8 dtype (see below). Why did the script version of the same commands (i.e. the .astype( np.int8, copy=False ) command) not provide the correct/same output as the idle terminal version?

>>> a = np.array( [0.40001298, 0.86285348, 0.78885307, 0.678035, 0.08950794, 0.15995076, 0.19581122, 0.29012991, 0.30003468, 0.89203236] )
>>> a
array([0.40001298, 0.86285348, 0.78885307, 0.678035  , 0.08950794,
       0.15995076, 0.19581122, 0.29012991, 0.30003468, 0.89203236])
>>> b = np.floor( norm.ppf( a, loc=15, scale=5.25 ) )
>>> b
array([13., 20., 19., 17.,  7.,  9., 10., 12., 12., 21.])
>>> b.astype( np.int8, copy=False )
array([13, 20, 19, 17,  7,  9, 10, 12, 12, 21], dtype=int8)
>>>

Also, I had tried explicitly declaring the dtype=np.int8 keyword in the np.floor() ufunc in my script. However, it had issue too. May I know how to resolve the TypeError?

    np_Xs = np.floor( norm.ppf( np_p_all, loc=15, scale=5.25 ), dtype=np.int8 )
TypeError: No loop matching the specified signature and casting was found for ufunc floor

Upvotes: 0

Views: 406

Answers (1)

krxat
krxat

Reputation: 523

When you do np_Xs.astype( np.int8, copy=False), it returns the array so you have to catch it back in the variable like

np_Xs = np_Xs.astype( np.int8, copy=False)

Now even if copy=False is set, it does not change the type inplace.

Upvotes: 2

Related Questions