Just to be sure, oou want to replace inf/nan with a numerical value in
Theano? This can be done like this:
import numpy as np
a = np.array([0,0,1,1,2], dtype='float')
b = np.array([0,1,0,1,3], dtype='float')
with np.errstate(divide='ignore', invalid='ignore'):
c = np.true_divide(a,b)
import theano.tensor as T
import theano
c=theano.tensor.as_tensor_variable(c)
inf_mask = T.isinf(c)
nan_mask = T.isnan(c)
inf_idx = inf_mask.nonzero()
nan_idx = nan_mask.nonzero()
c_without_inf = theano.tensor.set_subtensor(c[inf_idx], -999)
c_without_inf_nan = theano.tensor.set_subtensor(c_without_inf[nan_idx], 0)
c_without_inf_nan.eval()
array([ 0.00000000e+00, 0.00000000e+00, -9.99000000e+02,
1.00000000e+00, 6.66666667e-01])
I don't know if it make a difference between inf and -inf.
Fred
On Thu, Jul 28, 2016 at 11:52 PM Varglur <[email protected]> wrote:
> dear all,
>
> similar to problem in numpy here
> <http://stackoverflow.com/questions/26248654/numpy-return-0-with-divide-by-zero>
>
> if element wise divide in theano , are there corresponding function in
> theano as this answer in numpy ?
>
> numpy answer:
>
> import numpy as np
>
> a = np.array([0,0,1,1,2], dtype='float')
> b = np.array([0,1,0,1,3], dtype='float')
> with np.errstate(divide='ignore', invalid='ignore'):
> c = np.true_divide(a,b)
> c[c == np.inf] = 0
> c = np.nan_to_num(c)
> print('c: {0}'.format(c))
>
> Output:
>
> c: [ 0. 0. 0. 1. 0.66666667]
>
>
>
> how to process inf and nan problem when 0/0 or 1/0 problem in theano??
>
> thanks
>
>
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "theano-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
