Doe tensorflow has api similar to pytorch's “masked_fill_”

[jason-liew]

pytorch has squeeze , tf also has squeeze.
pytorch has unsqueeze, tf has expand_dims do the same thing with diff name.
pythorh has masked_fill , tf has ?

Please make sure that this is a feature request. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:feature_template

System information

• TensorFlow version (you are using): 2.2.0
• Are you willing to contribute it (Yes/No): Yes

Describe the feature and the current behavior/state.
masked_fill_ api doc in pytorch
Will this change the current api? How?
Maybe not. It will add a new api.
Who will benefit with this feature?
someone try to migrate from pytorch to tensorflow like me.
Any Other info.

[ravikyram]

@jason-liew

Can you please go through tf.boolean_mask and see this is the feature you are looking for?
Thanks!

[jason-liew]

No,it's different.

tensorflow:

 t1 = tf.constant([1,2,3])
 t1_mask = numpy.array([1,0,1])
 tf.boolean_mask(t1,t1_mask) # [1, 3]

pytorch:

p1 = torch.tensor([1,2,3])
p1_mask = torch.tensor([1,0,1])
p1.masked_fill_(p1_mask,5) # [5, 2, 5]

[ravikyram]

@jason-liew

Got it.Thanks!
Are you interested to submit a PR?

[aavishkarmishra]

@gowthamkpr Hi, I would like to work in this. Can anyone help with how and from where I can start ?

[aavishkarmishra]

@gowthamkp Should we have to add a new api or we can also override boolean_mask with one more parameter, which one will be better?

[rohan100jain]

Strawman proposal. Can this be implemented as

def masked_fill(t, mask, value):
return t * (1 - tf.cast(mask, tf.float32)) + value * tf.cast(mask, tf.float32)

Might have to do some dtype handling but roughly should work? What are the performance requirements for this?

[aavishkarmishra]

@ravikyram @gowthamkpr please review this PR #41975

[fsx950223]

tf.where([True, False, True],5,[1,2,3])

[alextp]

We recommend the tf.where implementation.