ValueError: mask cannot be scalar.

[logictensornetworks]

ValueError: mask cannot be scalar.

[sbadredd]

Hi Milena,

Can you send me the rest of your code so I can have a look?

At first glance, this seems to be a TensorFlow error, not related to ltn. If you write:

mask = (labels_position=="P")
tensor_P = features[mask]
x_P = ltn.variable("P", tensor_P)

• Does the error occur on the second line?
• If yes, can you check the value and type of mask, is it a tf.Tensor/np.ndarray, or a scalar value? It seems that in your case, mask could be just a boolean, which would cause the Tensorflow error mask cannot be scalar
• If mask is not a tf.Tensor, you need to convert labels_position to a tensor beforehand. Also, try writing mask = tf.equal(labels_position,"P") to be more explicit

Hope it helps!

[MilenaTenorio]

Hi!

Thank you very much for the feedback, I will carry out the suggestions sent.
I can tell you what worked correctly in my code and share the link to access it

It will certainly help me

[MilenaTenorio]

Dear,

I add the code suggested:

    mask = (labels_position=="P")
    tensor_P = features[mask]

here:
mask = (labels_position=="P")-> return: FALSE
and
tensor_P = features[mask] -> return:

ValueError: in converted code:

    <ipython-input-13-f5edfa7f8178>:8 axioms *
        tensor_P = features[mask]
    C:\Users\Milena\Anaconda3\lib\site-packages\tensorflow\python\ops\array_ops.py:600 _slice_helper
        return boolean_mask(tensor=tensor, mask=slice_spec)
    C:\Users\Milena\Anaconda3\lib\site-packages\tensorflow\python\ops\array_ops.py:1365 boolean_mask
        raise ValueError("mask cannot be scalar.")

    ValueError: mask cannot be scalar.

[MilenaTenorio]

I have made a summary of the code with information that I believe is relevant to the problem I am currently trying to solve.
It's shared: spatialrelation-resume.git

[MilenaTenorio]

Now I'll try to convert labels_position to a tensor beforehand!

Thanks for your support!

[sbadredd]

With pleasure!

Writing something like:

mask_P = tf.equal(tf.constant(labels_position),"P")
tensor_P = features[mask_P]

should help.

[sbadredd]

By the way, if the variable is empty (there are no values of label "P"), the code will return an error when you write your formula.
So maybe add something like:

axioms = []
if tf.reduce_sum(tf.cast(mask_P,tf.int32)) != 0:                     # Only if x_P is not empty...
    axioms.append(Forall(x_P,p([x_P,class_P],training=training)))    # ... add the axiom

In the future, I might add some verifications for cases where the variables are empty

[MilenaTenorio]

Thank you very much for the support

[gaokun12]

您的来信我已收到！