How can I do any constraints in the same layer?

Savano_Pereira · January 27, 2022, 2:10pm

Dears,
I want to do two weights constraints in the same layer. My weights should be non-negatives and w_1 > w_2 > w_3 > … > w_N. I would like to use tf.keras.constraints. Did you do this?

Regards.

Bhack · January 27, 2022, 4:52pm

Probably you could try with the custom constrain section:

Savano_Pereira · January 27, 2022, 5:00pm

Thanks Back. But my question is how to impose the constraints w_1 > w_2 > w_3 > … > w_N? Do you know?

Regards.

Bhack · January 27, 2022, 7:55pm

If you check the other “canned” constrains you can see that you need to just manipulate the weights that are not respecting your constrain.

In your case I suggest to check this implementation as a baseline to retrieve the indexes of the weights that you want to manipulate:

github.com

tensorflow/tensorflow/blob/v2.7.0/tensorflow/python/ops/check_ops.py#L2108-L2148

      
        
            @tf_export(
                'math.is_strictly_increasing',
                v1=[
                    'math.is_strictly_increasing', 'debugging.is_strictly_increasing',
                    'is_strictly_increasing'
                ])
            @dispatch.add_dispatch_support
            @deprecation.deprecated_endpoints('debugging.is_strictly_increasing',
                                              'is_strictly_increasing')
            def is_strictly_increasing(x, name=None):
              """Returns `True` if `x` is strictly increasing.
            
            
  Elements of `x` are compared in row-major order.  The tensor `[x[0],...]`
              is strictly increasing if for every adjacent pair we have `x[i] < x[i+1]`.
              If `x` has less than two elements, it is trivially strictly increasing.
            
            
  See also:  `is_non_decreasing`
            
            
  >>> x1 = tf.constant([1.0, 2.0, 3.0])
              >>> tf.math.is_strictly_increasing(x1)

This file has been truncated. show original