Using numpy roll in Keras

Question

I'm trying to make a custom regularizer in Keras and I need to be able to roll the coefficient array.

I know this may be impossible however any mechanism that can replicate this roll function would be extremely appreciated.

```

def __call__(self, x):
    regularization = 0.

    # Add components if they are given
    if self.l1:
        # \lambda ||x||
        regularization += self.l1 * K.sum(K.abs(x))
    if self.fuse:
        # \lambda \sum{ |x - x_+1| }
        regularization += self.fuse * K.sum(K.abs(x - np.roll(x, 1)))
    if self.abs_fuse:
        # \lambda \sum{ ||x| - |x_+1|| }
        regularization += self.abs_fuse * K.sum(K.abs(K.abs(x) - K.abs(np.roll(x, 1))))

```

What's the dimension of x? If x is a matrix then it's probably a bit weird. Calling np.roll on a weight matrix will first flatten the matrix, shift the element linearly, and finally reshape it back to a matrix. Is this behavior exactly what you want, or is it np.roll(x, 1, axis=1)? — Yu-Yang
– Yu-Yang, Commented Aug 1, 2017 at 3:11
the dimensionality of x is always (m, 1) . And I know from testing it on a toy example with numpy that roll is the behaviour I'm looking for — SARose
– SARose, Commented Aug 1, 2017 at 4:07
@Yu-Yang EDIT: np.roll(x, 1, axis=1) gives the same result as np.roll(x, 1) in this example but axis=1 is more explicit. — SARose
– SARose, Commented Aug 1, 2017 at 4:15
I've posted a possible way to implement it, please see if that's what you want. — Yu-Yang
– Yu-Yang, Commented Aug 1, 2017 at 4:22

Yu-Yang · Accepted Answer · 2017-08-01 04:22:17Z

1

Given that x is of shape (m, 1), a possible solution is to use tile:

def roll_reg(x):
    length = K.int_shape(x)[0]
    x_tile = K.tile(x, [2, 1])
    x_roll = x_tile[length - 1:-1]
    return K.sum(K.abs(x - x_roll))

It will result in some extra memory usage, but if x is a 1-dim vector, I guess the overhead won't be too bad.

answered Aug 1, 2017 at 4:22

Yu-Yang

14.6k2 gold badges57 silver badges62 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Using numpy roll in Keras

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related