numpy.switch

Here are the examples of the python api numpy.switch taken from open source projects. By voting up you can indicate which examples are most useful and appropriate.

1 Examples 7

Example 1

Project: pylearn2 Source File: gsn.py
    @staticmethod
    def _apply_clamping(activations, clamped, symbolic=True):
        """
        Resets the value of some layers within the network.

        Parameters
        ----------
        activations : list
            List of symbolic tensors representing the current activations.
        clamped : list of (int, matrix, matrix or None) tuples
            The first component of each tuple is an int representing the
            index of the layer to clamp.
            The second component is a matrix of the initial values for that
            layer (ie what we are resetting the values to).
            The third component is a matrix mask indicated which indices in
            the minibatch to clamp (1 indicates clamping, 0 indicates not).
            The value of None is equivalent to the 0 matrix (so no clamping).
            If symbolic is true then matrices are Theano tensors, otherwise
            they should be numpy matrices.
        symbolic : bool, optional
            Whether to execute with symbolic Theano tensors or numpy matrices.
        """
        for idx, initial, clamp in clamped:
            if clamp is None:
                continue

            # take values from initial
            clamped_val = clamp * initial

            # zero out values in activations
            if symbolic:
                activations[idx] = T.switch(clamp, initial, activations[idx])
            else:
                activations[idx] = np.switch(clamp, initial, activations[idx])
        return activations