Dandelion's activation module is mostly inherited from Lasagne except for the softmax() and log_softmax() activations.
You're recommended to refer to Lasagne.nonlinearities document for the following activations:
- sigmoid
- tanh
- relu
- softplus
- ultra_fast_sigmoid
- ScaledTanH
- leaky_rectify
- very_leaky_rectify
- elu
- SELU
- linear
- identity
softmax
Apply softmax to the last dimension of input x
softmax(x)
- x: theano tensor of any shape
log_softmax
Apply softmax to the last dimension of input x, in log domain
log_softmax(x)
- x: theano tensor of any shape