from akida.core import Layer, LayerType, LayerParams
[docs]
class FullyConnected(Layer):
"""This represents a Dense or Linear neural layer.
A standard Dense Layer in a network is converted to FullyConnected Layer
on Akida.
This layer optionally executes ReLU operation to the outputs of the Dense
Layer.
The FullyConnected layer accepts 1-bit, 2-bit or 4-bit input tensors.
The FullyConnected can be configured with 1-bit, 2-bit or 4-bit weights.
It multiplies the inputs by its internal unit weights, returning a 4D
tensor of values whose first dimension is the number of samples and the
last dimension represents the number of units.
It can optionally apply a step-wise ReLU activation to its outputs.
Args:
units (int): number of units.
name (str, optional): name of the layer. Defaults to empty string.
weights_bits (int, optional): number of bits used to quantize weights.
Defaults to 1.
activation (bool, optional): enable or disable activation
function. Defaults to True.
act_bits (int, optional): number of bits used to quantize the neuron
response. Defaults to 1.
"""
def __init__(self,
units,
name="",
weights_bits=1,
activation=True,
act_bits=1):
try:
params = LayerParams(
LayerType.FullyConnected, {
"units": units,
"weights_bits": weights_bits,
"activation": activation,
"act_bits": act_bits
})
# Call parent constructor to initialize C++ bindings
# Note that we invoke directly __init__ instead of using super, as
# specified in pybind documentation
Layer.__init__(self, params, name)
except BaseException:
self = None
raise