hxtorch.perceptron.nn.MACLayer

class hxtorch.perceptron.nn.MACLayer(num_sends: Optional[numbers.Integral] = None, wait_between_events: numbers.Integral = 5, mock: bool = False, *, input_transform: Optional[Callable[[torch.Tensor], torch.Tensor]] = None, weight_transform: Optional[Callable[[torch.Tensor], torch.Tensor]] = <function clamp_weight_>)

Bases: hxtorch.perceptron.nn.Layer

Layer that performs a multiply accumulate operation.

__init__(num_sends: Optional[numbers.Integral] = None, wait_between_events: numbers.Integral = 5, mock: bool = False, *, input_transform: Optional[Callable[[torch.Tensor], torch.Tensor]] = None, weight_transform: Optional[Callable[[torch.Tensor], torch.Tensor]] = <function clamp_weight_>)
Parameters

  • num_sends – Number of sends of the input. Values greater than 1 result in higher output to the neurons and increases the s/n ratio. For None this is automatically adjusted during initialization.

  • wait_between_events – Wait time between two successive vector inputs, in FPGA clock cycles. Shorter wait time can lead to saturation of the synaptic input.

  • mock – Enable mock mode.

  • input_transform – Function that receives the input and returns a tensor to be used as input to the chip.

  • weight_transform – Function that receives the weight and returns a tensor to be used as weight matrix on the chip.

Methods

__init__([num_sends, wait_between_events, …])

param num_sends

Number of sends of the input. Values greater than 1

reset_parameters([weight_mean, relu_shift])

Reset parameters to reasonable initialization values.
reset_parameters(weight_mean: numbers.Real = 0.0, relu_shift: numbers.Integral = 1) → None

Reset parameters to reasonable initialization values. Method based on Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification - He, K. et al. (2015)

Parameters

  • weight_mean – Mean value of the weight distribution

  • relu_shift – Bit shift assumed in subsequent ConvertingReLU