pulsar

zeta/nn/modules/__init__.py

@@ -21,3 +21,4 @@
 from zeta.nn.modules.simple_attention import simple_attention
 from zeta.nn.modules.spacial_transformer import SpacialTransformer
 from zeta.nn.modules.yolo import yolo
+from zeta.nn.modules.pulsar import Pulsar

zeta/nn/modules/pulsar.py

@@ -0,0 +1,69 @@
+import torch 
+from torch import nn
+import torch.nn.functional as F
+
+
+class LogGammaActivation(torch.autograd.Function):
+ """
+ PulSar Activation function that utilizes factorial calculus
+
+ PulSar Activation function is defined as:
+ f(x) = log(gamma(x + 1))
+ where gamma is the gamma function
+
+ The gradient of the PulSar Activation function is defined as:
+ f'(x) = polygamma(0, x + 2)
+ where polygamma is the polygamma function
+
+ Methods:
+ forward(ctx, input): Computes the forward pass
+ backward(ctx, grad_output): Computes the backward pass
+ """
+
+ @staticmethod
+ def forward(ctx, input):
+ """
+ Forward pass of the PulSar Activation function
+ 
+ """
+ #compute forward pass
+ gamma_value = torch.lgamma(input + 1)
+ ctx.save_for_backward(input, gamma_value)
+ return gamma_value
+
+ @staticmethod
+ def backward(ctx, grad_output):
+ """
+ Backward pass of the PulSar Activation function
+ """
+ #compute gradient for backward pass
+ input, gamma_value = ctx.saved_tensors
+ polygamma_val = torch.polygamma(0, input + 2)
+ return polygamma_val * grad_output
+
+
+class Pulsar(nn.Module):
+ """
+ Pulsar Activation function that utilizes factorial calculus
+
+ Pulsar Activation function is defined as:
+ f(x) = log(gamma(x + 1))
+ where gamma is the gamma function
+
+
+ Usage:
+ x = torch.tensor([1.0, 2.0, 3.0, 4.0, 5.0], requires_grad=True)
+ pulsar = Pulsar()
+ y = pulsar(x)
+ print(y)
+ y = y.backward(torch.ones_like(x))
+
+
+ 
+ """
+ def forward(self, x):
+ """
+ Forward pass of the PulSar Activation function
+ """
+ return LogGammaActivation.apply(x)
+