80 lines
2.3 KiB
Python
80 lines
2.3 KiB
Python
from typing import Any, Dict
|
|
import ast
|
|
import torch
|
|
from torch.distributions import Gamma
|
|
import scipy.stats as scst
|
|
|
|
|
|
class GammaEx(Gamma):
|
|
"""
|
|
A class to extend Gamma to add missing functionality.
|
|
|
|
pragma: skip_doc
|
|
"""
|
|
|
|
def icdf(self, q: torch.Tensor):
|
|
"""
|
|
Computes the inverse of cumulative distribution function (CDF) of the Gamma distribution.
|
|
|
|
Args:
|
|
q (torch.Tensor): Input tensor containing values where the ICDF is evaluated.
|
|
|
|
Returns:
|
|
torch.Tensor: Tensor containing the inverse of cumulative probabilities.
|
|
"""
|
|
x2 = q.detach().cpu().numpy()
|
|
a = self.concentration.detach().cpu().item()
|
|
b = self.rate.detach().cpu().item()
|
|
outputs = scst.gamma.ppf(x2, a, 1/b) # Note that you need to pass the inverse of b.
|
|
return torch.tensor(outputs, dtype=torch.float32).to(q.device)
|
|
|
|
|
|
class PtdGamma:
|
|
"""
|
|
Instantiates a Gamma distribution object.
|
|
|
|
Args:
|
|
alpha: Shape. You can specify a scalar or a tuple of float.
|
|
beta: Rate. You can specify a scalar or a tuple of float.
|
|
|
|
category: PyTorch wrapper - Distribution
|
|
"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls) -> Dict[str, Any]:
|
|
"""
|
|
Defines the input types for the function.
|
|
|
|
Returns:
|
|
Dict[str, Any]: A dictionary specifying the required input types.
|
|
"""
|
|
return {
|
|
"required": {
|
|
"alpha": ("STRING", {"default": ""}),
|
|
"beta": ("STRING", {"default": ""})
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES: tuple = ("PTDISTRIBUTION",)
|
|
FUNCTION: str = "f"
|
|
CATEGORY: str = "Distribution"
|
|
|
|
def f(self, alpha: str, beta: str) -> tuple:
|
|
"""
|
|
Instantiates a Gamma distribution object.
|
|
|
|
Args:
|
|
alpha (str): Concentration You can specify a scalar or a tuple of float.
|
|
beta (str): Rate. You can specify a scalar or a tuple of float.
|
|
|
|
Returns:
|
|
tuple: (torch.distributions.binomial.Gamma)
|
|
"""
|
|
c = ast.literal_eval(alpha)
|
|
r = ast.literal_eval(beta)
|
|
|
|
dist = GammaEx(
|
|
concentration=torch.tensor(c, dtype=torch.float32),
|
|
rate=torch.tensor(r, dtype=torch.float32))
|
|
return (dist,)
|