| import torch | |
| def kde(x, std=0.1): | |
| # use a gaussian kernel to estimate density | |
| x = x.half() # Do it in half precision | |
| scores = (-torch.cdist(x, x) ** 2 / (2 * std**2)).exp() | |
| density = scores.sum(dim=-1) | |
| return density | |
| import torch | |
| def kde(x, std=0.1): | |
| # use a gaussian kernel to estimate density | |
| x = x.half() # Do it in half precision | |
| scores = (-torch.cdist(x, x) ** 2 / (2 * std**2)).exp() | |
| density = scores.sum(dim=-1) | |
| return density | |