numerical stability

Author: rusty1s

torch_scatter/std.py

@@ -6,14 +6,16 @@
 
 def scatter_std(src, index, dim=-1, out=None, dim_size=None, unbiased=True):
     src, out, index, dim = gen(src, index, dim, out, dim_size, fill_value=0)
+    print('src', src.mean())
 
     tmp = scatter_add(src, index, dim, None, dim_size)
     count = scatter_add(torch.ones_like(src), index, dim, None, dim_size)
     mean = tmp / count.clamp(min=1)
 
-    var = (src - mean.gather(dim, index))**2
+    var = (src - mean.gather(dim, index))
+    var = var * var
     out = scatter_add(var, index, dim, out, dim_size)
     out = out / (count - 1 if unbiased else count).clamp(min=1)
-    out = torch.sqrt(out)
+    out = torch.sqrt(out.clamp(min=1e-12))
 
     return out