When i trained a neural network i found something interesting!

data
1

When i trained a neural network ,
code:
class linear_regression(nn.Module):
def init(self):
super().init()
self.weight = nn.Parameter(torch.rand(1,
requires_grad = True,
dtype=torch.float))
self.bias = nn.Parameter(torch.rand(1,
requires_grad=True,
dtype=torch.float))
in dtype i give dtype=float instead of torch.float.
when i trained it didn’t showed any error instead it accurately predicts the value rather than torch.float. why pytorch showing this behavior?
predicting value with dtype=torch.float

new
new699×533 17.7 KB

2

I don’t fully understand the issue, but note that Python’s float is a C double and PyTorch will thus use torch.float64 or torch.double as seen here:

weight = nn.Parameter(torch.rand(1, requires_grad = True, dtype=torch.float))
print(weight.dtype)
# torch.float32

weight = nn.Parameter(torch.rand(1, requires_grad = True, dtype=float))
print(weight.dtype)
# torch.float64
4

it was a logical error i train with wrong model