ReLU Issue

[SuperPauly]

I've tried 2 ways, both still produce a torch.Tensor type and output the correct values. But I still only get 1/4 correct.

Here is the 2 ways:

def relu(x: torch.Tensor) -> torch.Tensor:
    #return torch.as_tensor(list(map(lambda n: n if (n >= 0.0) else -0.0, x)))
    return torch.as_tensor([n if n >= 0.0 else -0.0 for n in x])

The error message I get is confusing, I think it means it's not compatible with multidimensional tensors? Or does this error mean something else?

Why does it say element 1 does not require gradient function and has no gradient function. Yet this throws an error? This error shouldn't exist because it cancels itself out in the sale sentence.

🧪 Testing: Implement ReLU (Easy)
──────────────────────────────────────────────────
  ✅ [1/4] Basic values (1.6ms)
  💥 [2/4] 2-D tensor
     RuntimeError: Boolean value of Tensor with more than one value is ambiguous
         return torch.as_tensor([n if n >= 0.0 else -0.0 for n in x])
                                 ^^^^^^^^
RuntimeError: Boolean value of Tensor with more than one value is ambiguous
  💥 [3/4] Gradient check
     RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn
         return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn
  💥 [4/4] Performance
     RuntimeError: Boolean value of Tensor with more than one value is ambiguous
         return torch.as_tensor([n if n >= 0.0 else -0.0 for n in x])
                                 ^^^^^^^^
RuntimeError: Boolean value of Tensor with more than one value is ambiguous
──────────────────────────────────────────────────
  📊 1/4 tests passed.
  Keep going! Use hint("relu") if you're stuck.

So I opened the interpreter and typed in the solution to the Relu challenge manually, with fixed arguments. That then gave me this error:

Surely there's more than one way to do this challenge without copying the solution verbatim.

Another thing I find confusing is on the solution it gets a green tick for being a 2-D Tensor, but it's not. It's a a 1-D Tensor?