YouTipPytorch Torch Neg
## PyTorch torch.neg() Function
The `torch.neg()` function in PyTorch is used to perform element-wise negation on an input tensor. It multiplies every element in the input tensor by $-1$, effectively flipping the sign of each numerical value.
---
### Function Definition
```python
torch.neg(input, *, out=None) -> Tensor
```
Alternatively, you can use the alias `torch.negative()` or the in-place version `torch.neg_()`.
### Parameters
| Parameter | Type | Description |
| :--- | :--- | :--- |
| `input` | Tensor | The input tensor containing the numerical values to be negated. |
| `out` | Tensor (Optional) | The output tensor where the result will be written. Defaults to `None`. |
### Return Value
* Returns a new tensor with the same shape and data type as `input`, where each element has its sign inverted.
---
## Usage Examples
### Example 1: Basic Element-Wise Negation
This example demonstrates how to negate a 1D tensor containing positive floating-point numbers.
```python
import torch
# Create a 1D tensor
x = torch.tensor([1.0, 2.0, 3.0])
# Perform element-wise negation
result = torch.neg(x)
print("Original Tensor:")
print(x)
print("\nNegated Tensor:")
print(result)
```
**Output:**
```text
Original Tensor:
tensor([1., 2., 3.])
Negated Tensor:
tensor([-1., -2., -3.])
```
---
### Example 2: Handling Mixed Signs and Zero
`torch.neg()` correctly handles positive numbers, negative numbers, and zero.
```python
import torch
# Create a tensor with positive, negative, and zero values
x = torch.tensor([-5, 0, 12], dtype=torch.int32)
# Negate the tensor
result = torch.neg(x)
print("Original Tensor:", x)
print("Negated Tensor: ", result)
```
**Output:**
```text
Original Tensor: tensor([-5, 0, 12], dtype=torch.int32)
Negated Tensor: tensor([ 5, 0, -12], dtype=torch.int32)
```
---
### Example 3: In-Place Negation (`neg_`) and Operator Alias
If you want to modify the tensor in-place without allocating new memory, you can use `neg_()`. You can also use the unary minus operator `-` as a shorthand.
```python
import torch
x = torch.tensor([[1.0, -2.0], [3.0, -4.0]])
# Method 1: Using the unary minus operator
neg_operator = -x
# Method 2: In-place negation
x.neg_()
print("Negated using operator (-x):")
print(neg_operator)
print("\nOriginal tensor modified in-place (x.neg_()):")
print(x)
```
**Output:**
```text
Negated using operator (-x):
tensor([[-1., 2.],
[-3., 4.]])
Original tensor modified in-place (x.neg_()):
tensor([[-1., 2.],
[-3., 4.]])
```
---
## Considerations
1. **Data Types**: `torch.neg()` supports integer, floating-point, and complex data types. For complex numbers, it negates both the real and imaginary parts.
2. **Boolean Tensors**: Attempting to use `torch.neg()` on a boolean (`torch.bool`) tensor is not supported in newer PyTorch versions and will raise a `RuntimeError`. For logical negation, use `torch.logical_not()` or the bitwise NOT operator `~` instead.
3. **Memory Efficiency**: For large tensors, using the in-place variant `input.neg_()` helps optimize memory usage by avoiding the allocation of a new tensor.