YouTipPytorch Torch Range
## PyTorch torch.range
The `torch.range` function in PyTorch is used to create a 1D tensor containing a sequence of values spaced at equal intervals within a specified range.
Unlike many Python range functions, `torch.range` **includes** the `end` value in the generated sequence (it is a closed interval $[start, end]$).
> **Deprecation Warning:** `torch.range` is deprecated in modern versions of PyTorch and may be removed in future releases. It is highly recommended to use `torch.arange` instead, which uses a half-open interval $[start, end)$ to match standard Python conventions.
---
### Syntax
```python
torch.range(start=0, end, step=1, out=None, dtype=None, layout=torch.strided, device=None, requires_grad=False)
```
### Parameter Definitions
| Parameter | Type | Description |
| :--- | :--- | :--- |
| `start` | *float/int* | The starting value of the sequence. Default: `0`. |
| `end` | *float/int* | The ending value of the sequence (inclusive). |
| `step` | *float/int* | The spacing between each pair of adjacent values. Default: `1`. |
| `out` | *Tensor, optional* | The alternative output tensor to store the result. |
| `dtype` | *torch.dtype, optional* | The desired data type of the returned tensor. If not specified, infers the type from other input arguments. |
| `layout` | *torch.layout, optional* | The desired layout of the returned Tensor. Default: `torch.strided`. |
| `device` | *torch.device, optional* | The desired device of the returned tensor (e.g., `'cpu'`, `'cuda'`). |
| `requires_grad` | *bool, optional* | If autograd should record operations on the returned tensor. Default: `False`. |
---
### Code Examples
#### Example 1: Basic Usage (Default Start and Step)
If you only provide one positional argument, it is treated as the `end` value. The sequence starts at `0` with a step size of `1`.
```python
import torch
# Create a sequence from 0 to 5 (inclusive)
x = torch.range(5.0)
print(x)
# Output: tensor([0., 1., 2., 3., 4., 5.])
```
#### Example 2: Specifying Start, End, and Step
You can customize the starting point and the step interval.
```python
import torch
# Create a sequence from 1 to 10 with a step size of 2 (inclusive of 10 if reachable)
y = torch.range(1, 10, 2)
print(y)
# Output: tensor([1., 3., 5., 7., 9.])
```
#### Example 3: Specifying Device and Data Type
You can explicitly define the tensor's data type and place it on a specific device (like a GPU).
```python
import torch
# Create an integer sequence on the GPU (if CUDA is available)
device = "cuda" if torch.cuda.is_available() else "cpu"
z = torch.range(start=1, end=5, step=1, dtype=torch.int32, device=device)
print(z)
# Output: tensor([1, 2, 3, 4, 5], device='cuda:0', dtype=torch.int32)
```
---
### Key Considerations & Best Practices
1. **Inclusive End Value:** `torch.range(start, end)` includes the `end` value in the output tensor. For example, `torch.range(1, 3)` yields `[1, 2, 3]`.
2. **Deprecation Warning:** Because the inclusive behavior of `torch.range` is inconsistent with Python's built-in `range()` and NumPy's `arange()`, PyTorch has deprecated this function.
3. **Recommended Alternative (`torch.arange`):** You should use `torch.arange` for modern PyTorch development. It excludes the `end` value (half-open interval $[start, end)$).
```python
# Deprecated (Inclusive)
torch.range(1, 3) # Returns: tensor([1., 2., 3.])
# Recommended (Exclusive)
torch.arange(1, 3) # Returns: tensor([1, 2])
```