YouTip

## PyTorch `torch.equal` Function The `torch.equal` function in PyTorch is a utility used to check if two tensors are completely identical. It returns a single boolean value: `True` if both tensors have the exact same shape and elements, and `False` otherwise. --- ### Function Definition ```python torch.equal(input, other) -> bool ``` #### Parameters: * **`input`** *(Tensor)*: The first tensor to compare. * **`other`** *(Tensor)*: The second tensor to compare. #### Returns: * **`bool`**: `True` if the tensors are equal in both shape and content, `False` otherwise. --- ## Code Examples The following example demonstrates how to use `torch.equal` to compare identical and non-identical tensors. ```python import torch # Create two identical tensors x = torch.tensor([1, 2, 3]) y = torch.tensor([1, 2, 3]) # Check if they are equal result = torch.equal(x, y) print(f"Are x and y equal? {result}") # Create two different tensors (different elements) a = torch.tensor([1, 2, 4]) b = torch.tensor([1, 2, 3]) result2 = torch.equal(a, b) print(f"Are a and b equal? {result2}") # Create two tensors with different shapes c = torch.tensor([[1, 2, 3]]) result3 = torch.equal(x, c) print(f"Are x and c equal? {result3}") ``` ### Output ```text Are x and y equal? True Are a and b equal? False Are x and c equal? False ``` --- ## Key Considerations When using `torch.equal`, keep the following behaviors in mind: ### 1. No Broadcasting Unlike element-wise comparison operators (such as `==` or `torch.eq`), `torch.equal` **does not** perform broadcasting. If the shapes of the two tensors do not match exactly, the function immediately returns `False` without raising an error. ### 2. `torch.equal` vs. `torch.eq` It is important to distinguish between `torch.equal` and `torch.eq` (or the `==` operator): * **`torch.equal(tensor1, tensor2)`**: Returns a **single boolean** (`True` or `False`). It checks if the entire contents and shapes are identical. * **`torch.eq(tensor1, tensor2)`**: Returns a **boolean tensor** of the same shape (after broadcasting), representing the element-wise comparison results. #### Comparison Example: ```python import torch tensor1 = torch.tensor([1, 2, 3]) tensor2 = torch.tensor([1, 5, 3]) # Element-wise comparison print(torch.eq(tensor1, tensor2)) # Output: tensor([ True, False, True]) # Global comparison print(torch.equal(tensor1, tensor2)) # Output: False ``` ### 3. Floating-Point Precision When comparing floating-point tensors (`torch.float32`, `torch.float64`), tiny numerical precision errors can cause `torch.equal` to return `False` even if the values are mathematically expected to be equal. For comparing floating-point tensors with a tolerance margin, use **`torch.allclose`** instead: ```python # Using torch.allclose for floating-point comparisons float_tensor1 = torch.tensor([1.0, 2.0]) float_tensor2 = torch.tensor([1.0000001, 2.0]) print(torch.equal(float_tensor1, float_tensor2)) # Output: False print(torch.allclose(float_tensor1, float_tensor2)) # Output: True ```