YouTip

## PyTorch torch.corrcoef The `torch.corrcoef` function in PyTorch is used to compute the Pearson correlation coefficient matrix of an input tensor. This matrix measures the linear correlation between multiple variables, with values ranging from -1 to 1. --- ## Introduction to Pearson Correlation Coefficient The Pearson correlation coefficient ($r$) between two variables $X$ and $Y$ is calculated as: $$r = \frac{\text{cov}(X, Y)}{\sigma_X \sigma_Y}$$ Where: * $\text{cov}(X, Y)$ is the covariance of $X$ and $Y$. * $\sigma_X$ and $\sigma_Y$ are the standard deviations of $X$ and $Y$, respectively. The resulting matrix contains values where: * **1** indicates a perfect positive linear relationship. * **-1** indicates a perfect negative linear relationship. * **0** indicates no linear relationship. --- ## Syntax ```python torch.corrcoef(input) ``` ### Parameters | Parameter | Type | Description | | :--- | :--- | :--- | | `input` | *Tensor* | A 2D matrix containing multiple variables and observations, or a 1D vector representing a single variable. | ### Return Value * **Tensor**: A correlation coefficient matrix. If the input is a 2D tensor of shape `(M, N)` (where $M$ represents the number of variables and $N$ represents the number of observations), the output will be a symmetric matrix of shape `(M, M)`. --- ## Code Examples ### Example 1: Computing the Correlation Matrix of a 2D Tensor This example demonstrates how to compute the correlation matrix for a 2D tensor where each row represents a variable and each column represents an observation. ```python import torch # Create a 2D tensor (3 variables, 4 observations each) x = torch.tensor([[1.0, 2.0, 3.0, 4.0], [2.0, 3.0, 4.0, 5.0], [3.0, 4.0, 5.0, 6.0]]) print("Input shape:", x.shape) print("Input Tensor:") print(x) # Compute the correlation coefficient matrix corr = torch.corrcoef(x) print("\nCorrelation Coefficient Matrix:") print(corr) ``` **Output:** ```text Input shape: torch.Size([3, 4]) Input Tensor: tensor([[1., 2., 3., 4.], [2., 3., 4., 5.], [3., 4., 5., 6.]]) Correlation Coefficient Matrix: tensor([[1.0000, 1.0000, 1.0000], [1.0000, 1.0000, 1.0000], [1.0000, 1.0000, 1.0000]]) ``` *Note: Since all three rows increase linearly at the exact same rate, they are perfectly positively correlated, resulting in a matrix of all $1.0$.* --- ### Example 2: Correlation Between Two Specific Variables To find the correlation between two 1D tensors (variables), stack them into a 2D tensor of shape `(2, N)` before passing them to `torch.corrcoef`. ```python import torch # Define two variables a = torch.tensor([1.0, 2.0, 3.0, 4.0, 5.0]) b = torch.tensor([2.0, 4.0, 6.0, 8.0, 10.0]) # Stack them to create a (2, N) shape tensor data = torch.stack([a, b]) # Compute the correlation matrix corr2 = torch.corrcoef(data) print("Correlation Matrix:") print(corr2) # Extract the correlation coefficient between variable 'a' and 'b' r_ab = corr2[0, 1] print(f"\nCorrelation coefficient between a and b: {r_ab.item():.4f}") # Output is 1.0000, indicating a perfect positive linear correlation ``` **Output:** ```text Correlation Matrix: tensor([[1.0000, 1.0000], [1.0000, 1.0000]]) Correlation coefficient between a and b: 1.0000 ``` --- ## Important Considerations 1. **Data Type**: `torch.corrcoef` requires floating-point data types (e.g., `torch.float32` or `torch.float64`). If you pass integer tensors, PyTorch will raise a `RuntimeError`. Always cast your input using `.float()` or `.double()` if necessary. 2. **Input Dimensions**: The input tensor must be 1D or 2D. * A 1D tensor of shape `(N,)` is treated as a single variable with $N$ observations and will return a $1 \times 1$ matrix containing `[[1.0]]`. * A 2D tensor of shape `(M, N)` is treated as $M$ variables with $N$ observations. 3. **NaN Values**: If any variable has zero variance (i.e., all elements in a row are identical), the standard deviation will be zero. This leads to division by zero during calculation, resulting in `nan` (Not a Number) values in the output matrix.