PyTorch Tensor Operations Cheat Sheet

PyTorch Tensor Operations Cheat Sheet

Creating Tensors

1. Create a tensor:

   Copy

   tensor = torch.tensor([1, 2, 3])

2. Create a zero tensor:

   Copy

   zeros = torch.zeros(3, 3)

3. Create an identity matrix:

   Copy

   identity = torch.eye(3)

4. Create a tensor with random values:

   Copy

   rand_tensor = torch.rand(3, 3)

5. Create a tensor with specific range:

   Copy

   range_tensor = torch.arange(0, 10, 2)

Indexing and Slicing

1. Index a tensor:

   Copy

   element = tensor[0]

2. Slice a tensor:

   Copy

   slice = tensor[1:3]

3. Access rows/columns:

   Copy

   row = tensor[0, :]
   col = tensor[:, 0]

Shape Manipulation

1. Get tensor shape:

   Copy

   shape = tensor.shape

2. Reshape tensor:

   Copy

   reshaped = tensor.view(2, 2)

3. Flatten tensor:

   Copy

   flattened = tensor.flatten()

Arithmetic Operations

1. Addition:

   Copy

   result = tensor1 + tensor2

2. Subtraction:

   Copy

   result = tensor1 - tensor2

3. Multiplication:

   Copy

   result = tensor1 * tensor2

4. Division:

   Copy

   result = tensor1 / tensor2

Element-wise Operations

1. Element-wise square root:

   Copy

   sqrt = torch.sqrt(tensor)

2. Element-wise exponentiation:

   Copy

   exp = torch.exp(tensor)

3. Element-wise logarithm:

   Copy

   log = torch.log(tensor)

4. Element-wise absolute value:

   Copy

   abs = torch.abs(tensor)

Reduction Operations

1. Sum of all elements:

   Copy

   sum = tensor.sum()

2. Mean of all elements:

   Copy

   mean = tensor.mean()

3. Maximum element:

   Copy

   max_val = tensor.max()

4. Minimum element:

   Copy

   min_val = tensor.min()

Matrix Operations

1. Matrix multiplication:

   Copy

   matmul = torch.matmul(matrix1, matrix2)

2. Element-wise matrix multiplication:

   Copy

   element_wise = matrix1 * matrix2

3. Matrix transposition:

   Copy

   transposed = matrix.T

4. Matrix inverse:

   Copy

   inverse = torch.inverse(matrix)

Broadcasting

1. Expand tensor dimensions:

   Copy

   expanded = tensor.unsqueeze(0)

2. Broadcast tensors for element-wise operations:

   Copy

   result = tensor1 + tensor2

GPU Acceleration

1. Move tensor to GPU:

   Copy

   tensor.to('cuda')

2. Perform operations on GPU:

   Copy

   result = tensor1 + tensor2

Conversion

1. Convert to NumPy array:

   Copy

   numpy_array = tensor.numpy()

2. Convert to Python scalar:

   Copy

   scalar = tensor.item()

Concatenation and Stacking

1. Concatenate tensors along a dimension:

   Copy

   concatenated = torch.cat([tensor1, tensor2], dim=0)

2. Stack tensors along a new dimension:

   Copy

   stacked = torch.stack([tensor1, tensor2], dim=0)

Element-wise Comparison

1. Element-wise equality:

   Copy

   equal = torch.eq(tensor1, tensor2)

2. Element-wise greater than:

   Copy

   gt = torch.gt(tensor1, tensor2)

3. Element-wise less than or equal:

   Copy

   le = torch.le(tensor1, tensor2)

Element-wise Logic Operations

1. Element-wise logical AND:

   Copy

   logical_and = tensor1 & tensor2

2. Element-wise logical OR:

   Copy

   logical_or = tensor1 | tensor2

3. Element-wise logical NOT:

   Copy

   logical_not = ~tensor

Random Number Generation

1. Generate random tensor from uniform distribution:

   Copy

   uniform = torch.rand(3, 3)

2. Generate random tensor from normal distribution:

   Copy

   normal = torch.randn(3, 3)

Element-wise Round Operations

1. Round elements to nearest integer:

   Copy

   rounded = torch.round(tensor)

2. Floor elements to nearest integer:

   Copy

   floored = torch.floor(tensor)

3. Ceiling elements to nearest integer:

   Copy

   ceiled = torch.ceil(tensor)

This comprehensive cheat sheet covers a wide range of tensor operations in PyTorch. Experiment with these operations to enhance your understanding and proficiency in working with tensors. For more advanced operations and details, refer to the official PyTorch documentation.

Feel free to use this cheat sheet as a reference guide for tensor operations in PyTorch!