๐ฅ PyTorch Tensors
PyTorch tensors are similar to NumPy arrays but can run on GPUs for accelerated computing.
๐ก Quick Tip:
Mastering this concept will significantly boost your Python data science skills!
๐ป Code Example:
import torch import torch.nn as nn import torch.optim as optim print("PyTorch version:", torch.__version__) # 1. Tensor basics x = torch.tensor([[1.0, 2.0], [3.0, 4.0]]) print("Shape:", x.shape, "| Dtype:", x.dtype) print("Matrix power:\n", x @ x.T) # Matrix multiply # 2. Autograd โ automatic gradient computation a = torch.tensor(3.0, requires_grad=True) b = torch.tensor(4.0, requires_grad=True) loss = a**2 + 2*b + a*b # f(a,b) = aยฒ + 2b + ab loss.backward() print(f"\nGradient df/da={a.grad:.1f}, df/db={b.grad:.1f}") # df/da = 2a + b = 10, df/db = 2 + a = 5 # 3. Build a simple linear regression model torch.manual_seed(42) X = torch.rand(100, 1) * 10 y = 3.5 * X + 2.0 + torch.randn(100, 1) model = nn.Linear(1, 1) optim_ = optim.SGD(model.parameters(), lr=0.01) loss_fn = nn.MSELoss() # 4. Training loop for epoch in range(200): y_pred = model(X) loss = loss_fn(y_pred, y) optim_.zero_grad() loss.backward() optim_.step() w = model.weight.item() b_val = model.bias.item() print(f"\nLearned โ weight: {w:.4f} | bias: {b_val:.4f}") print(f"Expected โ weight: 3.5000 | bias: 2.0000") # 5. Inference new_x = torch.tensor([[7.5]]) with torch.no_grad(): pred = model(new_x).item() print(f"\nPrediction for x=7.5: {pred:.2f}")
Keep exploring and happy coding! ๐ป
