Autograd as a dependency graph (step 7/7) · pytorch tensors and autograd

Checkpoint

One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.

Checkpoint. Prove training works by watching the loss fall. Write train(w, x, y, lr) that records the loss before training, runs two gradient-descent steps on pred = w * x, and returns:

loss 64.00 -> 8.29

(loss before -> loss after 2 steps, each to 2 decimals). Loss is (w*x - y)**2; each step uses gradient 2*(w*x - y)*x and update w = w - lr*grad.

Checkpoint

One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.

Checkpoint. Prove training works by watching the loss fall. Write train(w, x, y, lr) that records the loss before training, runs two gradient-descent steps on pred = w * x, and returns:

loss 64.00 -> 8.29

(loss before -> loss after 2 steps, each to 2 decimals). Loss is (w*x - y)**2; each step uses gradient 2*(w*x - y)*x and update w = w - lr*grad.

full-screen editor opens — close anytime to keep reading.

Autograd as a dependency graph — step 7 of 7

Autograd as a dependency graph — step 7 of 7