45. Your First Neural Network in PyTorch
45.4 Predict with Sigmoid
Why. After training, convert logits to probabilities and decisions.
What. torch.sigmoid(logits) → probability. Threshold 0.5 → class 0/1 for a demo.
How.
Predict
- Set
model.eval(). - Use
torch.no_grad()so you do not build a graph. - Print probs and hard labels.
model.eval()
with torch.no_grad():
logits = model(X)
probs = torch.sigmoid(logits)
preds = (probs >= 0.5).float()
print("probs:\n", probs.round(decimals=3))
print("preds:\n", preds)
Ravindra Bagale's Tip
eval + no_grad = inference mode. Training mode vegla; habit solid theva.