more typos

This commit is contained in:
Roland Grinis 2021-07-13 20:00:36 +01:00
parent cc4da94646
commit 3b0032cb2f

View File

@ -139,7 +139,7 @@ class Loss(torch.nn.Module):
def forward(self, x): def forward(self, x):
return self.loss(x, self.target) return self.loss(x, self.target)
# Generated TorchScript modules and serialise # Generate TorchScript modules and serialise them
torch.jit.script(Data()).save('data.pt') torch.jit.script(Data()).save('data.pt')
torch.jit.script(Net()).save('net.pt') torch.jit.script(Net()).save('net.pt')
torch.jit.script(Loss(y_train)).save('loss.pt') torch.jit.script(Loss(y_train)).save('loss.pt')
@ -165,12 +165,12 @@ NoaFloat {
// Set the model in training mode // Set the model in training mode
netModule.train(true) netModule.train(true)
// Loss function set for training regime // Loss function for training
lossModule.setBuffer("target", yTrain) lossModule.setBuffer("target", yTrain)
// Compute the predictions // Compute the predictions
val yPred = netModule.forward(xTrain) val yPred = netModule.forward(xTrain)
// Compute the training error // Compute the training loss
val loss = lossModule.forward(yPred) val loss = lossModule.forward(yPred)
println(loss) println(loss)
@ -195,7 +195,7 @@ NoaFloat {
// Finally validate the model // Finally validate the model
// Compute the predictions for the validation features // Compute the predictions for the validation features
netModule.forwardAssign(xVal, yPred) netModule.forwardAssign(xVal, yPred)
// Set the loss for the true values // Set the loss for validation
lossModule.setBuffer("target", yVal) lossModule.setBuffer("target", yVal)
// Compute the loss on validation dataset // Compute the loss on validation dataset
lossModule.forwardAssign(yPred, loss) lossModule.forwardAssign(yPred, loss)