def model_loss(model, dataset, train=False, optimizer=None):
performance=L1Loss()
score_metric=R2Score()
avg_loss=0
avg_score=0
count=0
for input, output in iter(dataset):
prediction=model.feed(input)
loss=performance(prediction,output)
score_metric.update([prediction,output])
score=score_metric.compute()
if(train):
optimizer.zero_grad()
loss.backward()
optimizer.step()
avg_loss+=loss.item()
avg_score+=score
count+=1
return avg_loss / count, avg_score / count
def model_loss(model, dataset, train = False, optimizer = None):
# Cycle through the batches and get the average L1 loss
performance = L1Loss()
score_metric = R2Score()
avg_loss = 0
avg_score = 0
count = 0
for input, output in iter(dataset):
# Get the model's predictions for the training dataset
predictions = model.feed(input)
# Get the model's loss
loss = performance(predictions, output)
# Get the model's R^2 score
score_metric.update([predictions, output])
score = score_metric.compute()
if(train):
# Clear any errors so they don't cummulate
optimizer.zero_grad()
# Compute the gradients for our optimizer
loss.backward()
# Use the optimizer to update the model's parameters based on the gradients
optimizer.step()
# Store the loss and update the counter
avg_loss += loss.item()
avg_score += score
count += 1
return avg_loss / count, avg_score / count
def model_loss(model, dataset, train=False, optimizer=None):
performance = L1Loss()
score_metric = R2Score()
avg_loss = 0
avg_score = 0
avg_mse = 0
count = 0
for input, output in iter(dataset):
predictions = model.feed(input)
loss = performance(predictions, output)
score_metric.update([predictions, output])
score = score_metric.compute()
mse = mean_squared_error(output.cpu(),
predictions.cpu().detach().numpy())
if (train):
optimizer.zero_grad()
loss.backward()
optimizer.step()
avg_loss += loss.item()
avg_score += score
count += 1
avg_mse += mse
return avg_loss / count, avg_score / count, avg_mse / count
def model_loss(model,dataset,train = False, optimizer = None):
performance = L1Loss()
score_metric = R2Score()
avg_loss = 0
avg_score = 0
count = 0
for input,output in iter(dataset):
predictions = model.feed(input)
loss = performance(predictions,output)
score_metric.update([predictions,output])
score = score_metric.compute()
if(train):
#clear any errors so they dont cummulate
optimizer.zero_grad()
loss.backward()
#use the optimizer to update the models parameters based on the gradients
optimizer.step()
avg_loss += loss.item()
avg_score += score
count += 1
return avg_loss/count, avg_score/count
def model_loss(model, dataset, train=False, optimizer= None):
#cycle through batches and get avg L1loss
performance=L1Loss()
score_metric=R2Score()
avg_loss=0
avg_score=0
count=0
for input, output in iter(dataset):
# get the model predictions for training dataset
predictions=model.feed(input)
#get the model loss
loss= performance(predictions, output)
#get the model r2 score
score_metric.update([predictions,output])
score= score_metric.compute()
if(train):
#clear any errors so that they dont commulate
optimizer.zero_grad()
#compute gradiennts for our optimizer
loss.backward()
# use the optimizer to update the model parameters based on gradients
optimizer.step()
avg_loss +=loss.item()
avg_score +=score
count +=1
return avg_loss / count, avg_score/count