PyTorch遍历历元，然后输出所有历元的最终值

问题描述

下面的代码计算MSE和MAE值，但我遇到一个问题，即在每个时期结束后，MAE和MSE的值不会得到store_MAE并不会存储MSE。它似乎仅使用最后一个纪元的值。任何想法我都需要在代码中执行以保存每个纪元的值，我希望这是有道理的。谢谢您的帮助

    global_step = 0
best_test_error = 10000
MAE_for_all_epochs = []
MSE_for_all_epochs = []
for epoch in range(4):
    print("Epoch %d" % epoch)
    model.train()
    for images,paths in tqdm(loader_train):
        images = images.to(device)
        targets = torch.tensor([metadata['count'][os.path.split(path)[-1]] for path in paths]) # B
        targets = targets.float().to(device)

        # forward pass:
        output = model(images) # B x 1 x 9 x 9 (analogous to a heatmap)
        preds = output.sum(dim=[1,2,3]) # predicted cell counts (vector of length B)
        
        # backward pass:
        loss = torch.mean((preds - targets)**2)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        # logging:
        count_error = torch.abs(preds - targets).mean()
        writer.add_scalar('train_loss',loss.item(),global_step=global_step)
        writer.add_scalar('train_count_error',count_error.item(),global_step=global_step)

        print("Step %d,loss=%f,count error=%f" % (global_step,count_error.item()))

        global_step += 1
    
    mean_test_error = 0
    model.eval()
    for images,paths in tqdm(loader_test):
        images = images.to(device)
        targets = torch.tensor([metadata['count'][os.path.split(path)[-1]] for path in paths]) # B
        targets = targets.float().to(device)

        # forward pass:
        output = model(images) # B x 1 x 9 x 9 (analogous to a heatmap)
        preds = output.sum(dim=[1,3]) # predicted cell counts (vector of length B)

        # logging:
        #error = torch.abs(preds - targets).sum().data
        #squared_error = ((preds - targets)*(preds - targets)).sum().data

        #runnning_mae += error
        #runnning_mse += squared_error

        loss = torch.mean((preds - targets)**2)
        count_error = torch.abs(preds - targets).mean()
        mean_test_error += count_error
        writer.add_scalar('test_loss',global_step=global_step)
        writer.add_scalar('test_count_error',global_step=global_step)
        
        global_step += 1
        #store_MAE = 0
        #store_MSE = 0

        mean_test_error = mean_test_error / len(loader_test)
        #store_MAE += mean_test_error
        MAE_for_all_epochs = np.append(MAE_for_all_epochs,mean_test_error)

        mse = math.sqrt(loss / len(loader_test))
        #store_MSE +=mse
        MSE_for_all_epochs = np.append(MSE_for_all_epochs,mse)

        print("Test count error: %f" % mean_test_error)
        print("MSE: %f" % mse)

    if mean_test_error < best_test_error:
        best_test_error = mean_test_error
        torch.save({'state_dict':model.state_dict(),'optimizer_state_dict':optimizer.state_dict(),'globalStep':global_step,'train_paths':dataset_train.files,'test_paths':dataset_test.files},checkpoint_path)

print("MAE Total: %f" % store_MAE)
print("MSE Total: %f" % store_MSE)
model_mae= MAE_for_all_epochs / epoch
model_mse= MSE_for_all_epochs / epoch
print("Model MAE: %f" % model_mae)
print("Model MSE: %f" % model_mse)

解决方法

np.append()将适用于您的情况，

#outside epochs loop
MAE_for_all_epochs = [] 

#inside loop
#replace this store_MAE with relevant variable

MAE_for_all_epochs = np.append(MAE_for_all_epochs,store_MAE) 

编辑：玩具示例：根据用法

import numpy as np
all_var = []

for e in range(1,10):
    var1 = np.random.random(1)
    all_var = np.append(all_var,var1)
    
print(all_var)
# output : [0.07660848 0.46824825 0.09432051 0.79462902 0.97798061 0.67299183 0.50996432 0.13084029 0.95100381]