deeplearning with pytorch (五)

dc爱傲雪和技术2024-03-10 14:46

.view()方法在PyTorch中用于重塑张量。这里它被用来将单个样本的张量重塑成模型所期望的输入形状。具体地，1,1,28,28意味着创建一个新的张量，其中：

第一个1代表批次大小（batch size），这里为1，因为你只预测一个样本。
第二个1可能代表颜色通道的数量，这在处理灰度图像时常见，意味着每个像素只有一个颜色值。对于RGB图像，这个数字会是3。
28,28代表图像的高度和宽度，这是典型的MNIST手写数字数据集的维度。

python 复制代码

#graph the loss at epoch
train_losses = [tl.item() for tl in train_losses]
plt.plot(train_losses, label= "training loss")
plt.plot(test_losses, label="validation loss")
plt.title("loss at epoch")
plt.legend()
#graph the accuracy at the end of each epoch
plt.plot([t/600 for t in train_correct], label = "training accuracy")
plt.plot([t/100 for t in test_correct], label = "validation accuracy")
plt.title("accuracy at the end of each epoch")
plt.legend()
test_load_everything = DataLoader(test_data, batch_size= 10000, shuffle= False)
with torch.no_grad():
    correct = 0
    for X_test, y_test in test_load_everything:
        y_val = model(X_test)
        predicted = torch.max(y_val, 1)[1]
        correct += (predicted == y_test).sum()

# did for correct 
correct.item()/len(test_data) * 100
## Send New Image Thru The Model
# grab an image
test_data[4143] #tensor with an image in it ... at end ,it shows the label
# grab just the data 
test_data[4143][0]
#reshape it 
test_data[4143][0].reshape(28,28)
# show the image 

plt.imshow(test_data[4143][0].reshape(28,28))
# pass the image thru our model
model.eval()
with torch.no_grad():
    new_prediction = model(test_data[4143][0].view(1,1,28,28)) #batch size of 1,1 color channel, 28x28 image
    
# check the new prediction, get probabilities
new_prediction
new_prediction.argmax()

完整的py文件见GitHub - daichang01/neraual_network_learning at dev