知识卡片

普通读书笔记卡 · 6694.f · 生成模型:GPT-6

评价过程要区分参数更新前后

理解等级:中级

示例先计算训练输出与损失,再step,接着算测试输出;因此同一行日志两项来自不同参数时点。若重复运行训练单元又不重置模型和优化器,轮数还会累计;需要明确起点和记录时机才能比较曲线。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      for epoch in range(num_epochs): optimizer.zero_grad() #1 outputs = model(X_train) loss_train = loss_func_train(outputs, y_train) loss_train.backward() epoch_loss_train = loss_train.item() optimizer.step() outputs_test = model(X_test) loss_test = loss_func_test(outputs_test, y_test).item() accuracy_test = measure_binary_accuracy(outputs_test, y_test) if epoch % 20 == 19: #2 print(f'Epoch {epoch}:' f' loss_train/test: {loss_train.item():.4f}/{loss_test:.4f},' f' accuracy_test: {accuracy_test:.4f}') Epoch 19: loss_train/test: 80.1816/75.3989, accuracy_test: 0.4275 Epoch 39: loss_train/test: 75.0748/74.4430, accuracy_test: 0.5933

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第05章《神经网络基础》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-5.html) - 结论依据:摘录中关于评价过程要区分参数更新前后的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:for epoch in range(num_epochs): optimizer.zero_grad() #1 outputs = model(X_train) loss_train = loss_func_train(outputs, y_train) loss_train.backward() epoch_loss_train = loss_train.item() optimizer.step() outputs_test = model(X_test) loss_test = loss_func_test(outputs_test, y_test).item() accuracy_test = measure_binary_accuracy(outputs_test, y_test) if epoch % 20 == 19: #2 print(f'Epoch {epoch}:' f' loss_train/test: {loss_train.item():.4f}/{loss_test:.4f},' f' accuracy_test: {accuracy_test:.4f}') Epoch 19: loss_train/test: 80.1816/75.3989, accuracy_test: 0.4275 Epoch 39: loss_train/test: 75.0748/74.4430, accuracy_test: 0.5933