知识卡片
多数据集加载器应累积而非覆盖
循环每个split时重新赋值整个dataloader字典,只会保留最后一次的键。应对各键保存对应加载器,再核对train、validation和test入口;否则表面遍历过所有split也不代表它们仍可访问。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 用词嵌入推理
原书出处 · 1 处
- · 查看原书依据
for k in dset.keys(): dataloader = { k: DataLoader( Word2VecDataset( dset[k], vocab_size=len(vocab)), batch_size=BATCH_SIZE, shuffle=True, num_workers=CPU_CORES - 1) } You need a one-hot encoder to turn your word pairs into one-hot vector pairs: def one_hot_encode(input_id, size): vec = torch.zeros(size).float() vec[input_id] = 1.0 return vec To dispel some of the magic of the examples you saw earlier, you’ll train the network from scratch, just as you did
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第06章《用词嵌入推理》(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-6.html)
- 结论依据:摘录中关于多数据集加载器应累积而非覆盖的定义、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 k in dset.keys(): dataloader = { k: DataLoader( Word2VecDataset( dset[k], vocab_size=len(vocab)), batch_size=BATCH_SIZE, shuffle=True, num_workers=CPU_CORES - 1) } You need a one-hot encoder to turn your word pairs into one-hot vector pairs: def one_hot_encode(input_id, size): vec = torch.zeros(size).float() vec[input_id] = 1.0 return vec To dispel some of the magic of the examples you saw earlier, you’ll train the network from scratch, just as you did