知识卡片
点积与批矩阵乘的高维规则不同
二维时dot与matmul通常给同样矩阵乘结果,但高维时其收缩及批次广播规则不同。使用前写清张量轴和预期shape,不能依据本章B=A的特例相等,推断任意B.dot(A.T)都等于A乘B转置。
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
原书出处 · 1 处
- · 查看原书依据
For the mathematicians in the audience, you may be more comfortable using the .dot() method rather than the np.matmul() function. Whenever you see .dot, you can imagine you are in math class, looking at the black dot you would see on a whiteboard between A and B:
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》附录C《附录C:向量与线性代数》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-c.html)
- 结论依据:摘录中关于点积与批矩阵乘的高维规则不同的定义、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 the mathematicians in the audience, you may be more comfortable using the .dot() method rather than the np.matmul() function. Whenever you see .dot, you can imagine you are in math class, looking at the black dot you would see on a whiteboard between A and B: