1.Scalar equation derivation of vector
Scalar equation has two rules for vector derivation, one is denominator layout, and the other is molecular layout. After expansion, they are the same as denominator or molecular dimension respectively.
Take for example:
2.Vector equation derivation of vector
Now we change the function into a vector function, which has different scalar function components in different dimensions. If the denominator layout is adopted, the function can be expanded according to the denominator dimension, and then the content of each row can be calculated separately.
3. Important formula
The denominator layout is used as the transpose of the original matrix ; the molecular layout is equal to the original matrix.
Take for example:
Example : Linear regression cost function
For the square error cost function in machine learning, using the matrix derivation rule can greatly reduce the amount of calculation.
4. Chain Rule
The derivation rule of the vector is different from that of the scalar. The scalar satisfies the commutative law, so it can be decomposed in chain according to the four arithmetic operations, while the vector does not satisfy the commutative law, so it needs to be replaced.