W4_A1_Computing Activation functions in Linear Activation Backward

[Fatma_Shaban]

I could not understand the equations flow in this exercise, th activation_cache is unknown and the g(Z) is unknown.

[Kic]

Hi @Fatma_Shaban ,

At the start of the linear_activation_backward() function, there are comment lines explaining the input arguments for this function:

Arguments:
dA – post-activation gradient for current layer l
cache – tuple of values (linear_cache, activation_cache) we store for computing backward propagation efficiently
activation – the activation to be used in this layer, stored as a text string: “sigmoid” or "relu

As you can see from the ‘cache’ argument, it contains the activation_cache for the use of the formula’s calculation. The content of this cache is obtained during the forward pass through the network.

[Fatma_Shaban]

Thanks for your help. I have used cache[1] to access the activation_cache. and assign the backward function to be equal to (dA_prev, dW, db)
However, I got the following error while running the test, even that my function returns 3 values

[saifkhanengr]

Note that in linear_activation_backward function, we have linear_cache and activation_cache. You have to use the correct one when calling the relu_backward, sigmoid_backward, and linear_backward functions.

[Kic]

Hi @Fatma_Shaban ,

Firstly, two functions, sigmoid_backward() or relu_backward(), are provided for you to calculate the dz depending on which activation function is specified to use for the calculation. So there is no need to do np.multiply().

Secondly, ‘cache’ consists of ‘linear_cache’ and ‘activation_cache’. When calling linear_backward(), the parameters pass to the function should be dz and the linear_cache, as it contains the activation output ‘A’ of previous layer, W and b, the weight matrix and bias of current layer respectively. These are the information needed when back propagating through the network. If you refer to linear_forward() funciton, you can see what the linear cache contains.

[Fatma_Shaban]

Thanks for the clear explaniation.

[Fatma_Shaban]

In the “L_model_backward”, I made the function. but, I got only one test passed and two test not passed. How could I d

ebuge my function, could I access the test files or the data entry to the function?

[saifkhanengr]

There are a lot of threads, like this one, in this forum. Please use the search bar to search the related posts by typing your error or any other keyword, before posting yours.

Moreover, you can check the test files and other sources for utility functions. Please read this.