W4_A1_UnboundError_variable A referenced before assignment

Course Q&A Deep Learning Specialization Neural Networks and Deep Learning
1

UnboundLocalError Traceback (most recent call last)
in
----> 1 parameters, costs = two_layer_model(train_x, train_y, layers_dims = (n_x, n_h, n_y), num_iterations = 2, print_cost=False)
2
3 print("Cost after first iteration: " + str(costs[0]))
4
5 two_layer_model_test(two_layer_model)

in two_layer_model(X, Y, layers_dims, learning_rate, num_iterations, print_cost)
48 # YOUR CODE STARTS HERE
49
—> 50 A1, cache1 = linear_activation_forward(X, W1, b1, activation = relu)
51 A2, cache2 = linear_activation_forward(A1, W2, b2, activation = sigmoid)
52

~/work/release/W4A2/dnn_app_utils_v3.py in linear_activation_forward(A_prev, W, b, activation)
209 A, activation_cache = relu(Z)
210
→ 211 assert (A.shape == (W.shape[0], A_prev.shape[1]))
212 cache = (linear_cache, activation_cache)
213

UnboundLocalError: local variable ‘A’ referenced before assignment

I think “linear_activation_forward” have some error in this assignment.
Could you help me solve this problem?
I tried several things… but I couldn’t solve this error.

2

Hello @yusinjeong5859 씨 :slight_smile:

This error is very tricky. I followed the error traceback to the definition of linear_activation_forward which is inside dnn_app_utils_v3.py (avaiable to you as well). If we look at its code:

image

then you can see that the function expected either "sigmoid" or "relu" as the value for activiation. If we have provided anything other than those, then A will never be assigned any value. Consequently, as it ran assert (A.shape == (W.shape[0], A_prev.shape[1])), we would have the local variable ‘A’ referenced before assignment error.

My suggestion is for you to go through the error traceback, check what you had supplied to linear_activation_forward for the activation parameter, and see if you can fix it.

Cheers,
Raymond

3

Hi,
I used the below statement while defining the activation function parameter in linear_activation_forward and linear_activation_backward utility funtion. Please use activation = ‘relu’ or ‘sigmoid’ while passing the parameter. it will resolve the issue.
e.g. A1, cache1 = linear_activation_forward(X, W1, b1, activation = “relu”)
or dA1, dW2, db2 = linear_activation_backward(dA2, cache2, activation = ‘sigmoid’)

4

It would be better to share your full error, rather than your code.

5

Hi,
The full error is mentioned in the message by yusinjeong5859 at the very beginning of the thread.

6

Hello, PreetamKumar.

Paul sir has already explained that in his reply.

the function expected either "sigmoid" or "relu" as the value for activiation . If we have provided anything other than those, then A will never be assigned any value. Consequently, as it ran assert (A.shape == (W.shape[0], A_prev.shape[1])) , we would have the local variable ‘A’ referenced before assignment error.

Please look at the activation function carefully, you are providing to the case.