def linear_activation_backward(dA, cache, activation):

    linear_cache, activation_cache = cache
    
    if activation == "relu":
        dZ = relu_backward(dA, activation_cache) 
        dA_prev, dW, db = linear_backward(dZ, linear_cache) # calculates updates of parameter in each layer
        
    if activation == "sigmoid":
        dZ = sigmoid_backward(dA, activation_cache)
        dA_prev, dW, db = linear_backward(dZ, linear_cache)
        
    if activation == "tanh":
        dZ = tanh_backward(dA, activation_cache)
        dA_prev, dW, db = linear_backward(dZ, linear_cache) 
    
    return dA_prev, dW, db