Community

ML Tutorial 46-47 TensorFlow Question

Hello Harrison,
    I really like the tutorials you filmed for Python, and I learn a lot from it.
    Recently, starting to learn Neural Network, and for TensorFlow, I am wondering how does it work in industrial environment. All these functions are encapsulated so well that only a couple of line codes can train and predict data. This is Question 1.
    For second question, it is about the codes in Tutorial 46,47. For number of nodes in each layer, I decreased n_number_layer3 to 50, then the accuracy goes low to 10% - 30%. But when I change like

n_nodes_l1 = 50 n_nodes_l2 = 500 n_nodes_l3 = 500 

or

n_nodes_l1 = 500 n_nodes_l2 = 50 n_nodes_l3 = 500 

,
the accuracy is still above 90%. Why does the last layer have so much influence on the output? If it does have much influence, then why are there 3 layers in our neural network, I mean, 2 layers totally works for our model?
    Thank you!