This is the first python Lab on TensorFlow, and it kinda too much to grasp on first TF lesson.

linear_layer = tf.keras.layers.Dense(units=1, activation = 'linear', )
→ Pls explain what and why Dense and what are units. What are we doing here, and the purpose of choosing Linear.

a1 = linear_layer(X_train[0].reshape(1,1))
thought X_train was already in 1,1 shape

linear_layer.set_weights([set_w, set_b])
→ THis is confusing. Pls explain why we are passing an array of weights

→ Am lost here as to why we are doing what we are doing

model = Sequential(
[
tf.keras.layers.Dense(1, input_dim=1, activation = 'sigmoid', name='L1')

→ Why this Sequential was not done for Linear one ?

Let me know if I am asking too many questions.
This is the first lab, I am confused as to its objective , and how and why the path is chosen to achieve this Neural nw

I will connect the relevant content in the lab to your question with some additions. It will take some time to think and understand, but it is also a good chance to do so

To start, please be reminded that the purpose of the lab is to introduce the idea of neural network, neural network layer, and neuron. A neural network can contain any number of layers, and a layer can contain any number of neurons.

In particular, a network with one layer that contains only one neuron, which is what the lab is showing, is actually a linear regression formula. Moreover, if we add a sigmoid activation to that only neuron, it becomes a logistic regression formula.

In other words, the lab wants to connect our existing knowledge about linear/logistic regression with neurons!

Before you continue, I recommend you to read the lab’s introductory paragraph, which states the objective again.

With my starting comment in mind, see if you can make sense the following graph:

Simply not necessary. It is good for us to see how one thing can be achieved in different ways. Remember, one of the lab’s jobs is to show us possibilities. Now, you know we can do it in two different ways, and you can verify yourself the other way for the case of logistic regression

Hey @rmwkwok , so X_train[0] is [1.] and when I reshape it I get [[1.]]
My question is why do we use the latter, is it like the example we are taking might work without reshaping but would need to reshape once we have a decent size of inputs? or is there any other reason?