TensorFlow Basics

Working Example (Linear Regression)

• Linear regression formula: y = mx + b
• Linear regression is one of the simple machine learning models.
• The x value will be given as an input, and the model will return the value of y depending on the calculated optimal values for m and b.
• The goal of training the model is to optimize the values for m and b that best fit the given training data (x and y).
• The work will start by guess values for m and b, then the training will optimize them to minimize the loss.
• Loss is the difference between the actual y value and the line itself.

General

• Constant nodes doesn’t hold its value until the session actually started.
• Tensor Shape: is the size of data where [5.0] has shape (1,) and [ [3.0], [2.0] ] has shape (2,1).
• Variable nodes needs a global variables initializer to initialize all the variables before it being used during session running.
• Assigning a new value for a variable needs to be taken during session running.

Node Types

• Constant Node: holds a constant value. The value can be float number, integer, or even a string.
• Operator Node: holds the type of the operator as well as the operands nodes.
• Placeholder: holds no value until we pass the value during session running.
• Variable Node: holds an initial value but it can be re-assigned by another one during the session. Also, it has some extra functions over the constant nodes.