Revisiting notMNIST

Let's start our effort incrementally by trying the technical changes on the notMNIST dataset we used in Chapter 2, Your First Classifier. You can write the code as you go through the chapter, or work on the book's repository at:

https://github.com/mlwithtf/mlwithtf/blob/master/chapter_02/training.py.

We will begin with the following imports:

    import sys, os 
    import tensorflow as tf 
    sys.path.append(os.path.realpath('../..')) 
    from data_utils import * 
    from logmanager import * 
    import math

There are not many substantial changes here. The real horsepower is already imported with the tensorflow package. You'll notice that we reuse our data_utils work from before. However, we'll need some changes there.

The only difference from before is the math package, which we will use for ancillary math functions, such as ceiling.

Program configurations

Now, let's look at our old program configurations, which are as follows:

    batch_size = 128 
    num_steps = 10000...