Chapter 5. Wrapping Up

In the previous chapter, we learned about another interface to TensorFlow and RNN models. This chapter will wrap up our look at TensorFlow, looking at how far we've come and where you can go from here. First, we'll review our progress on the font classification problem, then we'll briefly look at TensorFlow beyond deep learning and see where it will go in the future. At the end of the chapter, you will be familiar with the following concepts:

• Research evaluation

• A quick review of all the models

• The future of TensorFlow

• Some more TensorFlow projects

Let's now begin by looking at research evaluation in detail.

Research evaluation

In this section, we'll compare our models in the font classification problem. First, we should remind ourselves what the data looks like. Then, we'll inspect the simple logistic dense neural network and convolutional neural network models. You've come a long way in modeling with TensorFlow.

Before we move on from deep learning, however, let's look back and see how models compare on the font classification problem. First, let's look at the data again, so we don't lose sight of the problem. In fact, let's look at one image that includes all the letters and digits from every font, just to see what shapes we have:

# One look at a letter/digit from each font
# Best to reshape as one large array, then plot
all_letters = np.zeros([5*36,62*36])
for font in range(5):
    for letter in range(62):
        all_letters[font*36:(font+1)*36,
                letter*36:(letter+1)*36] = \
                train[9*(font*62 + letter)]

This would be a lot of subplots for Matplotlib to handle...

A quick review of all the models

Let's recap each of the models we built, to model these fonts and some of their strengths and weaknesses:

At a glance, recall that we slowly built up more complicated models and took into account the structure of the data to improve our accuracy.

The logistic regression model

First, we started with a simple logistic regression model:

This has 36x36 pixels plus 1 bias times 5 classes total weights, or 6,485 parameters that we need to train. After 1,000 training epochs, this model achieved about 40 percent validation accuracy. Your results may vary. This is relatively poor, but the model has some advantages.

Let's glance back at the code:

# These will be inputs
## Input pixels, flattened
x = tf.placeholder("float", [None, 1296])
## Known labels
y_ = tf.placeholder("float", [None,5])

# Variables
W = tf.Variable(tf.zeros([1296,5]))
b = tf.Variable(tf.zeros([5]))

# Just initialize
sess.run(tf.initialize_all_variables())

# Define model
y = tf.nn.softmax(tf.matmul...

The future of TensorFlow

In this section, we will observe how TensorFlow is changing, who is starting to use TensorFlow, and how you can make an impact contributing to it.

Since it was released in late 2015, TensorFlow has already seen several more releases:

TensorFlow is constantly being updated. Although it isn't an official Google product, it is also open source and hosted on GitHub. At the time of writing, TensorFlow is at release 1.2. The most recent release added distributed computing capabilities. These are beyond the scope of this book, but generally speaking, they allow computation across multiple GPUs on multiple machines for maximum parallelization. Under heavy development, more features are always just around the corner. TensorFlow is becoming more popular every day.

Several software companies have released machine learning frameworks recently, but TensorFlow stands out in adoption. Internally, Google is practicing what they preach. Their acclaimed DeepMind team has switched to...

Summary

In this chapter, we reviewed how we climbed from a humble logistic regression model to flying high with a deep convolutional neural network to classify fonts. We also discussed the future of TensorFlow. We finally recalled our TensorFlow models for font classification, reviewing their accuracy. We also took some time to discuss where TensorFlow is headed. Congratulations! You're now well-versed in TensorFlow. You've applied it to multiple research problems and models in this series, and learned how it's widely applicable.

The next step is to deploy TensorFlow in one of your own projects. Happy modeling!