Difficulty: beginner
Estimated Time: 10 minutes

This scenario is the continuation of the MNIST for beginner one and shows how to use TensorFlow to build deep convolutional network. The training is performed on the MNIST dataset that is considered a Hello world for the deep learning examples. The content is based on the official TensorFlow tutorial.

To take the most of this course you should know how to program in Python or other language that would allow you to understand Python syntax.

Congratulations!

You've completed TensorFlow: MNIST for experts scenario.

You've learned how to perform the following tasks:

  • Create dense, convolutional, max polling hidden layers
  • Connect the layers with flattening and reshaping techniques
  • Apply dropout

Don’t stop now! The next scenario will only take about 10 minutes to complete.

TensorFlow: MNIST for experts

Step 1 of 4

Training process

As mentioned in the MNIST for beginners tutorial, our deep learning process was defined by few steps:

  • Reading training/testing dataset (MNIST)
  • Defining the neural network architecture
  • Defining the loss function and optimiser method
  • Training the neural network based on data batches
  • Evaluating the performance over the test data

Here we cumulated almost every step from this list and will be working strictly on the neural network architecture design part.

The whole process was defined in the training.py file, and will be imported in the specific files including the code for different neural network architectures. The dense.py includes the simple one output layer network, which is the same as presented in the beginners tutorial. Later in the scenario, we will create a more complex model.

You can run the network training by using the following command:

python dense.py