Description
The aim of this assignment is to help you familiarize yourself with Keras and the Landmarks
image dataset that you will use for your final project.
1. Install Keras
Keras is now part of Google’s TensorFlow. The top-level installation page for TensorFlow is
here:
https://www.tensorflow.org/install.
As the installation page points out, you can run TensorFlow on Google Colab. But, it is good to
have alternatives and to be able to run on your own machine.
The installation instruction uses Miniconda, which is an installer that supports the creation of
separate installation environments. This helps prevent Python package conflicts.
If you have an Nvidia GPU on your machine, then you should follow the GPU setup instructions
and install the Nvidia GPU drive, CUDA and cuDDN. This will speed up running Keras.
2. Fashion MNIST Neural Network
Follow the instructions in Chapter 10 of Aurelien (Hands-on Machine Learning) to create a fourlayer neural network (1 Flatten Layer and 3 Dense Layers) and train it on the Fashion MNIST
dataset.
What to turn in:
• The CPU Times and Wall Times returned by fit() from the training process
• Generate loss and accuracy versus epoch plots (see Figure 10-11)
• The accuracy, precision and recall on the test test
• The precision and recall values by class_id on the test test. There are 10 classes.
3. Fashion MNIST Convolutional Neural Network
Repeat part problem 2, but this time create a convolution neural network using the Fashion
MNIST network in Chapter 14 of Aurelien.
