This is a tutorial on survival analysis, also referred to as time-to-event analysis or reliability analysis. You will learn how to train a convolutional neural network to predict time to a (generated) event from MNIST images, using a loss function specific to survival analysis.

There are two versions of the tutorial:

1. Using tf.Estimator and TensorFlow 1.X: tutorial_tf1.ipynb
2. Using Keras and TensorFlow 2.X: tutorial_tf2.ipynb

The easiest way to run this notebook is Google Colaboratory. If you want to run this notebook locally, you have to make sure the following dependencies are installed:

• numpy
• matplotlib
• pandas
• scikit-survival
• tensorflow