Survival Analysis for Deep Learning

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

Getting started

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