Hyperparameter Tuning with Neural Network Intelligence

Overview

Class Central Tips

In this 2-hour long guided project, we will learn the basics of using Microsoft's Neural Network Intelligence (NNI) toolkit and will use it to run a Hyperparameter tuning experiment on a Neural Network. NNI is an open source, AutoML toolkit created by Microsoft which can help machine learning practitioners automate Feature engineering, Hyperparameter tuning, Neural Architecture search and Model compression. In this guided project, we are going to take a look at using NNI to perform hyperparameter tuning. Please note that we are going to learn to use the NNI toolkit for hyperparameter tuning, and are not going to implement the tuning algorithms ourselves. We will use the popular MNIST dataset and train a simple Neural Network to learn to classify images of hand-written digits from the dataset. Once a basic script is in place, we will use the NNI toolkit to run a hyperparameter tuning experiment to find optimal values for batch size, learning rate, choice of activation function for the hidden layer, number of hidden units for the hidden layer, and dropout rate for the dropout layer. To be able to complete this project successfully, you should be familiar with the Python programming language. You should also be familiar with Neural Networks, TensorFlow and Keras. Note: This course works best for learners who are based in the North America region. We’re currently working on providing the same experience in other regions.

Syllabus

Hyperparameter Tuning with Microsoft NNI

In this 2-hour long guided project, we will learn the basics of using Microsoft's Neural Network Intelligence (NNI) toolkit and will use it to run a Hyperparameter tuning experiment on a Neural Network. NNI is an open source, AutoML toolkit created by Microsoft which can help machine learning practitioners automate Feature engineering, Hyperparameter tuning, Neural Architecture search and Model compression. In this guided project, we are going to take a look at using NNI to perform hyperparameter tuning. Please note that we are going to learn to use the NNI toolkit for hyperparameter tuning, and are not going to implement the tuning algorithms ourselves. We will use the popular MNIST dataset and train a simple Neural Network to learn to classify images of hand-written digits from the dataset. Once a basic script is in place, we will use the NNI toolkit to run a hyperparameter tuning experiment to find optimal values for batch size, learning rate, choice of activation function for the hidden layer, number of hidden units for the hidden layer, and dropout rate for the dropout layer.