Step #3: Use AutoML To Find the Optimal Hyperparameters

Develop and Tune Computer Vision Models using NVIDIA TAO AutoML (Latest Version)

This lab will use the link from the left-hand navigation pane throughout the course of the lab.

  • Jupyter Notebook

The key objective of this lab is to use AutoML capability in TAO to find the best set of hyperparameters for a given dataset. You will evaluate the model accuracy of several experiments and choose the best model for your dataset.

Open and run through the AutoML notebook, by clicking the Jupyter Notebook link in the left hand navigation pane and then selecting and running lab3-automl.ipynb, in the tutorial folder. This will take approximately 12-18 hours to complete

Notebook steps:

  1. Import data from previous steps. Setup dataset for training again and split into training and evaluation sets.

  2. List available pretrained models & select the one to be used

  3. Configure AutoML and hyperparameters. Depending on what model network is used there will be different hyperparameters that can be optimized. We can go with the default set or select additional parameters.

  4. Run AutoML. AutoML will iterate the training with varying hyperparameters to find the best model. This step will take several hours and can be polled to check status.

  5. View best model created by AutoML. The accuracy of the model created by AutoML should be better than the baseline without AutoML.

© Copyright 2022-2023, NVIDIA. Last updated on Mar 20, 2023.