Lightning AI Studios: Never set up a local environment again →

Log in or create a free Lightning.ai account to track your progress and access additional course materials  

Deep Learning Fundamentals

  • Pages

    • Final certification exam

    Deep Learning Fundamentals > Unit 6 > Unit 6.5
    Course Progress:

    Unit 6.5 Automating The Hyperparameter Tuning Process

    Slides

    Code

    Additional resources if you want to learn more

    Log in or create a free Lightning.ai account to access:

    • Quizzes
    • Completion badges
    • Progress tracking
    • Additional downloadable content
    • Additional AI education resources
    • Notifications when new units are released
    • Free cloud computing credits

    Quiz: 6.5 Automating The Hyperparameter Tuning Process - Part 1

    Grid optimization is the only hyperparameter search method that is guaranteed to find the optimal solution

    Correct. Grid search is an exhaustive search that is guaranteed to find the optimal solution. But it comes with a big asterisk: It requires that the optimal parameter is defined as a parameter on the grid.

    Incorrect. Other hyperparameters are often preferred because they are less expensive, but grid search is the only methid that guarantees to return the optimal solution given that the optimal solution is a choice on the grid.

    Please answer all questions to proceed.

    Quiz: 6.5 Automating The Hyperparameter Tuning Process - Part 2

    How could we change the batch size of our data module to 16 using the CLI?

    Incorrect. Hint: try to run python mlp_cli.py --help for more information.

    Incorrect. Hint: try to run python mlp_cli.py --help for more information.

    Incorrect. Hint: try to run python mlp_cli.py --help for more information.

    Correct. The datamodule specific setting can be accessed via data. ...

    Please answer all questions to proceed.

    Quiz: 6.5 Automating The Hyperparameter Tuning Process - Part 3

    In our custom CLI code, suppose we find that --model.hidden_units "[100, 200]" results in a worse performance. Can we change it to --model.hidden_units "[50, 100]" to reduce overfitting or would that be a problem given that our dataset has 100 input features?

    Correct. We can change it to "[100, 200]", "[100, 50]" or “[100, 100, 200]”, or something else because the 100 refers to the output size of the first hidden layer, not the number of inputs of the hidden layer.

    Incorrect. We can change it to "[100, 200]", "[100, 50]" or “[100, 100, 200]”, or something else because the 100 refers to the output size of the first hidden layer, not the number of inputs of the hidden layer.

    Please answer all questions to proceed.

    Quiz: 6.5 Automating The Hyperparameter Tuning Process - Part 4

    Using the Sweeper app, we saw an example of

    Correct. We can turn this into a GridSearch if we use Categorical distributions instead of LogUniform/Uniform distributions, for example.

    Correct. But we can turn this into a GridSearch if we use Categorical distributions instead of LogUniform/Uniform distributions, for example.

    Please answer all questions to proceed.
    Watch Video 1

    Unit 6.5

    Videos
    Follow along in a Lightning Studio

    DL Fundamentals 6: DL Tips & Tricks

    Sebastian
    Launch Studio →