Module For Hyperparamater Sweeps

Do you find it difficult to choose optimal hyperparameter values for training? What if you could experiment with a variety of hyperparameter values and their combinations? With this module, you can designate a range of values for a hyperparameter, specifying the step size for experimentation. Additionally, you have the flexibility to select combinations of various hyperparameters for your experiments.

How To Use This Module?

1. Install this module in your app where your dataset resides and your models will be stored
2. Open the installed module
3. In the first section, enter the model details
4. In the second section, enter the checkpoint model details (optional)
5. In the third section, enter the training parameter details (e.g. dataset details)
6. In the fourth section, enter the output parameter details (e.g. concept details)
7. In the fifth section, enter the model template details
8. In the sixth section, several hyperparamters would be displayed once the template is chosen i.e the sweepable parameters
9. To experiment on a hyperparamter, tick the Try a range of values checkbox and choose the min & max value for the range and specify the step size
10. In the seventh section, enter the non-sweepable parameter details
11. In the final section, using the previous sections data all possible combinations of hyperparameter values are generated and displayed
12. Select the combinations you want to train the model on and check the cost that will be incurred for each training on per hour basis
13. Once you click on Submit, model versions would be created on your Models section of the app
14. You can check for training & testing details on the Models page and see which model version performed the best by clicking the Model Training Page button