Hyperparameters Real time Deploy

Hi Mentors,

@paulinpaloalto @bahadir @eruzanski @Carina @neurogeek @lucapug @javier @kampamocha

Basically for particular datasets, we are doing Hyperparameter search using cross validation dataset, select the one which works best in terms of improving accuracy or reducing the value of the cost function J.

So In real time project, after model deployed, when the model is executed or functioning, will the model do the hyperparameter search to pick best learning_rate or the model will just use best learning_rate which we found earlier before deploy the model ?

The point of hyperparameters is that the model does not change or learn them. You have to select them and then evaluate whether they are acceptable or not.

1 Like

Hi @Anbu,

As @paulinpaloalto said, the algorithm does not learn the hyperparameters by itself, you need to tweak them by yourself. The hyperparameter search strategies and cross validation techniques can be helpful in doing so, until you are satisfied with the results and then you fix your model to evaluate and deploy.

After deployment you are no longer training your model, so you don’t need a learning rate and, other hyperparameters, like for example the number of layers, are fixed so you no longer need to tweak them.

Hope that helps.


Thanks Kampamocha and Paul :slight_smile:

1 Like