Woa, It's so amazing, I'm not good at coding, but currently I'm trying to make some machine learning application. All of your works inspired me a lot. Thank you so much and best wishes!
Cool! But I don’t get the point completely, I guess. Could not we save the model as a pickle file and load it from folder (directory) while uploading files (during app deployment).
Hi! Yes you are right that would be possible as well. As I mentioned in the video, I like to keep things separated and in addition to that I wanted to use the MLFlow model registry. First of all I like to separate the models from the application, because that easily allows you to swap the model without needing to restart the application. But the main reason is that we can use the MLFlow registry. During experimentation it is totally fine to start with loading from pickles. But once you try out different models ect. It can get messy very quickly. With MLFlow you can track all the models and can always look back at which loss / parameters /... the models had. Hope that makes sense :)
Woa, It's so amazing, I'm not good at coding, but currently I'm trying to make some machine learning application. All of your works inspired me a lot. Thank you so much and best wishes!
Happy that you find it useful! :)
thank you! nice video!
Cool! But I don’t get the point completely, I guess. Could not we save the model as a pickle file and load it from folder (directory) while uploading files (during app deployment).
Hi! Yes you are right that would be possible as well. As I mentioned in the video, I like to keep things separated and in addition to that I wanted to use the MLFlow model registry.
First of all I like to separate the models from the application, because that easily allows you to swap the model without needing to restart the application.
But the main reason is that we can use the MLFlow registry. During experimentation it is totally fine to start with loading from pickles. But once you try out different models ect. It can get messy very quickly. With MLFlow you can track all the models and can always look back at which loss / parameters /... the models had.
Hope that makes sense :)
nice vid., i have a question when will you re train the data from mlflow serving and how you know which data should be used for re training?
1. If you have additional data that can be added to your overall dataset.
2. If the performance is below to the minimum acceptable range.
@@patrickjoseroxas1771 Thanks Good Insight!
Would be great to delve in more detail on MLflow and how to use it for production purposes (deployment, best practises, etc)
Thanks!
Thanks for your comment :) I'll note it down and come back to it in the future!
Best AI content on RUclips!! Thanks so much!
Great stuff once more!