search

AutoViML / deep_autoviml

Build tensorflow keras model pipelines in a single line of code. Now with mlflow tracking. Created by Ram Seshadri. Collaborators welcome. Permission granted upon request.
Apache License 2.0
120 stars 36 forks source link

added MLflow integration #12

Closed rajeevkumarsah closed 2 years ago

rajeevkumarsah commented 2 years ago

MLflow Integration with AutoViML.docx

rajeevkumarsah commented 2 years ago

  1. Can you please set up defaults for each new string variable that yo have created? That way people can use default names if they don’t mind it. Rajeev: Default is already setup for all variables, that means user don't need to provide name unless otherwise. User only needs to specify "use_mlflow=True" to use MLflow.

  2. Can you tell me whether you can provide some README updates on how people will be able to view their experiments once online? Rajeev: Done

  3. Can you print a link to the experiments once the fit is done? Rajeev: Done.

  4. What happens if they fit repeatedly in a single session? Are all those strings causing the experiments to be updated or are they causing overwrites? Rajeev: It is not overridden, it creates multiple model with different timestamp . So user can compare the performance and pick the best one. MLflow Integration with AutoViML.docx

AutoViML commented 2 years ago

Hi @rajeevkumarsah 👍

I have gone through the changes and they are good so far. However I am keen on adding model related hyper Param’s to the mlflow tracking. That’s why I decided to seek help for mlflow. For example:

See the TF API:

https://www.mlflow.org/docs/latest/python_api/mlflow.tensorflow.html

I need to add that to AutoViML using the same way that this article has done to Sklearn models:

https://medium.com/the-point-collections/intro-to-mlflow-with-colab-part-2-2-ae03ffd3930b?source=linkShare-a069c5c5154e-1641910217

Can you do a second round of changes? I’ll approve the current changes for now but add these as comments for the next iteration.

AutoViML