Hi there!
🌟With new versions of Splunk’s Machine Learning Toolkit (MLTK) v5.3 and Python for Scientific Add-on (PSC) v3.0, we have some important and exciting news to share for all current & future users of MLTK.
It is first important to note that due to several breaking changes in the underlying Python libraries MLTK v5.3 and PSC 3.0 are not backwards compatible with previous versions of MLTK or PSC. - Questions❓see below!
What does this mean for me?
If you are actively using MLTK and PSC you might find that models trained using your current MLTK version do not work with the new MLTK release.
What do I need to do?
On upgrading to MLTK 5.3 and PSC 3.0 you will need to retrain your models. To do this you need to re-run the Splunk search that originally generated the model, i.e. to run the Splunk search that contains the fit command into your model.
What if I have models that are trained using partial fit?
If you have models that utilize partial_fit it is recommended that you delete your existing model and retrain a new model using as much historic data as deemed necessary (the amount of data needed depends heavily on the seasonal nature of the data source used to generate the model). It is recommended to periodically replace partial_fit models anyway to ensure you are not making decisions from models that are biased toward historic data.
Will I need to change any of my searches?
No, all existing searches that utilize MLTK search commands will still operate as expected after the upgrade.
