As MLFlow is all usable from the word go, the one place it really falls short is creation of ML jobs pipelines or scheduling of ML jobs. This is where Airflow really shines, though it would be unfair to compare them as they are two different beasts altogether. AirFlow allows expressing the actions as a DAG. Further deep dive will be done in coming days.

I was struggling getting some documents from my Android device to the PC, using the BT Dongle, I suddenly realized that long file names can make the transfer fail. Though this was not mentioned in the Win 10 file transfer dialog. It just mentioned that the file transfer was aborted. So to make transfers work just limit file names to max 30 characters.

In any field there are books which are the go-to books to get the in-depth knowledge of the subject. In ML, these books are often cited:

• Applied Predictive Modeling
  
• An Introduction to Statistical Learning: With Applications in R
  
• The Elements of Statistical Learning: Data Mining, Inference, and Prediction
  

I am quite excited to check them out for now.

MLFlow is proving to be a nimble tool for model lifecycle management. I was amazed by the ease with which a registered model can get into production. Kubeflow was expected to be an answer to all ML deployment issues, but its heavy infra requirements are too much of a let down. this is especially true for small projects with small deployment expectations.