Issue resolution

Monitoring Machine Learning in Python

Hakim Elakhrass

Co-founder and CEO of NannyML

Do nothing

• Works well with up-and-running good monitoring system
• Requires an opportunity cost analysis and good understanding of a use-case
• An example is overestimating the number of calls in call center

Retraining the model

• Train on both old and new data
  • Making the model more robust
  • Learn the model as many as possible distributions
• Fine-tune the old model with the new data
  • Simply refit the model with the new data
  • More effective than training a new model from scratch every time
• Weighting Data
  • Give more importance to the recent data

Reverting back to a previous model

Change business process

• Change the business rules
• Run manual analysis on predictions

Let's practice!

Monitoring Machine Learning in Python