Machine Learning is Easy, then what is difficult?

In the whole Machine Learning development, model development is the easy and least time-consuming effort. Operationalization is the hardest and that’s where all projects are either failing or struggling to justify investment.

It’s not a one time effort.

Most Machine Learning projects start with access to existing data and particular use case to solve. The analogy I use often is making a resolution to stay healthy.

Steps to stay healthy:

  1. Get a proper night’s sleep
  2. Start your day with exercise
  3. Make healthier food choices

Doing it once is Easy

Making it as a practice is HARD

day_routine-v4

The above analogy fits well for Machine Learning projects. It’s easy for project teams to work on point in-time data and build Machine learning models.

Steps to build Machine learning model:

  1. Fetch relevant point in-time data.
  2. Build Model
  3. Test, train and evaluate model results.

Again, doing it once is EASY

Integration is HARD

Automation/Doing it at scale is SUPER HARD

springMl-Architecture-v3.gif

In the next subsequent blogs, we’ll talk about platforms (Google, Amazon, Microsoft) that are investing signifcant resources to provide ML platform as a service compared to Predictive Apps (Salesforce, Oracle) that provide ML as APIs.