So we’ve arrived, seemingly at the end of our journey. I wouldn’t be ready to pull the rip-cord just yet. We still need to get the customer to agree that our model is fit for purpose, and present our results to them. We also need deploy the model as previously agreed in our Architecture Discussions.
What we are doing here is deploying the model and making it available for production use by the customer.
There are many considerations for how we could deploy a model. Model deployment can involve many things, but it always depends on how the business is looking to implement the model. Ideally the data scientist will have defined this up front with the customer in the “Architecture” part of this process, therefore putting it into action will be relatively simple.
Some Options for model Deployment:
- Deploy on Businesses Current Architecture.
- Deploy in a new architecture created specifically created for the project.
- Provide code package in handover to business as Jupyter Notebooks.
- Host as an API in a third party app and charge for usage on a pay as you go basis (Machine learning as a service)
- Integrate into business Information delivery tool (more below)
Industry Trends in Deployment
Quick side note – its worth touching on the current industry trend of ‘Citizen Data Science’, the idea here is that many of the leading Business Intelligence Tools are attempting to make it easy for an end user to avail of advanced analytic techniques, with no coding experience whatsoever.
An interesting approach is for the trained model to be uploaded into the BI tool, therefore enabled to be called at will by end user from with the Business Intelligence tool. More on this here
The final curtain – or so it may seem. I wont go into much depth here…stick everything into a report and accompanying presentation. I will post another article on producing the perfect report, but I don’t believe this should be too taxing – we already have all the information available to us, we just need to document and present it.
Two important things to take into account on closing:
- In your report always outline options for future direction, these can be things that were discovered and could warrant further investigation, this means potential project extensions. More work is always good in any situation.
- The feedback stage is iterative, it can be ongoing for months – be prepared to loop all the way back to the start of the process and repeat any steps – once the customer gets a better understanding of the undertaking, they will always have more questions for you to explore.
So we reach the end of this series of posts. Good Luck – these are ongoing work, they were somewhat rushed, and all put together on one flight, so I was losing sanity toward the end, meaning corners were probably cut – Do feel free to comment anything that you fell is missing or inaccurate.
Thanks for reading – i’m off to sleep before landing.