IP Insights Model ‘De-simplify’ — Part I — Intro
Hope you can get a feeling of how much more thoughts we put into a high quality and practical ML product.
For all the on-line blogs and posts we saw, many people created a super simplified scenario and shows how to use a certain model or technique to solve a problem. Yes, this is an important first step to bring you to the the data scientist path. However, the problem remained ‘unsolved’ with such a shallow touch. Being able to provide a thorough and mature ML solution is a key feature to make a good data scientist standout.
Because you did a thorough analysis and your solution just will work better than others. Your clients and stakeholders would really appreciate your work because you solution is so thorough, it saves many rounds of back and forth discussion and speed up the whole project.
In this blog, we use the AWS IP insight model as an example to demonstrate how to bring a toy example to production level.
Enjoy!
Contents
- Detailed Tutorials of the AWS IP Insights Model
Before showing you what are the extra spices, I will start with a detailed summary of the AWS IP Insights model itself. What is the difference of my IP Insights model tutorial with AWS blogs ?
I added more details explaining why and more visuals (flowcharts or results visualizations) to help you understand the pipeline.
Sounds good? If yes, don’t hesitate to check out the first point. If you are familiar with the AWS IP Insights model (maybe you are the developer :), please jump to the second point).
Modeling Deployment and Evaluation
2. Upgrades to recover the real-life challenges of the building a ML product
a. You never have a 💯 clean data set to start;
b. More in-depth model training and tuning techniques;
c. What is a good evaluation metric?
d. How to monitor the performance of the model in real-time ?
e. Is this the best solution for the problem?
f. Any other features can be used to improve the model’s performance.
3. Close the loop of responsible AI
a. A detailed validation on the model
b. A detailed monitoring plan of the model
c. Tools and Automation
d. Did our AI model bring in any vulnerabilities ?
