The end-goal of many machine learning projects is to get a valuable model into the hands of users that will benefit from it. No matter how good your ML model is; a lack of monitoring and insights can prevent users from being willing to use your product.
A monitoring dashboard is easy to build & serves two functions: a. Monitoring of driving performance metrics b. Delivering insights into the model & it’s predictions.
These two functions combined can greatly help towards the primary goal of getting a valuable model into the hands of users. A monitoring dashboard thus serves an important purpose. Luckily, such a dashboard can be built effortlessly using tools like streamlit. We will show how a minimalistic monitoring dashboard that serves both functions can be built using streamlit in only a few hours of work.
This talk will cover the following topics, linked to each other through real-world examples:
- My struggles of getting a model in use by willing users.
- Why you need to monitor performance metrics.
- Why performance metrics might not be enough & you might need more insights.
- How easy it is to do all this in streamlit.
I'm a machine learning engineer at the Data & AI consultancy GoDataDriven, particularly interested in getting machine learning models from problem to solution in production
visit the speaker at: Github
Experienced change manager with extensive operational experience in customer-oriented organizations in Transport, Retail and FMCG. Powerful in aligning stakeholders, especially in highly complex environments and always with an IT component. Likes to aim for results on a human-oriented way in a step-by-step approach. Always with lot of enthusiasm.