The diagram below shows the overall architecture.
The use case is to predict the coffee flavor (or no coffee at all) for a new visitor in our office.
By asking a few questions, we hope to surprise the visitor with the right kind of coffee.
Before we can predict, we need to have labeled data.
The table below shows a subset of the columns and of course the label "coffee" that we want to predict.
This data set is used in Watson Studio to train our model, in this case we went with the random forest algorithm.
In the Assistant we configure the dialog node to get the answer from a cloud-function
To connect the assistant with the machine learning model, we use an IBM Cloud Function:
The cloud-function parses the collected answers and calls the Watson Studio Scoring Endpoint:
We have captured the end to end solution in this video