(Skip this section if you already have a Node-RED starter application running.)
On your bluemix account navigate to the catalog. Once there select the Node-RED starter under boilerplates.
Follow the instructions to create an instance of the Node-RED starter app on BlueMix
You may also choose to link your new BlueMix app to github. This is not required for our prototyping environment, but can be a very useful tool.
For documentation on the Node-RED Starter feel free to visit https://www.ng.bluemix.net/docs/#starters/Node-RED/nodered.html#nodered
Node-red Sample Flow
Launch the Node-RED flow editor from the App overview page.
Feel free to access the original flow at http://generalclassifierapp.mybluemix.net/red/# You can also copy the nodes from this flow and import them to the clipboard.
Enter the Node-RED flow and use the top right corner menu to import from clipboard the following:
The resulting flow should look something like this:
How to use the General Architecture
The inputs should be defined by your use case, but starting points have been provided to help facilitate prototyping. There are other ways to send input, however these are the most common. They will each need to be configured to work with the devices that you intend to use.
The classification of translated speech is done by analyzing the natural language through multiple tiers.
Each step in the process represents a classifier on the NLC BlueMix service (one service many classifiers). For most applications you only need two tiers to get to actionable information, but this can be grown depending on the application. The number of outputs should be changed to represent the number of classes. This can be done after you have built your global classifier on the BlueMix service.
This architecture makes each step simple to solve, and in turn provides high accuracy with limited training data. This is most appropriate when we know the interactions that we want to support between user and device.
You will need to configure each of the nodes in this flow to fit with the classifier that you built.
For more information on how to build a classifier with the NLC service feel free to visit:
When building your classifier tree remember that each category should be obvious. Thus each individual classifier is small and accurate. Your classification categories should be as different from each other as possible.
As an example a global classification could be food, a tier 1 classification could be meats, a tier 2 classification could be fish. After the tier 2 classification is identified we can parse the string to find words like tuna or salmon. Thus in three layers we can have machine and human speaking the same language. More tiers means more specificity but also more outcomes.
These are among many possible outputs. Each demonstrates the capacity that BlueMix and Node-RED have to facilitate communication from human to device to cognitive cloud. Like the inputs, these will need to be configured for your environment and devices.