
Getting started begins with looking at the problem you are trying to solve.
Start by asking yourself – What prediction do you want from the analytics engine? What specific information drives value for you? The analytics engine is going to apply an Aapp, (Algorithm application) to analyze/test your sample. So, the question becomes: Is there a suitable Aapp you can license from the store, or should you build a new Aapp for your needs?
If there is no existing Aapp to detect your specific target of interest, you can create your own Aapp.
Let’s walk through the process.
The following are the steps someone coming to the analytics site for the first time would take to use all aspects of the site. These steps include creating an account, connecting a piece of hardware, capturing and storing training data, creating a new Aapp, and finally using the Aapp to make a prediction.
We are going to use an example where the problem is trying to distinguish between two types of plastic (PLA vs. ABS), even though they appear identical to the naked eye.
For this example, we are going to use an Ocean Optics spectrometer (FlameNIR), together with a light source and a simple fiber optic probe. In addition, we have installed Stream’s Cloud Connect software on the computer connected to spectrometer, which will send readings from the spectrometer directly into the folders in your new account. To see how to install Cloud Connect, click here. Expand to see the actual hardware used.
Let’s begin. Remember you can do this yourself… all at no cost.
Go to the analytics site
We begin by going to the Stream analytics website. The initial page shows the Aapps available in the online store. Viewing the details of each of these Aapps will show what the Aapp does, what hardware is required in order to use it, and which components the Aapp can identify.
We want to create an Aapp, so we need to create an account.
Create your account
Your account can be created for free by providing some simple information.
Connect your device
From the Connected Devices menu
When Cloud Connect is first installed, it will ask for a PIN number. Click the add button on the Connected Devices page and then add a Software Device, giving it an appropriate name for ease of identification. The new software device will then appear under the Connected Devices menu. Click “View Details” on your device to get the PIN and enter that into the Cloud Connect software to complete the process.
Begin to create the Aapp
Enter the basic information
Enter the name of the Aapp as it will appear in the online store. The description explains what the Aapp does. A vanity photo will make a more interesting icon. The hardware model sets what device you are using to capture the data, and the Hardware Profile contains the spectrometer’s settings.
The Aapp type can be either Classification – for a “this or that” problem; or Regression if the problem is a “how much” problem. This example would be Classification: Is the sample PLA or ABS?
Ready to start adding training data
The Aapp’s details
Across the top you will see the steps necessary to complete the Aapp. We are currently at the “waiting for Training Data” stage. Our two components are listed at the bottom, and the scan count (currently at 0/100) is colored red. This shows we have not added any training scans, and the minimum number of scans needed before training is allowed is one hundred. Once we have added at least one hundred scans, it will allow us to move to the next stage. Let’s add the training scans.
Scan the sample
Scan some pieces
In our example we are going to scan some ABS plastic to teach the Aapp what the reflectance spectrum of ABS looks like. The label input here identifies this scan is taken from sample ABS Rd2, and the component is set to ABS to tell the Aapp that this scan is ABS. This round of scanning is set to capture 150 independent scans.
A newly trained Aapp
Ready to be published
After training is complete your Aapp will be created. All that is left to do is to publish it so it is available for use. It can be published either privately or publicly. Before we publish, let’s have a quick look at how well the training for this Aapp performed, by looking at the Training Report.
Training Report details
How well did we do?
This report shows results color coded to give an idea of how well the training went. Green means acceptable, yellow means it is borderline and could use more training, and red indicates that it doesn’t work at all and may require much more training data, a rethink of the sampling method, or a combination of the two.
Publish the Aapp
The Aapp creator sets the price
This Aapp can now be published Privately or Publicly. For this example we will publish Privately. Either way, a price is entered which establishes the cost per prediction from this Aapp for future use. Each Aapp published has a set number of free predictions. For more detail on pricing click here.
Create Prediction Set
A special type of container
Think of a Prediction Set as a container which combines a sample to be tested, the Aapp which will do the test, and the result of the test – the prediction.
Here is where this set is named, and we specify information about which hardware (and its settings) is going to be able to put its data into this set.