Stream’s analytics site is the website which hosts Stream’s analytics platform. The whole idea is to use machine learning to analyze spectral and spatial data and provide a prediction.

The site, found at analytics.streamtechinc.com, is designed to be easy enough for non-technical people to use to solve difficult detection problems.

There are seven elements to the website which take advantage of the Stream analytics engine and its unique capabilities. These elements include:

  1. Stream Online Store – where all models (Aapps) are distributed from
  2. My Company – where secure accounts are created and managed
  3. Connected Devices – shows all the data capture devices registered to you
  4. Hardware – lists all the hardware devices that are supported
  5. File Storage – folders to store all your data used to train Aapps, or make predictions
  6. Aapp Creation – where machine learning models (Aapps) are trained
  7. Prediction Sets – where the results (Predictions) to analyzed data is viewed and saved

Below is a high-level overview of what each of these sections is about. On the following pages you can walk through a more detailed explanation and view screenshots of creating and using the site from start to finish.


Stream Online Aapp Store

Where Aapps are made available. View, subscribe, or try one out at no cost.

Once a new Aapp is created and tested, it can be published into the Aapp store if the intent is to allow the public to have access. Aapps may be made public only by the company that creates it. For these Aapps, access is restricted with a private hosting option.

Aapps created and made publicly available are generally built to detect a specific target and use a specific type of data capture device. These details can be viewed for each Aapp to help determine its suitability to your purpose.

All Aapp subscriptions include a free tier so the Aapp can be tried for free to ensure it preforms function you’re looking for.

(Read more...)

The person who creates the Aapp and publishes it to the store sets the price for the Aapp. A price per prediction is established at the time of publishing, and pricing tiers are then auto generated for increased levels of use. The Aapp publisher can also earn a royalty on revenue generated by future subscriptions.

It is up to the creator of the Aapp if it will be placed into the store (subject to review) or kept private.
If the Aapp is kept private, the cost per prediction is set by Stream, and there is no ability to earn royalties on its use.

One of the main reasons you may want to share your Aapp through the store, is because over time and with wider use and more data, there are ways in which the Aapp can be made more accurate and more robust. The specifics vary, but in general the more it’s used, the better it becomes; and with more variety of data the better it becomes. Privately hosted Aapps do not benefit from these advantages.

My Company

Account management

Create your own secure account and you can immediately begin using the analytics website and making Aapps, at no cost.
From your account you can see how many predictions have been made, or how much time and storage you have used each month or cumulatively to date.

Others within your organization can be invited to collaborate.
If you decide to purchase an Aapp, all that is here too.

 

(Read more...)

There are two types of users:

  1. The account owner: The person who creates the account and will manage the accounts users.
    1. Creating an account is free.
    2. Using the free tier to create and use an Aapp is also free.
    3. If the monthly usage exceeds the free limits – the account owner will be responsible to upgrade to the appropriate tier.
  2. Invited users: People invited by the account owner. These are typically employees of a company, or students of a school, where the main administrator of the account is someone else.  As an invited user there are some limitations, including not being able to purchase Aapps for that account, and viewing usage limits or billing details.

Connected Devices

All connect devices are listed here.

Whether you use a camera, spectrometer or other data capture device, the picture or file has to be sent to our analytics engine. As the analytics engine is in the cloud, there are two methods to get the data into your folder in your account:

  1. If your camera or device has built in support for ONSI then the image or data can be automatically uploaded to a folder in your account. The Cloud Connect software knows how to connect the hardware with your account. This is for both training data and prediction purposes.
  2. Secondly, you can manually drag and drop your files into your folder in your account. Any device that produces images or spectral data and outputs a standard file can be used. Handy for testing and trying things out.
(Read more...)

In order to create accurate predictions, the more information known about the data capture device, the better. In fact, the metadata from a device can be used by our analytics engine to supplement creation of an Aapp. For example, it may be useful (even critical) to know the aperture setting of the camera used in obtaining the training data. Later, when you want to make predictions, the analytics engine – using Cloud Connect software –  can  apply that setting so images captured for predictions are more accurate.

Stream’s Cloud Connect software can be installed on any Windows computer. Stream also sells a small hardware device which includes a variety of ports, Wi-Fi and all the required software as a convenient alternative to a laptop.

Hardware

Supported hardware at a glance.

There is a wide range of hardware that can produce data for Stream’s analytics platform.

As a general rule, if your data can be represented as a series of labelled values, it will probably work. This includes data from cameras, spectrometers, phased array devices, gamma, x-ray or MRI machines, and the list goes on. The most common data capture device is a color camera.

Supported hardware refers to those devices with direct support through an API and our Cloud Connect software.
Stream also supports the ability to simply drag and drop files, allowing data from an even broader range of devices to be used.
It might make sense to first build an Aapp by simply using labelled images from a camera or spectrometer scans you already have – dragged and dropped into your free account. If your Aapp looks like it has value, you can make it better by thinking about how it will be used in the real world. Are other users going to be happy with drag and drop, or are they going to be looking for a more automated process when using your Aapp to do a prediction?

(Read more...)

If you are wondering if your hardware can be used,  just drop us a note with the Contact Us form and we’ll call you back.

File Storage / Folders

Folders are used to store your images and files. You can create and name folders in whatever manner makes sense to you. Folders store both training data and sample data from which predictions are made.

Stream currently provides 5 GB of free storage space for each new account (or company). If you exceed this limit you can always purchase more.

(Read more...)

You will find that images produce larger file sizes than spectral data (spectrometer scans). You can keep an eye on your usage amount under the “My Company” menu item. If you don’t see “My Company” in the left-side menu when you are logged in, that means you are an invited user and not the account owner.

Create an Aapp

The Aapp section is used to create, edit and publish Aapps.

Creating a new Aapp is easy. Fill in the fields specifying what you are looking for and what type of camera or spectrometer or other hardware are you using to capture training data. This screen defines the type of device and the settings you will use to make predictions.

When creating a new Aapp, there are a few simple steps that need to be completed. The first step is to collect the training data, which is stored in folders you create. After a certain minimum amount of data is collected, the yellow “Train” button will unlock. When you push the “Train” button, the analytics engine will build your new Aapp and prepare a Training Report to show you how well your Aapp is performing.

(Read more...)

Your reports are available under the Training Reports icon and include different levels of detail. The most basic level is a color indicator. Red means the analytics engine has not been able to build an effective Aapp. Yellow indicates there is certainly something there, but not great, indicating that you may need more training data. Green means the predictions are good or better. Further detail is also shown, including accuracy and other performance metrics. Ultimately, as the creator you decide the level of accuracy required. As with all machine learning, more training data will produce better results.

Once you hit the Train button it can take many hours of processing time to determine how well the model is working. We provide 5 hours of training time every month per account for free. Our intent is to provide enough free training time to build one new Aapp every month. Of course, if you require more time, you can upgrade to a paid pricing tier for as much training time as you need.

Prediction Sets

Here is where you store all the images you want to analyze – and save all your past predictions.

A Prediction set is a container which stores a collection of files, which includes the scans of the samples you want to analyze, a link to the Aapp which does the analysis, and the final result or prediction.

The goal of the Stream analytics platform is to obtain “Predictions”, which is what we call the final result of the analysis, usually in the form of either a value or false-colored image. For example, in a regression Aapp built to detect the percentage of protein in a seed, the prediction might be ‘12.5%’.

All this information is stored in your account, in a single container you can name, and re-open anytime. This is called a Prediction Set.

(Read more...)

You can review your Predictions from any browser, using a phone, tablet, or computer.