Introducing signalAI in Compose
Altair EmployeeLeverage signalAI in a Compose environment
As discussed in a previous article, in the context of anomaly detection signalAI serves two purposes, namely outlier detection & novelty detection. As we speak, within Altair product line, users can leverage signalAI off the shelf with Knowledge studio. Expanding on this we are currently working on introducing signalAI’s functionalities with Altair Compose using a GUI based interface called signalAI director and in today’s article we are going to give a sneak peek into how the end-to-end process looks like before it gets released in Altair marketplace.
Starting with the workflow, there are two tabs within signalAI director that guide the user starting from training the anomaly detection model to making inference using the trained model. The first tab which is by default the landing page of the director is called the “training”. As the name suggests, this tab enables the user to import data into signalAI director as well as training the anomaly detection model. Users can import data stored in their local system in one of the text formats like .csv, .txt, .xls etc using the “Browse” button associated with “TrainData file” function. Along with importing the data, the users have the capability of selecting which folder they want to store their trained model as well giving them specific names and hence enabling them to better track the model performance later. Once the user is done with importing the input data into signalAI, he or she can select which attributes present in the input data they want to use using the “Independent Variable” panel. After the independent variable selection is done, the users can select which algorithm they want to train along with corresponding hyper parameters depending upon the type/nature of the use case they have at their hand. At the end of this tab, users click the “train” button at the bottom of the panel which initiates the training process with the selected attributes, algorithm and corresponding hyperparameters.
Once the training ends, the second tab called the “Viewer” is automatically populated with the results from the trained anomaly detection models. The summary of the results is then displayed in the “Results table” which showcases the user selected attributes along with two additional columns called anomaly flag & anomaly score which determines the outlier status of a particular sample. Here the anomaly flag is a binary score (-1 & 1) representing anomaly vs non-anomaly respectively and the anomaly score is helpful in judging how abnormal a particular sample is. By default, these results are basically on the input data set that the user has fed into signalAI during training, but the users can also perform inference on a completely new data set by using the “Browse” button at the bottom half of the viewer tab.
So, to conclude here, today we have learnt how we can use signalAI within the Compose environment starting from importing data to making inference on a new dataset with a trained model. In our next article, we will touch upon what are the algorithms we have within the signalAI director, their hyper parameters and how the users can better select a model and optimize it’s hyperparameters depending upon the problem at hand. Stay tuned!
Thanks for the read! Happy Learning!