Prediction Models¶
Overview¶
The Insights Hub Quality Prediction application provides a user interface for creating, training and evaluating the data using a relevant machine and processes data to produce the Machine Learning models called as Prediction models.
The Prediction model initially created is in the "Draft" status, by default. Following successful model training, it is saved as a new model version and its status changes to "In Training" and is referred as "Training Board". Once the model is run with real-time predictions, the status of the model is now "Running" and is referred as "Operator Board". The prediction models are displayed as shown in the below image.
① Displays the total number of Prediction Models
② Displays the total number of Training Models
③ Displays the total number of Running Models
All the available models can be accessed by selecting “See all Prediction Models”. "Training and Running" models can be accessed by selecting “See details” for the corresponding models as shown in the below image.
Prediction Model Workflow¶
This workflow outlines the key steps for setting up models and generating prediction results:
Parameters and limits for the Prediction Models¶
Parameters and limits for the Prediction Models are listed below:
Parameter | Limit |
---|---|
Number of Features for Training | 15 |
Number of data points for quality data (rows) | 20,000 |
Number of models | 3 x Number of max active models |
Number of active models | Quota parameter(Initial value is 3) |
Inference frequency | 60 sec |
Inference time window | 7 days |
Inference late arrival window | 1 hour |
Inference data points of all variables ¹ in inference time window | 10,000,000 (15 points per second for 7 days) |
Training time window | 1 year |
Training data points of all variable | 10,000,000 |
Features in preview | 15 |
Filter definitions | 5 |