Key Features
The key features of Model Fabric application are:
Data Exploration
-
Data ingestion through Cognitive Data Lake (CDL), Industry Cognitive Model (ICM) and local CSV/XLS file.
-
Provision to select “variables” (columns) or “instances” (rows) in a dataset for the experiment.
-
Operating Mode Conditions feature to filter the data based on specific conditions defined and display the number of rows and columns that are filtered in the dataset as per the conditions applied.
-
Statistical summary and visualization of each variable that helps in understanding the data distribution and variation.
-
Missing data and outlier data identification.
Model Recommender
-
Data Pre-Processing
-
Missing data and outlier data treatment, where the missing and outlier data in the dataset can be treated with the appropriate treatment methods available.
-
Capability to select variables based on zero variance, which means that the value of the variable is constant throughout the dataset without any change.
-
Capability to select variables through multicollinearity and relative importance techniques, to identify multicollinearity of data among the variables in dataset and avoid redundant data.
-
Capability to select variables based on Feature Selection and Feature Reduction techniques.
-
Capability to apply scaling technique to change the scale of values in the variables (finalized during the variable selection process for each target variable) measured on different scales to a notionally common scale.
-
-
Model Building
-
Pre-built learning algorithms based on different techniques.
-
Bayesian optimization based on timeout to find the optimal hyper parameters for each algorithm.
-
Validation techniques to critically evaluate model performance on unseen test data.
-
Testing, Training and K-fold metrics to evaluate the model.
-
K-fold validation technique to cross validate and evaluate the performance of a model.
-
Graphical representation of model performance and model sensitivity analysis.
-
Model Registry and Deployment
-
Capability to use models built/trained by Model Fabric or import and register your own models trained outside of Model Fabric application.
-
Hassle free model deployment on a single click.
-
Model versioning and deployment status tracking for each model.
Model Monitoring
-
Dashboard that provides a quick overview of models that are in production, models that are retrained, models with data drift and latency and models that are under-performing.
-
Provision to monitor the data drift, performance drift and latency of the models.
-
Capability to retrain a model with new dataset or same dataset but with new experiment
-
Allows you to view various versions of a model, compare the metrics and plots of a model with all the available versions, replace the model or archive the model as per the requirement.
-
Provision to schedule the model monitoring as per the requirement.
Model Fabric Insights
-
The Model Fabric Insights module in the Model Fabric application enables you to visualize and get an overview of all the types of models created, registered and deployed in the Model Fabric application across all the projects at an enterprise level through various tiles, graphs and charts. Eventually, this slice and dice of the complete data in the Model Fabric application provides actionable insights to the management.
-
The number of models in the respective tiles providing the business insights and industry insights are displayed in the chart view.
-
Ability to filter specific models through the pie charts in the respective tiles and accordingly visualize the required data.
Related Topics: