Nieme Explorer

Since v0.7, Nieme includes a User Interface (UI) that allows to browse into Nieme classes. Any set of class that have been saved with the Frame::save() function can be opened in the Nieme Explorer for visualization. This UI uses a combination of generic and specific widgets. The generic widget is a tree of attribute/value pairs that allows to display any set of classes. Specific widgets have been developed for important classes such as DoubleVector or Tables.

This image shows the results of a Cross Validation experiment: two tables for the training and testing folds and a global "time" value.

This image shows a learnt machine. A machine is made of its Architecture, its Loss, its Regularizers and its Learner. Moreover, the Explorer allows to browse into the learnt parameters. The explorer includes sorting options in order to quickly see which weights are the most influent.

The Explorer includes a complete UI for browsing and interactively manipulating Decision Processes and Policies. The two images below shows the User Interface for Decision Processes. The UI is made of different parts:

A toolbar for easy navigation: selecting a new instance, selecting a policy, and making steps with the current policy.
The current trajectory: this component summarizes the sequence of actions that have been taken in the decision process. For each step, the UI displays the action names, the time the action took and the immediate and cumulative rewards.
(left) The current state.
(right) The current set of actions from which the policy should choose. It is possible to register a set of Action Values in this component to see how well an action is scored.
(down) The feature descriptions. This component allows to display the feature representations of the current state and actions.


A Sequence Labeling Decision Process	Another Decision Process: Le Compte est Bon