Overview
This chapter aims to provide an overview of the main features of NuPIC Studio, including a brief explanation of how the features are organized in the main window.
This the main window of NuPIC Studio:
Except for the menu and tool bar, all window tools can be customized according to your taste. This means that you can move, resize, or hide any window tool.
Let's look at some tools...
This tool is the essence of NuPIC Studio. With it you will be able to model a hierarchy of regions as proposed by the HTM theory, check the source code corresponding to it, or simply select a node to see its details. A hierarchy is composed of regions and sensors:
- Region: represented by a trapezium. Every Region must have at least one lower region or a Sensor to feed it input.
- Sensor: represented by a rectangle. Its only role is feeding a region with acceptable inputs, i.e. a fixed array composed of 0's and 1's. Sensors collect input data from a file or a database. This data may be converted values (i.e. the array of 0's and 1's ready to be fed) or raw values (i.e. natural values like numbers, string, etc, which must be converted to an array using an Encoder).
The essence of the tool bar is to provide commands to initialize the simulation of a neural network and follow its changes with time. It also provides a slider bar which allows you to see its state in a timestep t, t-1, etc. This way you can check if the columns that were predicted are indeed currently active; which synapses were created or removed since the last iteration, etc.
When you initialize the simulation, this tool shows a 3D representation of the neural network. Using the mouse or keyboard you will be able to change camera position, hide or show a given elements (cells, synapses, etc) and their states (active, inactive, etc), and save your customized view for future use.
This tool allows you to see the details of a region or sensor when you click on it. In the case of regions, these details usually include its statistics and a list of elements like columns, cells, etc, and their states and statistics. In case of sensors, these details include the original value of the input (i.e. the value that was converted by an Encoder) and a list of bits and their states and statistics. Furthermore, when you select an element in the list, its representation in the 3D Simulation is highlighted (in blue).
This tool shows a report of the simulation as you interact through time. It can be useful for tuning your HTM network and testing which parameters produce best results in terms of performance and accuracy, for example.