NeoVis

New! Now works with OgmaNeo2 (through PyOgmaNeo2)!

NeoVis is a visualizer for OgmaNeo2 (using the Python interface, PyOgmaNeo2) hierarchies, built using 'dear imgui' (AKA ImGui) and the Simple and Fast Multimedia Library (AKA SFML).

Overview

NeoVis is an application that allows one to visualize in real-time the contents of an OgmaNeo2 hierarchy. It uses sockets to communicate with the client program. It can be connected and disconnected seamlessly, provided the client code contains the appropriate visualization update function.

Building

The following commands can be used to build the NeoVis library:

cd NeoVis
mkdir build; cd build
cmake ..
make

The cmake command can be passed a CMAKE_INSTALL_PREFIX to determine where to install the visualization executable.

On Windows it is recommended to use cmake-gui to define which generator to use and specify optional build parameters.

CMake

Version 3.1, and upwards, of CMake is the required version to use when building the library.

Client code

Include visadapter/VisAdapter.py into your Python code by whichever method you prefer. Then:

from visadapter.VisAdapter import VisAdapter

# ...

cs = pyogmaneo.PyComputeSystem(...)

# ...

h = pyogmaneo.PyHierarchy(...)

# ...

enc = pyogmaneo.PyImageEncoder(...)

# ...

# Initialize the VisAdapter
v = VisAdapter()

v.create(54000) # Port

# ...

# In simulation loop:
v.update(cs, h, [ enc ]) # Visualize hierarchy h and pre-encoder enc

Host operation

Once NeoVis is started, use the Connection button and Connection Wizard dialog box to open a connection to your hierarchy. Simply specify the address (localhost, if on same machine) of the client, and make sure that both applications are using the same port (default 54000). Once Connect! button has been pressed, and the status switches to "Connected", you should see several windows appear.

Each layer has windows for its hidden layer CSDR (Sparse Distributed Representation) and feed-forward weight matrices.

The CSDRs are organized into a "grid of grids", where each sub-grid represents a 1D column (wrapped into 2D for ease of visualization). You can right-click on any cell to show the corresponding feed-forward weight matrices. You can also hover over the weight matrix display and use the scroll wheel to cycle through different Z-levels of the weight matrices. If there are 3 or 6 Z-levels, these will be visualized as either RGB or two RGB (side by side) images, respectively.

At the moment, CSDR and feed-forward weight matrix visualizations are the only features of NeoVis. Despite this, we found it quite handy for debugging our programs. If your application isn't functioning properly, it may be a good idea to pear into the network with NeoVis!

Note

Please note that the visualizer only works with simulation-like environments. It assumes that the client is running in real-time, and it streams whatever the latest state is for visualization. This makes is particularly handy for visualizing the OgmaNeo2 hierarchy during reinforcement-learning-like tasks and other realtime applications.

Contributions

Refer to the CONTRIBUTING.md file for information on making contributions to NeoVis.

License and Copyright

The work in this repository is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. See the NEOVIS_LICENSE.md and LICENSE.md file for further information.

Dear ImGui is licensed under the MIT License.
SFML license https://github.com/SFML/SFML/blob/master/license.md

Contact Ogma via licenses@ogmacorp.com to discuss commercial use and licensing options.

NeoVis Copyright (c) 2017-2019 Ogma Intelligent Systems Corp. All rights reserved.