Container Based Execution Stack for Neural Networks (ConbexNN)
This a project features web services to train and evaluate neural networks using the Kubernetes container orchestration and a Java based microservice architecture.
Demo VM
See the project in Action by running a virtual machine. It comes preconfigured with Kubernetes running all necessary ConbexNN services and a neural network training set for testing.
You can try out the RESTful API and GUI.
Specs
- Format: OVA
- Size: 4.3 GB
- System Lubuntu 18.04.1 64bit
- Recommended RAM: 3GB
- Required disk space: 15 GB
Tested with VirtualBox 5.2.18 r124319 on macOS 10.13.6
Download
Import
In VirtualBox click
- File / Import Appliance
- Select the .ova file and start the import
- the machine can be started now
Credentials
- User:
conbexnn - Password:
conbexnn
Features
Startup
After Login the VM starts autostart.sh initalizing ConbexNN. The Iris testset and a ViNNSL neural network are automatically imported.
Browser
Firefox is preinstalled and opens predefined Tabs
- Kubernetes Dashboard
- VINNSL-NN-UI (user interface showing imported neural networks)
- Swagger API (documentation of the ConbexNN RESTful interface)
- Status Tab (shows status of imported neural networks)
ConbexNN Endpoints
https://localhost + endpoint
| endpoint | Service |
|---|---|
| /#/ | Vinnsl NN UI |
| /vinnsl | Vinnsl Service |
| /status | Vinnsl NN Status |
| /worker/queue | Worker Queue |
| /storage | Storage Service |
| /train/overview | DL4J Training UI (while training) |
Postman
Postman comes also preinstalled with the VM and contains a collection of requests supported by ConbexNN. You can view all neural networks, create new ones, start training, …