Now that you have a fully functional end-to-end framework for training and deploying a deep learning signal classification algorithm, the source code is yours to tailor, modify, and improve to fit your custom application.
When adding new layers to your neural network, make sure to confirm that they are supported by the TensorRT deployment framework. You may find a list of supported layers in the TensorRT List of Supported Ops.
Explore the AirPack Python API¶
AirPack is not just a script-based python tool for training an example DNN. It is a fully functional Python API with two main modules, each of which has sub-modules. Because the source code is provided, AirPack may be modified to suit the needs of each particular application.
airpack.deploy- Submodule to simplify the deployment of neural networks on the AIR-T. Included tools use ONNX for saving models, TensorRT for optimizing models for inference, and CUDA for zero-copy memory management.
airpack.tf2- TensorFlow 2 submodule containing the Default Neural Network: an untrained RF signal classifier model. Also included are tools for efficient file I/O during training and inference of the AirPack model. This package is designed to run in the Docker environments provided with AirPack.
AirPack Scripts Package¶
airpack_scripts.tf2- Demonstration scripts for training, benchmarking, and running inference on the training computer using the TensorFlow 2 Default Neural Network. The result will produce a trained model in the ONNX industry standard. This file will be transferred to the AIR-T for deployment. Note that these scripts are typically run within the provided Docker environment.