This bundle provides a generalized ECL interface to Keras over Tensorflow.
It provides Keras / Tensorflow operations parallelized over an HPCC cluster.
Tensorflow Models are created transparently on each HPCC node, and training, evaluation and predictions are done in a distributed fashion across the HPCC cluster.
GNN is designed to handle any type of Neural Network model that can be built using Keras. This includes Classical (Dense) Neural Networks as well as Convolutional and Recursive Networks (such as LSTM), or any combination of the above.
GNN currently supports Tensorflow 2.x versions. It also supports the use of GPUs in conjunction with Tensorflow, with certain restrictions in the supported topology. Specifically:
- All servers in a cluster must have the same GPU configuration
- The number of HPCC nodes must equal the number of GPUs.
One GPU will be allocated to each HPCC node. See GNNI module documentation for details.
The Module GNNI defines the ECL interface to Keras. It supports any Keras model (Functional or Sequential), and allows models with multiple inputs and outputs.
Input to GNNI is in the form of Tensor records. The built-in Tensor module defines these Tensors and provides functions for operating on them. Tensors provide an efficient N-dimensional representation for data into and out of GNNI.
Python3 and Tensorflow must be installed on each server running HPCC Systems Platform software. Tensorflow should be installed using su so that all users can see it, and must be installed using the same version of Python3 as is embedded in the HPCC Systems platform. The file Test/SetupTest.ecl can be used to test the environment. It will verify that Python3 and Tensorflow are correctly installed on each Thor node. To Install GNN, run:
ecl bundle install https://github.com/hpcc-systems/GNN.git
The files Test/ClassicTest.ecl and ClassificationTest.ecl show annotated examples of using GNN to create a simple Classical Neural Networks using the Keras Sequential model.
The file Test/FuncModelTest.ecl shows an example of building a classical regression / classification network with multiple inputs and outputs using the Keras Functional model.
The folder Test/HARTests contains tests that show how to create more sophisticated Convolutional and Recurrent networks.
The folder Test/PretrainedModelTest contains tests that show how to use the pre-trained models.
Programmer Documentation is available at: HPCC Machine Learning Library A tutorial on installing and running GNN is available at: Generalized Neural Network Blog