v2.5.0

• Improve audio python library and update audio training tutorial
• Add support for non-unrolled loops in the Value library
• Add new methods to Python API to support FastGRNN models
• Add easy node-creation functions
• Fix bugs in ReinterpretLayoutNode
• Fix importing of DenseNet models
• Improve FFT and MelFilterBank nodes so that non-power of 2 sized input buffers can be processed
• Fix darknet importer
• Add shuffling of test and validation sets to audio train_classifier
• Update OpenBLASSetup.cmake so it can find a locally built version of OpenBLAS
• Fix compiling for cortex-m4 targets
• Add a link back to the original model file name in our emitted header file
• Clean up Python API, removing "TensorShape" specific API in favor of the more general PortMemoryLayout
• Fix bug in Microphone and WavReader classes so they properly handle auto-resampling of audio with multiple channels
• Add an -auto_scale option to audio scripts to control whether audio is scaled or not before featurization
• Add ability to mix noise with audio during make_dataset process
• Add support for shuffling and cleaning the audio training list
• Add many new options to audio train_classifier, including ability to control learning rate schedulers
• Fix link error when building on a Mac
• Fix exception in DetectLowPrecisionConvolutionTransformation when running Compile in Debug mode