Preparing and testing the quantized TFLite model
As we know from Chapter 3, Building a Weather Station with TensorFlow Lite for Microcontrollers, the model requires quantization to 8 bits to run more efficiently on a microcontroller. However, how do we know if the model can fit into the Arduino Nano? Furthermore, how do we know if the quantized model preserves the accuracy of the floating-point variant?
These questions will be answered in this recipe, where we will show how to evaluate the program memory utilization and the accuracy of the quantized model generated by the TFLite converter. After analyzing the memory usage and accuracy validation, we will convert the TFLite model to a C-byte array.
The following Colab notebook (the Preparing and testing the quantized TFLite model section) contains the code referred to in this recipe:
prepare_model.ipynb:
