v1.2.0

• PlatformIO as project type
• Create torch models and load weights to esp32 usign generated include file
• Added inference only mode
• Fixed minor on device training issues
• Optimized data types to uint, where possible
• Added Python XOR torch example model for generating header file for esp32

v1.1.0

V1.1.0

• Added Xavier Initialization to optimizes weight distribution for faster training.

v1.0.0

V1.0.0

• Choose from multiple activation functions (Sigmoid, Relu, Softmax, TanH, LeakyRelu)
• Loss Calculator with output logging
• Use TanH and Softmax for the XOR example code
• Show output at every 10th epoch of total epochs, with a minimum interval of 1 epoch

V0.1.1

V0.1.1

• Fix spelling in logging and visual output code

0.1.0

V0.1.0

First functional version of a neural network fully trained and making predictions directly on an Arduino or ESP32.