🐜 NeuralNetworks

• 🛠️ Fixed:
• • a2707b8 Crucial Softmax issue: when used with ACTIVATION__PER_LAYER and not ALL_ACTIVATION_FUNCTIONS
• • 7b52a56 Potential issue with CATEGORICAL_CROSS_ENTROPY & BINARY_CROSS_ENTROPY when you USE_64_BIT_DOUBLE with REDUCE_RAM_WEIGHTS_LVL2
• ✨ Added:
• • 64dcb9d FRAM Examples
• • c572333 Example for SD migration to v3.0.0
• • dab6f55 Support for NN execution (partially) via external FRAM
• ⚙️ Improved:
• • a03b160 Unnecessary me->i_j++ logic
• • 26d2f20 Logic related to int and unsigned int
• • 2912a86 Unnecessary EEPROM-logic effecting sketch size
• • 6432955 Backpropagation algorithm, cutting flash memory usage by up to 200 bytes.
• • 9ac2b51 Prioritized "reduced-logic" over performance at FeedForward_Individual()
• ⚠️ Changed:
• • d4ce5e0 Optimized SD load() & save()

Warning

load() & save() previous implementations (although perfectly working) had significant design flaws, but the 3.0.0 release brings a much-improved versions of them. Note the breaking change! I’ve included a clear migration guide to help easily convert old NN-files to the new format via just a simple sketch. Alternatively, I'm providing limited backwards compatibility through save_old() and load_old(). However, please note that these legacy methods won't receive further updates or improvements over time.

🕸️ NeuralNetworks

• Fixed:
• • SD load(...) & save(...) for Teensy + return true
• • Conflicting macro definitions for Teensy & probably other MCUs\Boards
• Improved:
• • Removed redundant NULL-check above delete[] when REDUCE_RAM_DELETE_OUTPUTS is used

🕸️ NeuralNetworks

• Fixed:
• • #32 #39 warnings (+improvement)

🕸️ NeuralNetworks

• Added:
• • _2_ 0B00000001-optimization, disabling backpropagation when it's not automatically disabled. helps reduce the size of sketch
• Fixed:
• • Missing MULTIPLY_BY_INT_IF_QUANTIZATION in Layer::FeedForward_Individual(...) (not an issue with EEPROM or PROGMEM)
    
• Improved:
• • Ignore * _inputs during compilation when there's NO_BACKPROP

🕸️ NeuralNetworks

• Fixed:
• • Crucial mistake made in v2.3.6. Forgot to Increment i in destructor (even though I tested it before releasing it)

🕸️ NeuralNetworks

• Fixed:
• • Destructor, deallocating last-layer's outputs when not defined as REDUCE_RAM_DELETE_OUTPUTS resulting in potential unadentified behaviours
    
• Improved:
• • NN.Feedforward(...) logic, reduced program space
• • Reduced program space when NO_BACKPROP and, SD.h not used (above #include <NeuralNetwork.h>)

🕸️ NeuralNetworks

• Improved:
• • REDUCE_RAM_DELETE_OUTPUTS-optimization logic
• • Removed redundant nullification of deleted pointers

🕸️ NeuralNetworks

• Added:
• • _2_OPTIMIZE 0B00000010 aka REDUCE_RAM_STATIC_REFERENCE_FOR_MULTIPLE_NN_OBJECTS. Or in other words: REDUCE_RAM_STATIC_REFERENCE-optimization support for when using multiple NN objects in your sketch interchangeably. Meaning no more need for manual updating of NN.me ...

🕸️ NeuralNetworks

• Added:
• • int16_t&int8_t quantization support (for all storage-medias)
• • Training example with tensorflow + quantization
• • Pre-trained int8_t-quantized NN example
• Fixed:
• • #38 ESP issue with SD.h

🕸️ NeuralNetworks

• Fixed:
• • #37 issue due to binary constants marked as deprecated in new cores
• • FIXED: error: extended character X is not valid in an identifier

Warning

Due to #37 being an issue for many versions of new arduino-cores, I highly-suggest to replace binary-constants in your sketches with the prefix 0 like so: #define _1_OPTIMIZE B00000000 --> #define _1_OPTIMIZE 0B00000000.