Syntra is a lightweight neural network framework built in C++ with support for model saving, training from CSV datasets, multiple activation functions, and early stopping. It is basically zeus nn V2
- Feedforward neural networks
- Activation functions: Sigmoid, ReLU, Tanh
- Model saving/loading to
.txt(I could have made a proprietary format, but we have enough of those) - CSV-based dataset input (optional)
- Training logs to file
- Early stopping
- It exists :D
g++ -std=c++17 -O2 -o syntra main.cpp./syntra [csv_file.csv] [flags]| Flag | Description |
|---|---|
--epochs |
Training epochs (default: 1000) |
--lr |
Learning rate (default: 0.1) |
--batch |
Batch size (default: 4) |
--activation |
sigmoid, relu, or tanh |
--earlystop |
Patience in epochs (default: 20) |
--save |
Output model file |
--log |
Output log file |
./syntra --epochs 1500 --lr 0.05 --batch 2 --activation tanh --earlystop 50 --save model.txt --log out.log./syntra dataset.csv --epochs 500 --lr 0.01 --activation reluEach CSV row should contain input values followed by output labels.
Licensed under the BSD 3-Clause License.