Implemented a fully connected neural network in pure NumPy and compared the performance with Pytorch on the Iris and MNIST dataset.
# use nn implemented here
python train.py
# use nn in pytorch
python pytorch_train.py| nn-numpy | nn-pytorch | |
|---|---|---|
| Iris | 0.978 | 0.956 |
| MNIST | 0.943 | 0.974 |