Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/raphaelsenn/mnist-numpy
MNIST (Digits & Fashion) Neural Network (MLP) from scratch in raw NumPy, benchmarked against PyTorch.
https://github.com/raphaelsenn/mnist-numpy
deep-learning deep-neural-networks machine-learning mnist mnist-classification neural-network
Last synced: about 22 hours ago
JSON representation
MNIST (Digits & Fashion) Neural Network (MLP) from scratch in raw NumPy, benchmarked against PyTorch.
- Host: GitHub
- URL: https://github.com/raphaelsenn/mnist-numpy
- Owner: raphaelsenn
- License: mit
- Created: 2024-12-13T05:08:38.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2024-12-17T07:32:03.000Z (about 2 months ago)
- Last Synced: 2025-02-04T15:47:31.545Z (7 days ago)
- Topics: deep-learning, deep-neural-networks, machine-learning, mnist, mnist-classification, neural-network
- Language: Python
- Homepage:
- Size: 2.85 MB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# mnist-numpy
MNIST (Digits & Fashion) Neural Network from scratch in raw NumPy, benchmarked against PyTorch.
## Using PyTorch
##### 2-Layer MLP
* 784 Input neurons
* 512 Hidden neurons
* 10 Output neurons
#### Using Stochastic Gradient Descent and Cross-Entropy:
##### Training (lr = 0.01):
```console
Epoch: 0 Cross-Entropy: 0.4144 Accuracy: 0.8893
Epoch: 10 Cross-Entropy: 0.1618 Accuracy: 0.9552
Epoch: 20 Cross-Entropy: 0.1011 Accuracy: 0.9722
Epoch: 30 Cross-Entropy: 0.0714 Accuracy: 0.9811
Epoch: 40 Cross-Entropy: 0.0542 Accuracy: 0.9861
Epoch: 50 Cross-Entropy: 0.0422 Accuracy: 0.9901
Epoch: 60 Cross-Entropy: 0.0346 Accuracy: 0.9925
Epoch: 70 Cross-Entropy: 0.0278 Accuracy: 0.9946
Epoch: 80 Cross-Entropy: 0.0234 Accuracy: 0.9956
Epoch: 90 Cross-Entropy: 0.0194 Accuracy: 0.9970
```
##### Testing:
```console
Accuracy on Testing set: 0.9812
```
## Using NumPy
##### 2-Layer MLP
* 784 Input neurons
* 512 Hidden neurons
* 10 Output neurons
##### Training (lr = 0.095):
```console
Epoch: 0 Cross-Entropy: 0.3059 Accuracy: 0.9133
Epoch: 10 Cross-Entropy: 0.1696 Accuracy: 0.9520
Epoch: 20 Cross-Entropy: 0.1355 Accuracy: 0.9615
Epoch: 30 Cross-Entropy: 0.1185 Accuracy: 0.9666
Epoch: 40 Cross-Entropy: 0.1089 Accuracy: 0.9685
Epoch: 50 Cross-Entropy: 0.0996 Accuracy: 0.9716
Epoch: 60 Cross-Entropy: 0.0920 Accuracy: 0.9743
Epoch: 70 Cross-Entropy: 0.0884 Accuracy: 0.9746
Epoch: 80 Cross-Entropy: 0.0824 Accuracy: 0.9767
Epoch: 90 Cross-Entropy: 0.0881 Accuracy: 0.9727
```
##### Testing:
```console
Accuracy on Testing set: 0.9615
```
## License
* MNIST Digits: Public domain, created by Yann LeCun and collaborators.
* Fashion-MNIST: MIT License, created by Zalando Research.