https://github.com/nathanielsimard/numpy-mnist-classifier

Feed forward neural network using Numpy for MNIST classification.
https://github.com/nathanielsimard/numpy-mnist-classifier

deep-learning machine-learning mnist mnist-classification mnist-classifier numpy

Last synced: 21 days ago
JSON representation

Feed forward neural network using Numpy for MNIST classification.

• Host: GitHub
• URL: https://github.com/nathanielsimard/numpy-mnist-classifier
• Owner: nathanielsimard
• Created: 2019-01-11T00:20:49.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2019-02-01T01:14:14.000Z (almost 6 years ago)
• Last Synced: 2024-12-05T17:47:04.432Z (about 1 month ago)
• Topics: deep-learning, machine-learning, mnist, mnist-classification, mnist-classifier, numpy
• Language: Python
• Homepage:
• Size: 490 KB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Numpy Mnist Classifier

This project was done for learning purpose.

The goal was to implement a deep neural network to do supervised learning.

The MNIST data set was used due to its small size, making the learning process fast enough on a personal laptop.

To try the project you first need to install the dependencies, note that `python3` is required.

```bash

pip install -r requirements/basic.txt

```

## Usage

It only takes a small amount of code to test some models :

```python

    from classifier import nn, training

    from data import mnist

    # The MNIST data set will be automatically downloaded and cached.

    training_data, validation_data, test_data = mnist.load()

    # Create a Neural Network with one hidden layer.

    model = nn.NeuralNetwork([784, 30, 10], learning_rate=0.02, batch_size=50)

    # Train the model with early stopping regularization.

    model_training = training.EarlyStoppingRegularization(model,

                                                          training_data,

                                                          validation_data,

                                                          test_data,

                                                          max_steps_without_progression=2)

    model_training.train()

    # It is possible to save the result which serializes the model and create a report.

    result.save('models/mnist-example')

    # It is possible to load the trained model for futur uses.

    model_trained = nn.load('models/mnist-example/model.pkl)

```

## Report Example

## Model

- Layers : [784, 30, 10]

- Activation : sigmoid

- Learning Rate : 0.02

- Batch Size : 50

## Training

- Method : early stopping regularization

- Epochs : 69

## Data

Size :

- Training : 50000

- Test : 10000

- Validation : 10000

### Sample

![graph](./assets/sample.png)

## Accuracy and Loss

|   | Training | Test |

|---|---|---|

| Accuracy | 97.392% | 95.430%  |

| Loss | 0.046 | 0.081 |

![graph](./assets/result.png)