https://github.com/akash-nath29/autoencoder-from-scratch
Implementation of an Autoencoder from scratch only using numpy
https://github.com/akash-nath29/autoencoder-from-scratch
Last synced: about 2 months ago
JSON representation
Implementation of an Autoencoder from scratch only using numpy
- Host: GitHub
- URL: https://github.com/akash-nath29/autoencoder-from-scratch
- Owner: Akash-nath29
- Created: 2025-01-16T09:24:31.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-01-16T09:39:37.000Z (4 months ago)
- Last Synced: 2025-01-16T10:56:14.409Z (4 months ago)
- Language: Jupyter Notebook
- Size: 4.22 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Autoencoder from Scratch
This project implements an autoencoder from scratch using NumPy. The autoencoder is trained on the MNIST dataset to compress and reconstruct images.
## Project Structure
```
autoencoder_from_scratch/
│
├── data/
│ ├── mnist_train.csv
│ ├── mnist_test.csv
│
├── functions/
│ ├── activations.py
│ ├── loss_functions.py
│ ├── utils.py
│
├── model/
│ ├── autoencoder.py
│ ├── encoder.py
│ ├── decoder.py
│ ├── optimizer.py
│
├── train.py
├── dataloader.py
├── model.ipynb
├── README.md
```
## Hyperparameters
The hyperparameters for the model can be adjusted in `train.py`:
```python
input_size = 784
hidden_size = 256
output_size = 784
learning_rate = 0.001
batch_size = 64
epochs = 300
```
## Model Performance
### Loss Over Epochs
The loss over epochs shows how the model's loss decreases as training progresses. This indicates that the model is learning to reconstruct the input images more accurately over time.
### Accuracy Over Epochs
The accuracy over epochs shows how the model's accuracy improves as training progresses. This indicates that the model is becoming better at reconstructing the input images.
### Output
The output section shows examples of the original, encoded, and reconstructed images. This demonstrates the effectiveness of the autoencoder in compressing and reconstructing the input images.
## Installation
1. Clone the repository:
```sh
git clone https://github.com/yourusername/autoencoder_from_scratch.git
cd autoencoder_from_scratch
```
2. Install the required packages:
```sh
pip install numpy pandas matplotlib
```
3. Ensure the MNIST dataset CSV files are in the `data/` directory.
## Usage
### Training the Autoencoder
To train the autoencoder, run the `train.py` script:
```sh
python train.py
```
### Jupyter Notebook
You can also explore the project using the provided Jupyter Notebook `model.ipynb`.
### Evaluating the Model
After training, the model can be evaluated on the test dataset. The evaluation script is included in `train.py` and `model.ipynb`.
### Visualizing Results
The notebook and script include code to visualize the original, encoded, and reconstructed images.
## Saving the Model
The trained model weights are saved to `model_weights.json` after training.
## License
This project is licensed under the MIT License.