Perceptron Neural Network Application

This repository contains a neural network application implementing a multilayer perceptron for image recognition tasks, such as digit recognition using the MNIST dataset. The app allows you to train, predict, save, and load models easily.

---

Features

• Create Custom Models: Define the number of neurons in hidden layers.
• Train on Custom Data: Train the model using your own dataset.
• Train on Random Images: Train the model on randomly selected images.
• Make Predictions: Predict the class label for an input image.
• Save and Load Models: Save the current model to disk and load it for later use.
• Load Default Model: Reset the model to the default pretrained configuration.
• Reset Training: Revert the current model to its state before any training.

---

Installation

1. Clone the repository:

   git clone https://github.com/mcaramba563/final_project_git.git
   cd final_project_git

2. Install dependencies:

   pip install -r requirements.txt

Usage

Start the App

Run the app in interactive mode:

python main.py

Features and Commands

1. Create a Custom Model

Create a model with specific hidden layer sizes:

make_custom_model layer1_size layer2_size ...

Example:

make_custom_model 400 256 128

2. Train on Custom Data

Train the model on labeled data from a specified file(the file containes pathes to images and labels):

train path_to_dataset_file number_of_epochs learning_rate

Example:

train dataset/train_data.txt 5 0.01

where dataset/train_data.txt:

images/mnist_png/test/0/10.png 0

3. Train on Random Images from default dataset

Train the model on randomly selected images:

train_on_random_images number_of_images number_of_epochs learning_rate

Example:

train_on_random_images 20000 2 0.01

4. Make Predictions

Predict the class of a given image:

predict path_to_image

Example:

predict images/mnist_png/test/0/10.png

5. Save the Model

Save the current model to a file:

save_model path_to_model_file

Example:

save_model models/my_model.npy

6. Load a Custom Model

Load a previously saved model:

load_custom_model path_to_model_file

Example:

load_custom_model models/my_model.npy

7. Load Default Model

Reset the model to its default pretrained state:

load_default_model

8. Reset Training

Revert the current model to its initial state before training:

reset_training

9. Exit the Application

Exit the interactive mode:

exit

Example Workflow

1. Start the app:

   python main.py

2. Create a custom model with hidden layers of sizes 400, 256, and 128:

   make_custom_model 400 256 128

3. Train on custom data for 5 epochs with a learning rate of 0.01:

   train dataset/train_data.txt 5 0.01

4. Save the trained model:

   save_model models/my_trained_model.npy

5. Load the default pretrained model:

   load_default_model

6. Predict the class of a test image:

   predict images/mnist_png/test/0/10.png

7. Exit the app:

   exit