Forward Feeding Backpropagating Neural Network
Projects | | Links:
This project involved constructing a Forward Feeding Back Propagating Neural Network using Sigmoid activation in Python.
This project involved implementing a forward feeding, backpropagating neural network in Python, and demonstrates core machine learning principles. The network architecture is a basic neural network with input, hidden and output layers, intaking numerical data. The forward feed incorporates a sigmoid activation function to introduce non-linearity and the loss is calculated using a Root Mean Squared Error (RMSE) loss metric. The overall network is constructed manually using a doubly linked list.
Key Features and Highlights
- Input Handling: Data is preprocessed and split into train and test sets.
- Forward Propagation: The network computes weighted sums and applies sigmoid activation to propagate data through the hidden and output layers.
- Backpropagation: Errors are backpropagated with weight adjustments.
- Error Metric: The RMSE was used to evaluate and monitor the performance of the network during training.
Tools Used
- Numpy
- JSON
- PyQt5
- Matplotlib
Experimentation & Results
The network was tested using the well known Iris dataset and the Sine dataset.
Iris Dataset
The network was trained twice using the Iris dataset; for 10001 epochs at a learning rate of 0.7 and for 20001 epochs at a learning rate of 0.3.
Iris dataset training results
Sine Dataset
The Sine dataset was used to evaluate the impact of network of varying sizes; evaluating across layers and nodes per layer.
Sine dataset network size experiment results
