ff-neural-network
A Python module for the generation and training of an entry-level feedforward neural network.
This repository serves as a repurposing of a 2019 project I did as an initiation into machine learning.
Usage
Creating a network:
network = Network(layer_sizes, bias_value)
layer_sizes: Number of neurons in each layer.
Ex: [2, 5, 1] will generate a network that can be visualized as such:
bias_value: Value of the bias nodes (standardized at 1):
Bias nodes are added to a feed-forward neural network to help facilitate
learning patterns. They function like an input node that always
produces a value of 1.0 or other constant.
network.randomize()
- Initializes the weights between all neurons with a random value.
network.train(input_data, target_data, learning_rate)
-
input_data: The input data, a good approach is to have it normalized into a proper range. -
target_data: The data that the model learns from. -
learning_rate: Controls how quickly or slowly the network model learns the problem.
Example
For an (output = X) pattern learning data:
| X | Y | Target |
|---|---|---|
| 0 | 1 | 0 |
| 1 | 0 | 1 |
| 1 | 1 | 1 |
Which should lead to:
| X | Y | Output |
|---|---|---|
| 0 | 0 | ~0 |
from network import Network
from data_set import DataSet
# Initializing a network with a 2-2-1 structure
network = Network([2, 2, 1], 1.0)
# Randomizing initial weights between all neurons
network.randomize()
# Initializing data_set with input and output training data
inputs = [[0, 1], [1, 0], [1, 1]]
outputs = [[0], [1], [1]]
data_set = DataSet(inputs, outputs)
# Training the network for 10000 intervals
for _ in range(10000):
for index in range(0, data_set.get_size()):
network.train(data_set.get_input(index),data_set.get_target(index), 1.0)
# Printing output prediction for input = [0, 0]
print(network.calculate_outputs([0, 0]))
We get :
output : [0.0023672395614975253]
