Neural Network subpackage (battelle.nn)

Neuron(weights, bias, activation_function)	An instance of a neuron
Layer(neuron_count[, activation_function, ...])	An instance of a layer
NeuralNetwork()	Instance of a neural network.
ActivationFunction(f, fp)	An instance of a activation function
NeuralNetwork.train(x, y[, learning_rate])	Train the neural network with dataset of features x and of corresponding labels y.
NeuralNetwork.predict(x)	Makes a prediction for features x
NeuralNetwork.get_metrics(features, labels, ...)	Compute metrics for a binary classification using data from features and labels.

This module allows the user to create, train and visualize simple neural networks.

Here is a simple example of binary classification:

import numpy as np
import matplotlib.pyplot as plt

from battelle.nn import *


# List of features
x = np.random.random(size=(300, 2)) - 0.5
x = [list(xi) for xi in x]

# List of corresponding labels (xor)
y = np.array([np.sign(np.prod(k)) for k in x])

# Define neural network
nn = NeuralNetwork()
nn.add_layer(Layer(5, "tanh", input_dim=2))
nn.add_layer(Layer(4, "tanh"))
nn.add_layer(Layer(1, "tanh"))

# Train the neural network in epochs
NUM_EPOCHS = 100
for i in range(NUM_EPOCHS):
    print(f"Epoch {i+1}/{NUM_EPOCHS}")
    nn.train(x, y, learning_rate=0.03)
    print(f"Loss: {np.mean([(nn.predict(x[k]) - y[k])**2 for k in range(len(x))])}")

# Plot training data
plt.subplot(1, 2, 1)
for i, xi in enumerate(x):
    if y[i] == 1:
        plt.plot(xi[0], xi[1], "xr")
    else:
        plt.plot(xi[0], xi[1], "xb")
plt.title("Testing data")

# Plot neural net result
plt.subplot(1, 2, 2)
xrange = np.linspace(-0.5, 0.5, 20)
yrange = np.linspace(-0.5, 0.5, 20)
for x in xrange:
    for y in yrange:
        if nn.predict([x, y]) > 0:
            plt.plot(x, y, "xr")
        else:
            plt.plot(x, y, "xb")

plt.title("Neural net result")
plt.show()

This code creates the following graph: