python-neural-net

Simple neural network classifier written in Python 3 using numpy.

Supports several tuning parameters and outputs accuracy reports & graphs to visualize convergence using matplotlib.

Parameters

• x: 2-D array of training data
• y: 1-D array of target labels (0,1)
• num_layers: Number of hidden layers to use for the model (>= 1, default: 2)
• num_nodes: Number of nodes per hidden layer (>= 1, default: 2)
• lr: Learning rate (default: 0.01)
• max_iter: Number of iterations to run (default: 10,000)

Input

Training data used in the code (source: https://stackabuse.com/creating-a-neural-network-from-scratch-in-python/) are simple examples that can illustrate the neural network's ability to fit the data and make reasonable predictions not seen in the training set.

Person	Smoking	Obesity	Exercise	Diabetic
Person1	0	1	0	1
Person2	0	0	1	0
Person3	1	0	0	0
Person4	1	1	0	1
Person5	1	1	1	1

Prediction test on unseen data:

Person	Smoking	Obesity	Exercise	Diabetic
Person1	1	0	1	?

Output

TRAINING OUTPUTS:
0.989994 --> 1
0.019510 --> 0
0.020174 --> 0
0.987970 --> 1
0.978687 --> 1
Accuracy: 100.00%

PREDICTION
0.008062 --> 0

NOTE: This will vary when a random seed is implemented