Fast Artificial Neural Network (FANN)¶
FANN, for Fast Artificial Neural Network, is a PECL extension that wraps the FANN C library, a lightweight library for creating, training, and running multilayer feedforward artificial neural networks.
The extension exposes functions to build a network with a chosen number of layers and neurons, train it against a data set using algorithms such as backpropagation, and then run inference on new inputs. Networks can be saved to and loaded from disk, so a model can be trained once and reused across requests without retraining.
FANN predates the current generation of PHP machine-learning libraries, and its API is comparatively low-level: it does not provide dataset preprocessing, model selection, or GPU acceleration. It remains useful for small, fast, embeddable networks where pulling in a heavier stack such as TensorFlow via a bridge is unnecessary.
<?php
// Create a network: 2 inputs, one hidden layer of 3 neurons, 1 output.
$ann = fann_create_standard(3, 2, 3, 1);
fann_set_activation_function_hidden($ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output($ann, FANN_SIGMOID_SYMMETRIC);
// Train on a data file of input/output pairs (FANN training file format).
fann_train_on_file($ann, 'xor.data', 5000, 500, 0.0001);
$output = fann_run($ann, [1, -1]);
print_r($output);
fann_destroy($ann);
?>
See also FANN library.
Related : Machine Learning, Algorithm, GNU Multiple Precision (GMP)
