README

coral-media/ext-fann is a PHP extension that wraps the Fast Artificial Neural Network (FANN) C library using vendored sources (lib/libfann).

Requirements

• PHP >= 8.1
• A C toolchain (phpize, make, compiler)
• No system libfann-dev required (vendored build)

Install (PIE)

From the extension root:

pie install --with-fann

Build From Source (manual)

phpize
./configure --with-fann
make -j"$(nproc)"
make test

The extension module will be built as fann.

Minimal Example

<?php
$ann = fann_create_standard(3, 2, 3, 1);
fann_set_training_algorithm($ann, FANN_TRAIN_INCREMENTAL);
fann_set_learning_rate($ann, 0.7);
fann_set_activation_function_hidden($ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output($ann, FANN_SIGMOID_SYMMETRIC);

for ($epoch = 0; $epoch < 200; $epoch++) {
    fann_train($ann, [0.0, 0.0], [0.0]);
    fann_train($ann, [0.0, 1.0], [1.0]);
    fann_train($ann, [1.0, 0.0], [1.0]);
    fann_train($ann, [1.0, 1.0], [0.0]);
}

$out = fann_run($ann, [0.0, 1.0]);
var_dump($out);

fann_destroy($ann);

Tests

Run all PHPT tests:

make test

API Coverage

Implemented functions are grouped below. Full API reference: docs/api/index.md.

Network Creation, Persistence, and Lifecycle

• fann_create_standard, fann_create_standard_array
• fann_create_sparse, fann_create_sparse_array
• fann_create_shortcut, fann_create_shortcut_array
• fann_create_from_file, fann_copy, fann_destroy
• fann_save, fann_save_to_fixed
• fann_print_parameters, fann_print_connections

Execution and Basic Training

• fann_run, fann_test
• fann_train, fann_train_on_data, fann_train_on_file
• fann_train_epoch, fann_test_data
• fann_cascadetrain_on_data, fann_cascadetrain_on_file

Training Data I/O and Dataset Utilities

• fann_create_train, fann_read_train_from_file
• fann_save_train, fann_save_train_to_fixed
• fann_destroy_train
• fann_shuffle_train_data
• fann_scale_input_train_data, fann_scale_output_train_data, fann_scale_train_data
• fann_merge_train_data, fann_duplicate_train_data, fann_subset_train_data
• fann_length_train_data, fann_num_input_train_data, fann_num_output_train_data

Training Metrics and Error Handling

• fann_get_MSE, fann_reset_MSE, fann_get_bit_fail
• fann_get_errno, fann_reset_errno, fann_reset_errstr
• fann_get_errstr, fann_print_error, fann_set_error_log

Training Algorithm and Hyperparameters

• fann_set_training_algorithm, fann_get_training_algorithm
• fann_set_learning_rate, fann_get_learning_rate
• fann_set_learning_momentum, fann_get_learning_momentum
• fann_set_train_error_function, fann_get_train_error_function
• fann_set_train_stop_function, fann_get_train_stop_function
• fann_set_bit_fail_limit, fann_get_bit_fail_limit
• fann_set_quickprop_decay, fann_get_quickprop_decay
• fann_set_quickprop_mu, fann_get_quickprop_mu
• fann_set_rprop_increase_factor, fann_get_rprop_increase_factor
• fann_set_rprop_decrease_factor, fann_get_rprop_decrease_factor
• fann_set_rprop_delta_min, fann_get_rprop_delta_min
• fann_set_rprop_delta_max, fann_get_rprop_delta_max
• fann_set_rprop_delta_zero, fann_get_rprop_delta_zero
• fann_set_sarprop_weight_decay_shift, fann_get_sarprop_weight_decay_shift
• fann_set_sarprop_step_error_threshold_factor, fann_get_sarprop_step_error_threshold_factor
• fann_set_sarprop_step_error_shift, fann_get_sarprop_step_error_shift
• fann_set_sarprop_temperature, fann_get_sarprop_temperature

Activation Functions and Steepness

• fann_set_activation_function_hidden, fann_set_activation_function_output
• fann_set_activation_function, fann_set_activation_function_layer
• fann_get_activation_function
• fann_set_activation_steepness, fann_set_activation_steepness_layer
• fann_set_activation_steepness_hidden, fann_set_activation_steepness_output
• fann_get_activation_steepness

Network Introspection, Topology, and Weights

• fann_get_num_input, fann_get_num_output, fann_get_num_layers
• fann_get_total_neurons, fann_get_total_connections
• fann_get_network_type, fann_get_connection_rate
• fann_get_layer_array, fann_get_bias_array, fann_get_connection_array
• fann_set_weight_array, fann_set_weight
• fann_randomize_weights, fann_init_weights

Input/Output Scaling (Runtime + Parameter Setup)

• fann_set_input_scaling_params, fann_set_output_scaling_params, fann_set_scaling_params
• fann_clear_scaling_params
• fann_scale_train, fann_descale_train
• fann_scale_input, fann_scale_output, fann_descale_input, fann_descale_output

Cascade Training Controls

• fann_get_cascade_output_change_fraction, fann_set_cascade_output_change_fraction
• fann_get_cascade_output_stagnation_epochs, fann_set_cascade_output_stagnation_epochs
• fann_get_cascade_candidate_change_fraction, fann_set_cascade_candidate_change_fraction
• fann_get_cascade_candidate_stagnation_epochs, fann_set_cascade_candidate_stagnation_epochs
• fann_get_cascade_weight_multiplier, fann_set_cascade_weight_multiplier
• fann_get_cascade_candidate_limit, fann_set_cascade_candidate_limit
• fann_get_cascade_max_out_epochs, fann_set_cascade_max_out_epochs
• fann_get_cascade_min_out_epochs, fann_set_cascade_min_out_epochs
• fann_get_cascade_max_cand_epochs, fann_set_cascade_max_cand_epochs
• fann_get_cascade_min_cand_epochs, fann_set_cascade_min_cand_epochs
• fann_get_cascade_num_candidates
• fann_get_cascade_activation_functions_count, fann_get_cascade_activation_functions, fann_set_cascade_activation_functions
• fann_get_cascade_activation_steepnesses_count, fann_get_cascade_activation_steepnesses, fann_set_cascade_activation_steepnesses
• fann_get_cascade_num_candidate_groups, fann_set_cascade_num_candidate_groups

Constants

• Full exposure of currently registered FANN_* constants (training algorithms, activation functions, stop/error functions, net types, and error codes).

Vendored Dependency

• FANN source is vendored under lib/libfann
• License file: lib/libfann/COPYING.txt