fann_create_train_from_callback

(PECL fann >= 1.0.0)

fann_create_train_from_callback — Creates the training data struct from a user supplied function

説明

fann_create_train_from_callback(
    int $num_data,
    int $num_input,
    int $num_output,
    callable $user_function
): resource

Creates the training data struct from a user supplied function. As the training data are numerable (data 1, data 2...), the user must write a function that receives the number of the training data set (input, output) and returns the set.

パラメータ

num_data

The number of training data

num_input

The number of inputs per training data

num_output

The number of ouputs per training data

user_function

The user supplied function with following parameters:

num - The number of the training data set
num_input - The number of inputs per training data
num_output - The number of ouputs per training data

The function should return an associative array with keys input and output and two array values of input and output.

戻り値

成功した場合にニューラルネットワークトレーニングリソース、エラー時に false を返します。

例

例1 fann_create_train_from_callback() example

<?php
function create_train_callback($num_data, $num_input, $num_output) {
    return array(
        "input" => array_fill(0, $num_input, 1),
        "output" => array_fill(0, $num_output, 1),
    );
}

$num_data = 3;
$num_input = 2;
$num_output = 1;
$train_data = fann_create_train_from_callback($num_data, $num_input, $num_output, "create_train_callback");
if ($train_data) {
    // Do something with $train_data
}
?>

注意

注意:
この関数は、fann拡張モジュールが libfann > 2.2 以上でビルドされている場合にだけ利用可能です。

参考

fann_read_train_from_file() - Reads a file that stores training data
fann_train_on_data() - Trains on an entire dataset for a period of time
fann_destroy_train() - Destructs the training data
fann_save_train() - Save the training structure to a file

Improve This Page

Learn How To Improve This Page • Submit a Pull Request • Report a Bug

＋add a note

User Contributed Notes 1 note

down

geekgirljoy at gmail dot com ¶

8 years ago

This code can be used to read training data from MySQL rather than a text file.

<?php

// MySQL for This Example:
/*
CREATE TABLE `TrainingSets` (
  `ID` int(11) NOT NULL,
  `Name` varchar(150) COLLATE utf8mb4_unicode_ci NOT NULL,
  `TrainingData` text COLLATE utf8mb4_unicode_ci NOT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

ALTER TABLE `TrainingSets` ADD PRIMARY KEY (`ID`);
  
INSERT INTO `TrainingSets` (`ID`, `Name`, `TrainingData`) VALUES(1, 'XOR', '-1 -1\n-1\n-1 1\n1\n1 -1\n1\n1 1\n-1');

ALTER TABLE `TrainingSets` MODIFY `ID` int(11) NOT NULL AUTO_INCREMENT, AUTO_INCREMENT=2;
*/

// This function calls pulls the TrainingData from MySQL
function get_training_data_from_db($id) {
    $table_name = "TrainingSets";
    $field = "TrainingData";
    $connection=mysqli_connect("host","username","password","database"); // change to your DB credentials
    $result=mysqli_query($connection,"SELECT $field FROM $table_name");
    $data=mysqli_fetch_assoc($result);
    mysqli_close($connection);

    return $data[$field];
}

// This function prepares the newline delimited data to be handed off to FANN
/*
Example of "newline delimited data" (like XOR in a Plain Text File) stored in MySQL:
-1 -1
-1
-1 1
1
1 1
-1
1 -1
1
*/
function prepare_data_from_db($training_data) {
    $training_data = explode( "\n", $training_data ); // convert training data rows to array
    $num_data = count($training_data);
    
    // Sift the data and split inputs and outputs 
    for($i=0;$i<$num_data;$i++) {
      if($i % 2) { // $training_data[$i] is Output
       $training_data['outputs'][] = explode( " ", $training_data[$i]);
      }else{ // $training_data[$i] is Input
       $training_data['inputs'][] = explode( " ", $training_data[$i]);
      }
    }
    // remove the unsifted data
    foreach ($training_data as $key => $value) {
        if (is_numeric($key)) {
            unset($training_data[$key]);
        }
    }
    return $training_data; // returned the prepaired associative array
}

// This function hands the prepared data over to FANN
function create_train_callback($num_data, $num_input, $num_output) {
    global $training_data;
    global $current_dataset;
   
    $dataset = array("input" => $training_data['inputs'][$current_dataset],
                    "output" => $training_data['outputs'][$current_dataset]);
    $current_dataset++;

    return $dataset;
}

// Initialize the program variables
$record_id = 1; // the 'ID' for the training data in MySQL
$current_dataset = 0;
$num_input = 2;
$num_output = 1;
$num_layers = 3;
$num_neurons = 3;
$desired_error = 0.001;
$max_epochs = 500000;
$epochs_between_reports = 1000;

$training_data = get_training_data_from_db($record_id); // Get the Training Data from MySQL
$training_data = prepare_data_from_db($training_data); // Prepare the data
$num_data = count($training_data["input"]); // How many sets are there?

// Hand the data over to FANN
$train_data = fann_create_train_from_callback($num_data, $num_input, $num_output, "create_train_callback");

// Test for $train_data 
if ($train_data) {
    
    // Create $ann
    $ann = fann_create_standard($num_layers, $num_input, $num_neurons, $num_output);

    // Test for $ann
    if ($ann) {
        fann_set_activation_function_hidden($ann, FANN_SIGMOID_SYMMETRIC);
        fann_set_activation_function_output($ann, FANN_SIGMOID_SYMMETRIC);

        // Train XOR ANN with training data obtainied from MySQL
        if (fann_train_on_data($ann, $train_data, $max_epochs, $epochs_between_reports, $desired_error)){
           print('XOR trained.<br>' . PHP_EOL);

           // Test $ann
           $input = array(-1, 1);
           $calc_out = fann_run($ann, $input);
           printf("xor test (%f,%f) -> %f\n", $input[0], $input[1], $calc_out[0]);
           
           // destore $ann 
           fann_destroy($ann);
        }
    }
}
?>

＋add a note