Located in /ANN_Network.php (line 51)
ANN_Filesystem | --ANN_Network
[708] | protected | static | string : | getDefaultFilename () |
[1096] | public | static | void : | loadFromFile ([string $filename = null]) |
[1369] | public | static | ANN_Network : | loadFromHost (string $username, string $password, string $host) |
[352] | protected | void : | activate () | |
[1430] | protected | void : | adjustLearningRate () | |
[159] | public | ANN_Network : | __construct ([integer $numberOfHiddenLayers = 2], [integer $numberOfNeuronsPerLayer = 4], [integer $numberOfOutputs = 1]) | |
[1054] | protected | void : | calculateMaxTrainingLoops () | |
[312] | protected | void : | createHiddenLayers (integer $numberOfHiddenLayers, integer $numberOfNeuronsPerLayer) | |
[339] | protected | void : | createOutputLayer (integer $numberOfOutputs) | |
[1401] | protected | void : | detectOutputType () | |
[664] | protected | integer : | getCountInputs () | |
[1281] | protected | float : | getNetworkError () | |
[466] | protected | integer : | getNextIndexInputsToTrain ([boolean $reset = FALSE]) | |
[1141] | public | integer : | getNumberHiddenLayers () | |
[1155] | public | integer : | getNumberHiddens () | |
[1126] | public | integer : | getNumberInputs () | |
[1170] | public | integer : | getNumberOutputs () | |
[250] | public | array : | getOutputs () | |
[288] | public | array : | getOutputsByInputKey (integer $keyInput) | |
[497] | public | integer : | getTotalLoops () | |
[508] | protected | boolean : | isEpoch () | |
[573] | protected | boolean : | isTrainingComplete () | |
[611] | protected | boolean : | isTrainingCompleteByEpoch () | |
[628] | protected | boolean : | isTrainingCompleteByInputKey (integer $keyInput) | |
[1262] | protected | void : | logNetworkErrors () | |
[1207] | public | void : | logNetworkErrorsToFile (string $filename) | |
[1225] | protected | void : | logWeights () | |
[1188] | public | void : | logWeightsToFile (string $filename) | |
[752] | public | void : | printNetwork ([integer $level = 0]) | |
[834] | protected | void : | printNetworkDetails1 () | |
[995] | protected | void : | printNetworkDetails2 () | |
[1112] | public | void : | saveToFile ([string $filename = null]) | |
[1339] | public | void : | saveToHost (string $username, string $password, string $host) | |
[1517] | public | void : | setBackpropagationAlgorithm ([integer $algorithm = self::ALGORITHM_BACKPROPAGATION]) | |
[1463] | public | void : | setDynamicLearningRate ([boolean $dynamicLearningRate = TRUE]) | |
[189] | public | void : | setInputs (array $inputs) | |
[233] | protected | void : | setInputsToTrain (array $inputs) | |
[534] | public | void : | setLearningRate ([float $learningRate = 0.5]) | |
[1067] | public | void : | setMaxTrainingLoopsFactor ([integer $maxTrainingLoopsFactor = 230]) | |
[555] | public | void : | setMomentum ([float $momentum = 0.95]) | |
[1569] | public | void : | setOutputErrorTolerance ([float $outputErrorTolerance = 0.02]) | |
[213] | public | void : | setOutputs (array $outputs) | |
[721] | protected | void : | setOutputType ([string $type = 'linear']) | |
[1551] | public | void : | setQuickPropMaxWeightChangeFactor ([float $quickPropMaxWeightChangeFactor = 2.25]) | |
[1495] | public | void : | setWeightDecay ([float $weightDecay = 0.05]) | |
[1479] | public | void : | setWeightDecayMode ([boolean $weightDecayMode = TRUE]) | |
[397] | public | boolean : | train () | |
[1304] | public | ANN_Network : | trainByHost (string $username, string $password, string $host) | |
[683] | protected | void : | training (array $outputs) | |
[1081] | public | void : | __wakeup () |
Adjusting learning rate dynamically
If network error of current epoch is higher than the network error of the previous epoch the learning rate is adjusted by minus 1 per cent of current learning rate. Otherwise the learning rate is adjusted by plus 1 per cent of current learning rate. So, learning rate increases faster than decreasing does. But if learning rate reaches 0.9 it switches back to 0.5 to avoid endless training. The lowest learning rate is 0.5 also to avoid endless training.
Log network errors while training in CSV format
Log weights while training in CSV format
Selecting propagation algorithm
EXPERIMENTAL
Setting the learning rate disables dynamic learning rate automatically.
Setting the percentage of output error in comparison to the desired output
Parameter setting for QuickProp algorithm
EXPERIMENTAL
Inherited From ANN_Filesystem
ANN_Filesystem::loadFromFile()
ANN_Filesystem::saveToFile()
Back propagation (default)
Individual learning rate (EXPERIMENTAL)
iRProp- (EXPERIMENTAL)
iRProp+ (EXPERIMENTAL)
Quick propagation (EXPERIMENTAL)
RProp (EXPERIMENTAL)
RProp- (EXPERIMENTAL)
RProp+ (EXPERIMENTAL)
Documentation generated on Tue, 16 Dec 2008 18:36:07 +0100 by phpDocumentor 1.4.1