Trainer API

The Machine Learning Trainer API provides functionality for training neural network models.

Remark: In order to access files, a proper privilege has to be defined additionally:

Since: 7.0

Table of Contents

Summary of Interfaces and Methods

Interface Method
MachineLearningTrainer
Dataset createFileDataset (Path train, optional Path? validate, optional Path test)
Model createModel (optional Path configPath)
Layer
void setProperty (DOMString name, any value)
void dispose ()
Optimizer
void setProperty (DOMString name, any value)
void dispose ()
Dataset
void setProperty (DOMString name, any value, DatasetMode mode)
void dispose ()
CompileOptions
RunOptions
Model
void compile (optional CompileOptions? options)
void run (RunOptions? options, SuccessCallback successCallback, optional ErrorCallback? errorCallback)
DOMString summarize (optional VerbosityLevel? level)
void addLayer (Layer layer)
void setDataset (Dataset dataset)
void setOptimizer (Optimizer optimizer)
void saveToFile (DOMString path, SaveFormat format)
void load (DOMString path, SaveFormat format)
void dispose ()

1. Type Definitions

1.1. OptimizerType

Specifies type of Optimizer. 
  enum OptimizerType { "OPTIMIZER_ADAM", "OPTIMIZER_SGD", "OPTIMIZER_UNKNOWN" };

Since: 7.0

  • OPTIMIZER_ADAM - Adam Optimizer
  • OPTIMIZER_SGD - Stochastic Gradient Descent Optimizer
  • OPTIMIZER_UNKNOWN - Unknown Optimizer

1.2. DatasetMode

Specifies mode of a Dataset selected for property setting. 
  enum DatasetMode { "MODE_TRAIN", "MODE_VALID", "MODE_TEST" };

Since: 7.0

  • MODE_TRAIN - Refers to data used as a training part of Dataset
  • MODE_VALID - Refers to data used as a validation part of Dataset
  • MODE_TEST - Refers to data used as a test part of Dataset

1.3. LayerType

Specifies type of Layer. 
  enum LayerType { "LAYER_INPUT", "LAYER_FC", "LAYER_BN", "LAYER_CONV2D", "LAYER_POOLING2D", "LAYER_FLATTEN", "LAYER_ACTIVATION",
    "LAYER_ADDITION", "LAYER_CONCAT", "LAYER_MULTIOUT", "LAYER_EMBEDDING", "LAYER_RNN", "LAYER_LSTM", "LAYER_SPLIT", "LAYER_GRU",
    "LAYER_PERMUTE", "LAYER_DROPOUT", "LAYER_BACKBONE_NNSTREAMER", "LAYER_CENTROID_KNN", "LAYER_PREPROCESS_FLIP",
    "LAYER_PREPROCESS_TRANSLATE", "LAYER_PREPROCESS_L2NORM", "LAYER_LOSS_MSE", "LAYER_LOSS_CROSS_ENTROPY_SIGMOID",
    "LAYER_LOSS_CROSS_ENTROPY_SOFTMAX", "LAYER_UNKNOWN" };

Since: 7.0

  • LAYER_INPUT - Input Layer type
  • LAYER_FC - Fully Connected Layer type
  • LAYER_BN - Batch Normalization Layer type
  • LAYER_CONV2D - Convolution 2D Layer type
  • LAYER_POOLING2D - Pooling 2D Layer type
  • LAYER_FLATTEN - Flatten Layer type
  • LAYER_ACTIVATION - Activation Layer type
  • LAYER_ADDITION - Addition Layer type
  • LAYER_CONCAT - Concat Layer type
  • LAYER_MULTIOUT - Multi Output Layer type
  • LAYER_EMBEDDING - Embedding Layer type
  • LAYER_RNN - RNN Layer type
  • LAYER_LSTM - LSTM Layer type
  • LAYER_SPLIT - Split Layer type
  • LAYER_GRU - GRU Layer type
  • LAYER_PERMUTE - Permute Layer type
  • LAYER_DROPOUT - Dropout Layer type
  • LAYER_BACKBONE_NNSTREAMER - Backbone using NNStreamer
  • LAYER_CENTROID_KNN - Centroid KNN Layer type
  • LAYER_PREPROCESS_FLIP - Preprocess flip Layer type
  • LAYER_PREPROCESS_TRANSLATE - Preprocess translate Layer type
  • LAYER_PREPROCESS_L2NORM - Preprocess L2Normalization Layer type
  • LAYER_LOSS_MSE - Mean Squared Error Loss Layer type
  • LAYER_LOSS_CROSS_ENTROPY_SIGMOID - Cross Entropy with Sigmoid Loss Layer type
  • LAYER_LOSS_CROSS_ENTROPY_SOFTMAX - Cross Entropy with Softmax Loss Layer type
  • LAYER_UNKNOWN - Unknown

1.4. VerbosityLevel

Specifies the level of verbosity of the model summary. 
  enum VerbosityLevel { "SUMMARY_MODEL", "SUMMARY_LAYER", "SUMMARY_TENSOR" };

Since: 7.0

  • SUMMARY_MODEL - Overview of model summary with one-line layer information
  • SUMMARY_LAYER - Detailed model summary with layer properties
  • SUMMARY_TENSOR - Model summary layer's including weight information

1.5. SaveFormat

Specifies the output file format of model saving operation 
  enum SaveFormat { "FORMAT_BIN", "FORMAT_INI", "FORMAT_INI_WITH_BIN" };

Since: 7.0

  • FORMAT_BIN - Model saved in binary file
  • FORMAT_INI - Model configuration saved in INI file
  • FORMAT_INI_WITH_BIN - Model saved in two files, INI file contains configuration and reference to binary file

2. Interfaces

2.1. MachineLearningTrainer

The MachineLearningTrainer interface provides access to the Trainer API. 
  [NoInterfaceObject] interface MachineLearningTrainer {
    Layer createLayer(LayerType type) raises(WebAPIException);
    Dataset createFileDataset(Path train, optional Path? validate, optional Path test) raises(WebAPIException);
    Optimizer createOptimizer(OptimizerType type) raises(WebAPIException);
    Model createModel(optional Path configPath) raises(WebAPIException);
  };

Since: 7.0

Methods

createLayer
Creates a neural network layer. 
Layer createLayer(LayerType type);

Since: 7.0

Parameters:

  • type: Type of the layer.

Exceptions:

  • WebAPIException

    • with error type NotSupportedError, if the feature is not supported.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type AbortError, if any other error occurs.

Code example:

var l1 = tizen.ml.trainer.createLayer("LAYER_INPUT");
console.log("Type of layer: " + l1.type);

Output example:

Type of layer: LAYER_INPUT
createFileDataset
Creates a dataset with files to feed to a neural network. 
Dataset createFileDataset(Path train, optional Path? validate, optional Path test);

Since: 7.0

Each file provided to the function should match format described in machine learning guide

Parameters:

  • train: Path for file with data for training. Check Filesystem API for more information about valid path in Tizen.
  • validate [optional] [nullable]: Path for file with data for validating. Check Filesystem API for more information about valid path in Tizen.
  • test [optional]: Path for file with data for testing. Check Filesystem API for more information about valid path in Tizen.

Exceptions:

  • WebAPIException

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type SecurityError, if the application does not have the privilege to access the storage. For more information, see Storage privileges.

    • with error type AbortError, if any other error occurs.

Code example:

var trainsetFile = "documents/trainingSet.dat";
var validsetFile = "documents/valSet.dat";
var dataset = tizen.ml.trainer.createFileDataset(trainsetFile, validsetFile);
console.log("Type of dataset: " + dataset.type);

Output example:

Type of dataset: DATASET_FILE
createOptimizer
Creates a neural network optimizer. 
Optimizer createOptimizer(OptimizerType type);

Since: 7.0

Parameters:

  • type: Type of the optimizer.

Exceptions:

  • WebAPIException

    • with error type NotSupportedError, if the feature is not supported.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type AbortError, if any other error occurs.

Code example:

var opt = tizen.ml.trainer.createOptimizer("OPTIMIZER_ADAM");
console.log("Type of optimizer: " + opt.type);

Output example:

Type of optimizer: OPTIMIZER_ADAM
createModel
Creates model basing on config file. If optional configPath is not provided, empty model is created. 
Model createModel(optional Path configPath);

Since: 7.0

See the Trainer Guide to learn about configuration file structure.

Even if the model is created from a file, switching, modifying, or setting a component is possible until you compile model with Model::compile.

Parameters:

  • configPath [optional]: Path to config file. Only INI formatted files *.ini are supported to create a model from a file. Check Filesystem API for more information about valid path in Tizen.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if file does not exist.

    • with error type SecurityError, if the application does not have the privilege to access the storage. For more information, see Storage privileges.

    • with error type AbortError, if any other error occurs.

Code example:

var configFile = "documents/config.ini";
var m = tizen.ml.trainer.createModel(configFile);
console.log("Model with config:
" + m.summarize());

var m2 = tizen.ml.trainer.createModel();
console.log("Model without config:
" + m2.summarize());

Output example:

Model with config:
===================<N9nntrainer13NeuralNetworkE at 0xb8095df0>
graph contains 3 operation nodes
Model without config:
===================<N9nntrainer13NeuralNetworkE at 0xb8093f38>
graph contains 0 operation nodes
model is empty!

2.2. Layer

The Layer interface. 
  [NoInterfaceObject] interface Layer {
    readonly attribute DOMString name;
    readonly attribute LayerType type;
    void setProperty(DOMString name, any value) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };

Since: 7.0

Attributes

  • readonly DOMString name
    The name of the layer.

    Name is unique to this layer in a model and might be changed once layer is added to the model to keep the name unique. name value can be changed using Layer::setProperty() method.

    Since: 7.0

  • readonly LayerType type
    The type of the layer.

    Since: 7.0

Methods

setProperty
Sets the value of layer's property. 
void setProperty(DOMString name, any value);

Since: 7.0

Properties available for every LayerType:

Property name Property type Description
name (string) An identifier for each layer
input_shape (string) Formatted string as dim1:dim2:dim3:dim4.
First layer of the model must have input_shape. Other can be omitted as it is calculated at compile phase.
activation (categorical) Activation type
tanh hyperbolic tangent
sigmoid sigmoid function
relu relu function
softmax softmax function
weight_initializer (categorical) Weight initializer
zeros Zero initialization
lecun_normal LeCun Normal Initialization
lecun_uniform LeCun Uniform Initialization
xavier_normal Xavier Normal Initialization
xavier_uniform Xavier Uniform Initialization
he_normal He Normal Initialization
he_uniform He Uniform Initialization
bias_initializer (categorical) Bias initializer, same category as weight_initializer
weight_regularizer (categorical) Weight regularizer
l2norm L2 weight regularizer
weight_regularizer_constant (double) Weight regularizer constant
flatten (boolean) Flatten shape from c:h:w to 1:1:c*h*w

Properties available only for LAYER_INPUT LayerType:

Property name Property type Description
normalization (boolean) Normalize input if true
standardization (boolean) Standardize input if true

Properties available only for LAYER_FC LayerType:

Property name Property type Description
unit (long) number of outputs

Parameters:

  • name: Name of the property.
  • value: The new value of the property.

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the layer does not exist anymore.

Code example:

var l1 = tizen.ml.trainer.createLayer("LAYER_INPUT");
l1.setProperty("input_shape", "1:1:62720");
console.log("Layer type: " + l1.type);

Output example:

Layer type: LAYER_INPUT
dispose
Destroys layer object and releases the resources associated by the layer object. 
void dispose();

Since: 7.0

Layer will not be destroyed and function will throw exception if Layer is added to model. Such layer will be destroyed on Model disposal.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if the layer does not exist anymore.

    • with error type NoModificationAllowedError, if the layer is attached to a model.

Code example:

var m = tizen.ml.trainer.createModel();
var l1 = tizen.ml.trainer.createLayer("LAYER_INPUT");
l1.dispose();
try
{
  m.addLayer(l1);
}
catch (e)
{
  console.log("Adding disposed layer will fail.");
}

Output example:

Adding disposed layer will fail.

2.3. Optimizer

The Optimizer interface. 
  [NoInterfaceObject] interface Optimizer {
    readonly attribute OptimizerType type;
    void setProperty(DOMString name, any value) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };

Since: 7.0

Attributes

  • readonly OptimizerType type
    The type of the optimizer.

    Since: 7.0

Methods

setProperty
Sets the value of optimizer's property. 
void setProperty(DOMString name, any value);

Since: 7.0

Properties available for all optimizers:

Property name Property type
learning_rate float

Properties available for SGD and Adam optimizers:

Property name Property type
decay_rate float
decay_steps float

Optimizers use exponential decay type.

Properties available for Adam optimizers:

Property name Property type
beta1 float
beta2 float
epsilon float

Parameters:

  • name: Name of the property.
  • value: The new value of the property.

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the optimizer does not exist anymore.

Code example:

var opt = tizen.ml.trainer.createOptimizer("OPTIMIZER_ADAM");
opt.setProperty("learning_rate", "0.0001");
console.log("Optimizer type: " + opt.type);

Output example:

Optimizer type: OPTIMIZER_ADAM
dispose
Destroys optimizer object and releases the resources associated by the optimizer object. 
void dispose();

Since: 7.0

Optimizer will not be destroyed and function will throw exception if optimizer is added to model. Such optimizer will be destroyed on Model disposal.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if the optimizer does not exist anymore.

    • with error type NoModificationAllowedError, if the optimizer is attached to a model.

Code example:

var m = tizen.ml.trainer.createModel();
var opt = tizen.ml.trainer.createOptimizer("OPTIMIZER_ADAM");
opt.dispose();
try
{
  m.setOptimizer(opt);
}
catch (e)
{
  console.log("Setting disposed optimizer will fail.");
}

Output example:

Setting disposed optimizer will fail.

2.4. Dataset

The Dataset interface. 
  [NoInterfaceObject] interface Dataset {
    readonly attribute  type;
    void setProperty(DOMString name, any value, DatasetMode mode) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };

Since: 7.0

Attributes

  • readonly DatasetType type
    The type of the dataset.

    Since: 7.0

Methods

setProperty
Sets the value of dataset's property. 
void setProperty(DOMString name, any value, DatasetMode mode);

Since: 7.0

Properties available for all types of dataset:

Property name Description
buffer_size Buffer size (unsigned long)

Properties available only for file type dataset:

Property name Description
train_data Path to training data
val_data Path to validation data
test_data Path to testing data
label_data Path to label data

Parameters:

  • name: Name of the property.
  • value: The new value of the property.
  • mode: The mode of dataset content for which property should be set

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the dataset does not exist anymore.

Code example:

var trainsetFile = "documents/trainingSet.dat";
var validsetFile = "documents/valSet.dat";
var dataset = tizen.ml.trainer.createFileDataset(trainsetFile, validsetFile);
dataset.setProperty("buffer_size", "100", "MODE_TRAIN");
console.log("Type of dataset: " + dataset.type);

Output example:

Type of dataset: DATASET_FILE
dispose
Destroy dataset object and release associated resources 
void dispose();

Since: 7.0

Dataset will not be destroyed and function will throw exception if dataset is added to model. Such dataset will be destroyed on Model disposal.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if the dataset does not exist anymore.

    • with error type NoModificationAllowedError, if the dataset is attached to a model.

Code example:

var m = tizen.ml.trainer.createModel();
var trainsetFile = "documents/trainingSet.dat";
var validsetFile = "documents/valSet.dat";
var dataset = tizen.ml.trainer.createFileDataset(trainsetFile, validsetFile);
dataset.dispose();
try
{
  m.setDataset(dataset);
}
catch (e)
{
  console.log("Setting disposed dataset will fail.");
}

Output example:

Setting disposed dataset will fail.

2.5. CompileOptions

The CompileOptions dictionary represents options for compiling the model. 
  dictionary CompileOptions {
    DOMString loss;
    unsigned long batch_size;
  };

Since: 7.0

Dictionary members

DOMString loss
Loss function type. Available loss functions:
Value Description
"mse" Mean Squared Error
"cross" Cross Entropy (sigmoid/softmax depending of activation function of last layer)

Since: 7.0

unsigned long batch_size
Batch size.

Since: 7.0

Remark: This property can be redefined in run() stage using RunOptions.batch_size.

2.6. RunOptions

The RunOptions dictionary represents options for running model training. 
  dictionary RunOptions {
    unsigned long epochs;
    DOMString save_path;
    unsigned long batch_size;
  };

Since: 7.0

Dictionary members

unsigned long epochs
Maximum epochs.

Since: 7.0

DOMString save_path
Path to save trained model.

Since: 7.0

unsigned long batch_size
Batch size.

Since: 7.0

Remark: This property can be provided on both compile() and run() stages.

2.7. Model

The Model interface. 
  [NoInterfaceObject] interface Model {
    void compile(optional CompileOptions? options) raises(WebAPIException);
    void run(RunOptions? options, SuccessCallback successCallback, optional ErrorCallback? errorCallback) raises(WebAPIException);
    DOMString summarize(optional VerbosityLevel? level) raises(WebAPIException);
    void addLayer(Layer layer) raises(WebAPIException);
    void setDataset(Dataset dataset) raises(WebAPIException);
    void setOptimizer(Optimizer optimizer) raises(WebAPIException);
    void saveToFile(DOMString path, SaveFormat format) raises(WebAPIException);
    void load(DOMString path, SaveFormat format) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };

Since: 7.0

Methods

compile
Compiles the model. Once compiled, any modification to the properties of model or layers/dataset/optimizer in the model will be restricted. Further addition of layers or changing the optimizer/dataset of the model will not be permitted and will cause NoModificationAllowedError. 
void compile(optional CompileOptions? options);

Since: 7.0

Parameters:

  • options [optional] [nullable]: Hyperparameters for model training.

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if model does not exist anymore.

    • with error type NoModificationAllowedError, if the model is already compiled.

    • with error type AbortError, if any internal error occurs.

Code example:

/* assume that m is a valid model with all options set, not yet compiled */
var compileOpts = {loss: "cross", batch_size: "16"};
m.compile(compileOpts);
console.log("Compilation succeeded");

Output example:

Compilation succeeded
run
Trains the neural network model. 
void run(RunOptions? options, SuccessCallback successCallback, optional ErrorCallback? errorCallback);

Since: 7.0

The ErrorCallback method is called with an argument of type WebAPIException with an error:

  • InvalidValuesError - if options contain an invalid value.
  • NotFoundError - if the model does not exist anymore.
  • InvalidStateError - if the model is not compiled.
  • AbortError - if any other error occurs.

Parameters:

  • options [nullable]: Hyperparameters for model training.
  • successCallback: Callback function to invoke when the training along with requested validation and testing completes successfully.
  • errorCallback [optional] [nullable]: Callback function to invoke when an error occurs.

Exceptions:

  • WebAPIException

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

Code example:

/* assume that m is a valid, compiled model with all options set */
var runOpts = {epochs: 2, save_path: outputFile};
m.run(runOpts,
    function()
    {
      console.log("Training successful");
    },
    function(e)
    {
      console.log("error " + JSON.stringify(e));
    });

Output example:

Training successful
summarize
Gets summary of the model. 
DOMString summarize(optional VerbosityLevel? level);

Since: 7.0

Parameters:

  • level [optional] [nullable]: Verbosity level of the summary.
    • Default value: SUMMARY_MODEL.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if the model does not exist anymore.

    • with error type AbortError, if any internal error occurs.

Code example:

/* assume that m is a valid model with all options set */
console.log(m.summarize());

Output example:

===================<N9nntrainer13NeuralNetworkE at 0xb830b5d8>
graph contains 3 operation nodes
addLayer
Adds layer to neural network model. 
void addLayer(Layer layer);

Since: 7.0

The layer is added to the end of the existing layers in the model.

Parameters:

  • layer: Layer to add.

Exceptions:

  • WebAPIException

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the model or the layer does not exist anymore.

    • with error type NoModificationAllowedError, if the model is already compiled or model compilation or training is in progress.

    • with error type AbortError, if any internal error occurs.

Code example:

var m = tizen.ml.trainer.createModel();
var l1 = tizen.ml.trainer.createLayer("LAYER_INPUT");
m.addLayer(l1);
console.log("Added layer: " + l1.type);

Output example:

Added layer: LAYER_INPUT
setDataset
Sets the dataset (data provider) for the neural network model. 
void setDataset(Dataset dataset);

Since: 7.0

The dataset will provide training, validation and test data for the model.

Parameters:

  • dataset: Dataset to set.

Exceptions:

  • WebAPIException

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the model or the dataset does not exist anymore.

    • with error type NoModificationAllowedError, if model compilation or training is in progress.

    • with error type AbortError, if any internal error occurs.

Code example:

var trainsetFile = "documents/trainingSet.dat";
var validsetFile = "documents/valSet.dat";
var m = tizen.ml.trainer.createModel();
var dataset = tizen.ml.trainer.createFileDataset(trainsetFile, validsetFile);
m.setDataset(dataset);
console.log("Set dataset: " + dataset.type);

Output example:

Set dataset: DATASET_FILE
setOptimizer
Sets the optimizer for the neural network model. 
void setOptimizer(Optimizer optimizer);

Since: 7.0

Parameters:

  • optimizer: Optimizer to set.

Exceptions:

  • WebAPIException

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type NotFoundError, if the model or the optimizer does not exist anymore.

    • with error type NoModificationAllowedError, if the model is already compiled or model compilation or training is in progress.

    • with error type AbortError, if any internal error occurs.

Code example:

var m = tizen.ml.trainer.createModel();
var opt = tizen.ml.trainer.createOptimizer("OPTIMIZER_ADAM");
m.setOptimizer(opt);
console.log("Set optimizer: " + opt.type);

Output example:

Set optimizer: OPTIMIZER_ADAM
saveToFile
Saves model to file in selected format for future reuse 
void saveToFile(DOMString path, SaveFormat format);

Since: 7.0

Remark: Using FORMAT_INI_WITH_BIN creates two files: *.ini (containing parameters of a model) and *.bin (containing data of a learnt model)

Parameters:

  • path: Path to output file. If file already exists exception will be thrown
  • format: Selected output format

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type NoModificationAllowedError, if a destination file already exists or model compilation or training is in progress.

    • with error type NotFoundError, if the model does not exist anymore.

    • with error type SecurityError, if the application does not have the privilege to access the storage. For more information, see Storage privileges.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type AbortError, if any internal error occurs.

Code example:

var m = tizen.ml.trainer.createModel();
/* set all properties, add layers, set optimizer and dataset */
/* compile prepared model before saving */
m.saveToFile("documents/train_model.ini", "FORMAT_INI");
console.log("Successfully saved to file");

Output example:

Successfully saved to file
load
Loads model from a file in selected format. 
void load(DOMString path, SaveFormat format);

Since: 7.0

Remark: Using FORMAT_INI_WITH_BIN requires two files: *.ini (containing parameters of a model) and *.bin (containing data of a learnt model). Loading of binary data is done during compilation of a model.

Parameters:

  • path: Path to file storing a model. If file does not exist exception will be thrown
  • format: Selected file format

Exceptions:

  • WebAPIException

    • with error type InvalidValuesError, if any of the input parameters contain an invalid value.

    • with error type NotFoundError, if file does not exist or the model does not exist anymore.

    • with error type NoModificationAllowedError, if model compilation or training is in progress.

    • with error type SecurityError, if the application does not have the privilege to access the storage. For more information, see Storage privileges.

    • with error type TypeMismatchError, if any input parameter is not compatible with the expected type for that parameter.

    • with error type AbortError, if any internal error occurs.

Code example:

var m = tizen.ml.trainer.createModel();
m.load("documents/train_model.ini", "FORMAT_INI_WITH_BIN");
console.log("Successfully loaded a model:");
console.log(m.summarize());

Output example:

Successfully loaded a model:
===================<N9nntrainer13NeuralNetworkE at 0xb8095df0>
graph contains 3 operation nodes
dispose
Destroys model object and releases the resources associated by the model object. 
void dispose();

Since: 7.0

When model is disposed all attached objects (layers, dataset and optimizer) are also destroyed.

Exceptions:

  • WebAPIException

    • with error type NotFoundError, if the model does not exist anymore.

    • with error type NoModificationAllowedError, if model compilation or training is in progress.

Code example:

var m = tizen.ml.trainer.createModel();
var l1 = tizen.ml.trainer.createLayer("LAYER_INPUT");
m.dispose();
try
{
  m.addLayer(l1);
}
catch (e)
{
  console.log("Adding layer to disposed model will fail.");
}

Output example:

Adding layer to disposed model will fail.

3. Related Feature

Method tizen.systeminfo.getCapability() can be used in application runtime to check whether this API is supported.

To guarantee that the ML application runs on a device with the ML feature, declare the following feature requirements in the config file:

  • http://tizen.org/feature/machine_learning

    • To guarantee that the ML application runs on a device with the ML Trainer feature, declare the following feature requirements in the config file:

  • http://tizen.org/feature/machine_learning.training

    • For more information, see Application Filtering.

    4. Full WebIDL

    module Trainer {
  enum OptimizerType { "OPTIMIZER_ADAM", "OPTIMIZER_SGD", "OPTIMIZER_UNKNOWN" };
  enum DatasetMode { "MODE_TRAIN", "MODE_VALID", "MODE_TEST" };
  enum LayerType { "LAYER_INPUT", "LAYER_FC", "LAYER_BN", "LAYER_CONV2D", "LAYER_POOLING2D", "LAYER_FLATTEN", "LAYER_ACTIVATION",
    "LAYER_ADDITION", "LAYER_CONCAT", "LAYER_MULTIOUT", "LAYER_EMBEDDING", "LAYER_RNN", "LAYER_LSTM", "LAYER_SPLIT", "LAYER_GRU",
    "LAYER_PERMUTE", "LAYER_DROPOUT", "LAYER_BACKBONE_NNSTREAMER", "LAYER_CENTROID_KNN", "LAYER_PREPROCESS_FLIP",
    "LAYER_PREPROCESS_TRANSLATE", "LAYER_PREPROCESS_L2NORM", "LAYER_LOSS_MSE", "LAYER_LOSS_CROSS_ENTROPY_SIGMOID",
    "LAYER_LOSS_CROSS_ENTROPY_SOFTMAX", "LAYER_UNKNOWN" };
  enum VerbosityLevel { "SUMMARY_MODEL", "SUMMARY_LAYER", "SUMMARY_TENSOR" };
  enum SaveFormat { "FORMAT_BIN", "FORMAT_INI", "FORMAT_INI_WITH_BIN" };
  dictionary CompileOptions {
    DOMString loss;
    unsigned long batch_size;
  };
  dictionary RunOptions {
    unsigned long epochs;
    DOMString save_path;
    unsigned long batch_size;
  };
  [NoInterfaceObject] interface MachineLearningTrainer {
    Layer createLayer(LayerType type) raises(WebAPIException);
    Dataset createFileDataset(Path train, optional Path? validate, optional Path test) raises(WebAPIException);
    Optimizer createOptimizer(OptimizerType type) raises(WebAPIException);
    Model createModel(optional Path configPath) raises(WebAPIException);
  };
  [NoInterfaceObject] interface Layer {
    readonly attribute DOMString name;
    readonly attribute LayerType type;
    void setProperty(DOMString name, any value) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };
  [NoInterfaceObject] interface Optimizer {
    readonly attribute OptimizerType type;
    void setProperty(DOMString name, any value) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };
  [NoInterfaceObject] interface Dataset {
    readonly attribute  type;
    void setProperty(DOMString name, any value, DatasetMode mode) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };
  [NoInterfaceObject] interface Model {
    void compile(optional CompileOptions? options) raises(WebAPIException);
    void run(RunOptions? options, SuccessCallback successCallback, optional ErrorCallback? errorCallback) raises(WebAPIException);
    DOMString summarize(optional VerbosityLevel? level) raises(WebAPIException);
    void addLayer(Layer layer) raises(WebAPIException);
    void setDataset(Dataset dataset) raises(WebAPIException);
    void setOptimizer(Optimizer optimizer) raises(WebAPIException);
    void saveToFile(DOMString path, SaveFormat format) raises(WebAPIException);
    void load(DOMString path, SaveFormat format) raises(WebAPIException);
    void dispose() raises(WebAPIException);
  };
};
    Except as noted, this content - excluding the Code Examples - is licensed under Creative Commons Attribution 3.0 and all of the Code Examples contained herein are licensed under BSD-3-Clause.
    For details, see the Content License.