internal static ModelProto ConvertToOnnxProtobuf(this ModelOperationsCatalog catalog, ITransformer transform, IDataView inputData, int opSetVersion)
{
var env = catalog.GetEnvironment();
var ctx = new OnnxContextImpl(env, "model", "ML.NET", "0", 0, "machinelearning.dotnet", OnnxVersion.Stable, opSetVersion);
return(ConvertToOnnxProtobufCore(env, ctx, transform, inputData));
}
/// <summary>
/// Performs image classification using transfer learning.
/// usage of this API requires additional NuGet dependencies on TensorFlow redist, see linked document for more information.
/// <format type="text/markdown">
/// <]
/// ]]>
/// </format>
/// </summary>
/// <param name="catalog"></param>
/// <param name="options">An <see cref="Options"/> object specifying advanced options for <see cref="ImageClassificationEstimator"/>.</param>
public static ImageClassificationEstimator ImageClassification(
this ModelOperationsCatalog catalog, Options options)
{
options.EarlyStoppingCriteria = options.DisableEarlyStopping ? null : options.EarlyStoppingCriteria ?? new EarlyStopping();
var env = CatalogUtils.GetEnvironment(catalog);
return(new ImageClassificationEstimator(env, options, DnnUtils.LoadDnnModel(env, options.Arch, true)));
}
internal static ModelProto ConvertToOnnxProtobuf(this ModelOperationsCatalog catalog, ITransformer transform, IDataView inputData)
{
var env = catalog.Environment;
var ctx = new OnnxContextImpl(env, "model", "ML.NET", "0", 0, "machinelearning.dotnet", OnnxVersion.Stable);
var outputData = transform.Transform(inputData);
LinkedList <ITransformCanSaveOnnx> transforms = null;
using (var ch = env.Start("ONNX conversion"))
{
SaveOnnxCommand.GetPipe(ctx, ch, outputData, out IDataView root, out IDataView sink, out transforms);
return(SaveOnnxCommand.ConvertTransformListToOnnxModel(ctx, ch, root, sink, transforms, null, null));
}
}
/// <summary>
/// Create the ML context.
/// </summary>
/// <param name="seed">Random seed. Set to <c>null</c> for a non-deterministic environment.</param>
/// <param name="conc">Concurrency level. Set to 1 to run single-threaded. Set to 0 to pick automatically.</param>
public MLContext(int?seed = null, int conc = 0)
{
_env = new LocalEnvironment(seed, conc, MakeCompositionContainer);
_env.AddListener(ProcessMessage);
BinaryClassification = new BinaryClassificationContext(_env);
MulticlassClassification = new MulticlassClassificationContext(_env);
Regression = new RegressionContext(_env);
Clustering = new ClusteringContext(_env);
Ranking = new RankingContext(_env);
Transforms = new TransformsCatalog(_env);
Model = new ModelOperationsCatalog(_env);
Data = new DataOperations(_env);
}
/// <summary>
/// Create the ML context.
/// </summary>
/// <param name="seed">Random seed. Set to <c>null</c> for a non-deterministic environment.</param>
public MLContext(int?seed = null)
{
_env = new LocalEnvironment(seed);
_env.AddListener(ProcessMessage);
BinaryClassification = new BinaryClassificationCatalog(_env);
MulticlassClassification = new MulticlassClassificationCatalog(_env);
Regression = new RegressionCatalog(_env);
Clustering = new ClusteringCatalog(_env);
Ranking = new RankingCatalog(_env);
AnomalyDetection = new AnomalyDetectionCatalog(_env);
Transforms = new TransformsCatalog(_env);
Model = new ModelOperationsCatalog(_env);
Data = new DataOperationsCatalog(_env);
}
/// <summary>
/// Performs image classification using transfer learning.
/// usage of this API requires additional NuGet dependencies on TensorFlow redist, see linked document for more information.
/// <format type="text/markdown">
/// <]
/// ]]>
/// </format>
/// </summary>
/// <param name="catalog"></param>
/// <param name="featuresColumnName">The name of the input features column.</param>
/// <param name="labelColumnName">The name of the labels column.</param>
/// <param name="scoreColumnName">The name of the output score column.</param>
/// <param name="predictedLabelColumnName">The name of the output predicted label columns.</param>
/// <param name="validationSet">Validation set.</param>
public static ImageClassificationEstimator ImageClassification(
this ModelOperationsCatalog catalog,
string featuresColumnName,
string labelColumnName,
string scoreColumnName = "Score",
string predictedLabelColumnName = "PredictedLabel",
IDataView validationSet = null
)
{
var options = new ImageClassificationEstimator.Options()
{
FeaturesColumnName = featuresColumnName,
LabelColumnName = labelColumnName,
ScoreColumnName = scoreColumnName,
PredictedLabelColumnName = predictedLabelColumnName,
ValidationSet = validationSet
};
return(ImageClassification(catalog, options));
}
/// <summary>
/// Retrain the dnn model on new data.
/// </summary>
/// <param name="catalog"></param>
/// <param name="inputColumnNames"> The names of the model inputs.</param>
/// <param name="outputColumnNames">The names of the requested model outputs.</param>
/// <param name="labelColumnName">Name of the label column.</param>
/// <param name="dnnLabel">Name of the node in DNN graph that is used as label during training in Dnn.
/// The value of <paramref name="labelColumnName"/> from <see cref="IDataView"/> is fed to this node.</param>
/// <param name="optimizationOperation">The name of the optimization operation in the Dnn graph.</param>
/// <param name="modelPath">Path to model file to retrain.</param>
/// <param name="epoch">Number of training iterations.</param>
/// <param name="batchSize">Number of samples to use for mini-batch training.</param>
/// <param name="lossOperation">The name of the operation in the Dnn graph to compute training loss (Optional).</param>
/// <param name="metricOperation">The name of the operation in the Dnn graph to compute performance metric during training (Optional).</param>
/// <param name="learningRateOperation">The name of the operation in the Dnn graph which sets optimizer learning rate (Optional).</param>
/// <param name="learningRate">Learning rate to use during optimization (Optional).</param>
/// <param name="addBatchDimensionInput">Add a batch dimension to the input e.g. input = [224, 224, 3] => [-1, 224, 224, 3].
/// This parameter is used to deal with models that have unknown shape but the internal operators in the model require data to have batch dimension as well.</param>
/// <param name="dnnFramework"></param>
/// <remarks>
/// The support for retraining is under preview.
/// </remarks>
public static DnnEstimator RetrainDnnModel(
this ModelOperationsCatalog catalog,
string[] outputColumnNames,
string[] inputColumnNames,
string labelColumnName,
string dnnLabel,
string optimizationOperation,
string modelPath,
int epoch = 10,
int batchSize = 20,
string lossOperation = null,
string metricOperation = null,
string learningRateOperation = null,
float learningRate = 0.01f,
bool addBatchDimensionInput = false,
DnnFramework dnnFramework = DnnFramework.Tensorflow)
{
var options = new Options()
{
ModelLocation = modelPath,
InputColumns = inputColumnNames,
OutputColumns = outputColumnNames,
LabelColumn = labelColumnName,
TensorFlowLabel = dnnLabel,
OptimizationOperation = optimizationOperation,
LossOperation = lossOperation,
MetricOperation = metricOperation,
Epoch = epoch,
LearningRateOperation = learningRateOperation,
LearningRate = learningRate,
BatchSize = batchSize,
AddBatchDimensionInputs = addBatchDimensionInput,
ReTrain = true
};
var env = CatalogUtils.GetEnvironment(catalog);
return(new DnnEstimator(env, options, DnnUtils.LoadDnnModel(env, modelPath, true)));
}
/// <summary> /// Convert the specified <see cref="ITransformer"/> to ONNX format and writes to a stream. /// </summary> /// <param name="catalog">The class that <see cref="ConvertToOnnx(ModelOperationsCatalog, ITransformer, IDataView, Stream)"/> attached to.</param> /// <param name="transform">The <see cref="ITransformer"/> that will be converted into ONNX format.</param> /// <param name="inputData">The input of the specified transform.</param> /// <param name="stream">The stream to write the protobuf model to.</param> /// <returns>An ONNX model equivalent to the converted ML.NET model.</returns> public static void ConvertToOnnx(this ModelOperationsCatalog catalog, ITransformer transform, IDataView inputData, Stream stream) => ConvertToOnnxProtobuf(catalog, transform, inputData).WriteTo(stream);
internal ExplainabilityTransforms(ModelOperationsCatalog owner)
{
_env = owner._env;
}
internal ExplainabilityTransforms(ModelOperationsCatalog owner) : base(owner)
{
}
protected SubCatalogBase(ModelOperationsCatalog owner)
{
Environment = owner.Environment;
}
/// <summary> /// Load TensorFlow model into memory. This is the convenience method that allows the model to be loaded once and subsequently use it for querying schema and creation of /// <see cref="TensorFlowEstimator"/> using <see cref="TensorFlowModel.ScoreTensorFlowModel(string, string, bool)"/>. /// </summary> /// <param name="catalog">The transform's catalog.</param> /// <param name="modelLocation">Location of the TensorFlow model.</param> /// <example> /// <format type="text/markdown"> /// <] /// ]]> /// </format> /// </example> public static TensorFlowModel LoadTensorFlowModel(this ModelOperationsCatalog catalog, string modelLocation) => TensorFlowUtils.LoadTensorFlowModel(CatalogUtils.GetEnvironment(catalog), modelLocation);
/// <summary> /// Convert the specified <see cref="ITransformer"/> to ONNX format and writes to a stream. /// </summary> /// <param name="catalog">The class that <see cref="ConvertToOnnx(ModelOperationsCatalog, ITransformer, IDataView, Stream)"/> attached to.</param> /// <param name="transform">The <see cref="ITransformer"/> that will be converted into ONNX format.</param> /// <param name="inputData">The input of the specified transform.</param> /// <param name="stream">The stream to write the protobuf model to.</param> /// <param name="outputColumns">List of output columns we want to keep.</param> /// <returns>An ONNX model equivalent to the converted ML.NET model.</returns> public static void ConvertToOnnx(this ModelOperationsCatalog catalog, ITransformer transform, IDataView inputData, Stream stream, params string[] outputColumns) => ConvertToOnnxProtobuf(catalog, transform, inputData, outputColumns).WriteTo(stream);
/// <summary>
/// Performs image classification using transfer learning.
/// usage of this API requires additional NuGet dependencies on TensorFlow redist, see linked document for more information.
/// <format type="text/markdown">
/// <]
/// ]]>
/// </format>
/// </summary>
/// <param name="catalog"></param>
/// <param name="featuresColumnName">The name of the input features column.</param>
/// <param name="labelColumnName">The name of the labels column.</param>
/// <param name="scoreColumnName">The name of the output score column.</param>
/// <param name="predictedLabelColumnName">The name of the output predicted label columns.</param>
/// <param name="arch">The architecture of the image recognition DNN model.</param>
/// <param name="epoch">Number of training iterations. Each iteration/epoch refers to one pass over the dataset.</param>
/// <param name="batchSize">The batch size for training.</param>
/// <param name="learningRate">The learning rate for training.</param>
/// <param name="disableEarlyStopping">Whether to disable use of early stopping technique. Training will go on for the full epoch count.</param>
/// <param name="earlyStopping">Early stopping technique parameters to be used to terminate training when training metric stops improving.</param>
/// <param name="metricsCallback">Callback for reporting model statistics during training phase.</param>
/// <param name="statisticFrequency">Indicates the frequency of epochs at which to report model statistics during training phase.</param>
/// <param name="framework">Indicates the choice of DNN training framework. Currently only tensorflow is supported.</param>
/// <param name="modelSavePath">Optional name of the path where a copy new graph should be saved. The graph will be saved as part of model.</param>
/// <param name="finalModelPrefix">The name of the prefix for the final mode and checkpoint files.</param>
/// <param name="validationSet">Validation set.</param>
/// <param name="testOnTrainSet">Indicates to evaluate the model on train set after every epoch.</param>
/// <param name="reuseTrainSetBottleneckCachedValues">Indicates to not re-compute cached trainset bottleneck values if already available in the bin folder.</param>
/// <param name="reuseValidationSetBottleneckCachedValues">Indicates to not re-compute validataionset cached bottleneck validationset values if already available in the bin folder.</param>
/// <param name="trainSetBottleneckCachedValuesFilePath">Indicates the file path to store trainset bottleneck values for caching.</param>
/// <param name="validationSetBottleneckCachedValuesFilePath">Indicates the file path to store validationset bottleneck values for caching.</param>
/// <remarks>
/// The support for image classification is under preview.
/// </remarks>
public static ImageClassificationEstimator ImageClassification(
this ModelOperationsCatalog catalog,
string featuresColumnName,
string labelColumnName,
string scoreColumnName = "Score",
string predictedLabelColumnName = "PredictedLabel",
Architecture arch = Architecture.InceptionV3,
int epoch = 100,
int batchSize = 10,
float learningRate = 0.01f,
bool disableEarlyStopping = false,
EarlyStopping earlyStopping = null,
ImageClassificationMetricsCallback metricsCallback = null,
int statisticFrequency = 1,
DnnFramework framework = DnnFramework.Tensorflow,
string modelSavePath = null,
string finalModelPrefix = "custom_retrained_model_based_on_",
IDataView validationSet = null,
bool testOnTrainSet = true,
bool reuseTrainSetBottleneckCachedValues = false,
bool reuseValidationSetBottleneckCachedValues = false,
string trainSetBottleneckCachedValuesFilePath = "trainSetBottleneckFile.csv",
string validationSetBottleneckCachedValuesFilePath = "validationSetBottleneckFile.csv"
)
{
var options = new ImageClassificationEstimator.Options()
{
ModelLocation = ModelLocation[arch],
InputColumns = new[] { featuresColumnName },
OutputColumns = new[] { scoreColumnName, predictedLabelColumnName },
LabelColumn = labelColumnName,
TensorFlowLabel = labelColumnName,
Epoch = epoch,
LearningRate = learningRate,
BatchSize = batchSize,
EarlyStoppingCriteria = disableEarlyStopping ? null : earlyStopping == null ? new EarlyStopping() : earlyStopping,
ScoreColumnName = scoreColumnName,
PredictedLabelColumnName = predictedLabelColumnName,
FinalModelPrefix = finalModelPrefix,
Arch = arch,
MetricsCallback = metricsCallback,
StatisticsFrequency = statisticFrequency,
Framework = framework,
ModelSavePath = modelSavePath,
ValidationSet = validationSet,
TestOnTrainSet = testOnTrainSet,
TrainSetBottleneckCachedValuesFilePath = trainSetBottleneckCachedValuesFilePath,
ValidationSetBottleneckCachedValuesFilePath = validationSetBottleneckCachedValuesFilePath,
ReuseTrainSetBottleneckCachedValues = reuseTrainSetBottleneckCachedValues,
ReuseValidationSetBottleneckCachedValues = reuseValidationSetBottleneckCachedValues
};
if (!File.Exists(options.ModelLocation))
{
if (options.Arch == Architecture.InceptionV3)
{
var baseGitPath = @"https://raw.githubusercontent.com/SciSharp/TensorFlow.NET/master/graph/InceptionV3.meta";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"InceptionV3.meta");
}
baseGitPath = @"https://github.com/SciSharp/TensorFlow.NET/raw/master/data/tfhub_modules.zip";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"tfhub_modules.zip");
ZipFile.ExtractToDirectory(Path.Combine(Directory.GetCurrentDirectory(), @"tfhub_modules.zip"), @"tfhub_modules");
}
}
else if (options.Arch == Architecture.ResnetV2101)
{
var baseGitPath = @"https://aka.ms/mlnet-resources/image/ResNet101Tensorflow/resnet_v2_101_299.meta";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"resnet_v2_101_299.meta");
}
}
else if (options.Arch == Architecture.MobilenetV2)
{
var baseGitPath = @"https://tlcresources.blob.core.windows.net/image/MobileNetV2TensorFlow/mobilenet_v2.meta";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"mobilenet_v2.meta");
}
}
}
var env = CatalogUtils.GetEnvironment(catalog);
return(new ImageClassificationEstimator(env, options, DnnUtils.LoadDnnModel(env, options.ModelLocation, true)));
}
/// <summary>
/// Performs image classification using transfer learning.
/// </summary>
/// <param name="catalog"></param>
/// <param name="featuresColumnName">The name of the input features column.</param>
/// <param name="labelColumnName">The name of the labels column.</param>
/// <param name="outputGraphPath">Optional name of the path where a copy new graph should be saved. The graph will be saved as part of model.</param>
/// <param name="scoreColumnName">The name of the output score column.</param>
/// <param name="predictedLabelColumnName">The name of the output predicted label columns.</param>
/// <param name="checkpointName">The name of the prefix for checkpoint files.</param>
/// <param name="arch">The architecture of the image recognition DNN model.</param>
/// <param name="dnnFramework">The backend DNN framework to use, currently only Tensorflow is supported.</param>
/// <param name="epoch">Number of training epochs.</param>
/// <param name="batchSize">The batch size for training.</param>
/// <param name="learningRate">The learning rate for training.</param>
/// <remarks>
/// The support for image classification is under preview.
/// </remarks>
public static DnnEstimator ImageClassification(
this ModelOperationsCatalog catalog,
string featuresColumnName,
string labelColumnName,
string outputGraphPath = null,
string scoreColumnName = "Score",
string predictedLabelColumnName = "PredictedLabel",
string checkpointName = "_retrain_checkpoint",
Architecture arch = Architecture.ResnetV2101,
DnnFramework dnnFramework = DnnFramework.Tensorflow,
int epoch = 10,
int batchSize = 20,
float learningRate = 0.01f)
{
var options = new Options()
{
ModelLocation = arch == Architecture.ResnetV2101 ? @"resnet_v2_101_299.meta" : @"InceptionV3.meta",
InputColumns = new[] { featuresColumnName },
OutputColumns = new[] { scoreColumnName, predictedLabelColumnName },
LabelColumn = labelColumnName,
TensorFlowLabel = labelColumnName,
Epoch = epoch,
LearningRate = learningRate,
BatchSize = batchSize,
AddBatchDimensionInputs = arch == Architecture.InceptionV3 ? false : true,
TransferLearning = true,
ScoreColumnName = scoreColumnName,
PredictedLabelColumnName = predictedLabelColumnName,
CheckpointName = checkpointName,
Arch = arch,
MeasureTrainAccuracy = false
};
if (!File.Exists(options.ModelLocation))
{
if (options.Arch == Architecture.InceptionV3)
{
var baseGitPath = @"https://raw.githubusercontent.com/SciSharp/TensorFlow.NET/master/graph/InceptionV3.meta";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"InceptionV3.meta");
}
baseGitPath = @"https://github.com/SciSharp/TensorFlow.NET/raw/master/data/tfhub_modules.zip";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"tfhub_modules.zip");
ZipFile.ExtractToDirectory(Path.Combine(Directory.GetCurrentDirectory(), @"tfhub_modules.zip"), @"tfhub_modules");
}
}
else if (options.Arch == Architecture.ResnetV2101)
{
var baseGitPath = @"https://aka.ms/mlnet-resources/image/ResNet101Tensorflow/resnet_v2_101_299.meta";
using (WebClient client = new WebClient())
{
client.DownloadFile(new Uri($"{baseGitPath}"), @"resnet_v2_101_299.meta");
}
}
}
var env = CatalogUtils.GetEnvironment(catalog);
return(new DnnEstimator(env, options, DnnUtils.LoadDnnModel(env, options.ModelLocation, true)));
}
/// <summary> /// Load TensorFlow model into memory. This is the convenience method that allows the model to be loaded once and subsequently use it for querying schema and creation of /// <see cref="TensorFlowEstimator"/> using <see cref="TensorFlowModel.ScoreTensorFlowModel(string, string, bool)"/>. /// usage of this API requires additional NuGet dependencies on TensorFlow redist, see linked document for more information. /// <see cref="TensorFlowModel"/> also holds references to unmanaged resources that need to be freed either with an explicit /// call to Dispose() or implicitly by declaring the variable with the "using" syntax/> /// /// <format type="text/markdown"> /// <] /// ]]> /// </format> /// </summary> /// <param name="catalog">The transform's catalog.</param> /// <param name="modelLocation">Location of the TensorFlow model.</param> /// <param name="treatOutputAsBatched">If the first dimension of the output is unknown, should it be treated as batched or not.</param> /// <example> /// <format type="text/markdown"> /// <] /// ]]> /// </format> /// </example> public static TensorFlowModel LoadTensorFlowModel(this ModelOperationsCatalog catalog, string modelLocation, bool treatOutputAsBatched) => TensorFlowUtils.LoadTensorFlowModel(CatalogUtils.GetEnvironment(catalog), modelLocation, treatOutputAsBatched);
