The Amazon SageMaker NTM algorithm.
NTM(role, train_instance_count, train_instance_type, num_topics, encoder_layers=None, epochs=None, encoder_layers_activation=None, optimizer=None, tolerance=None, num_patience_epochs=None, batch_norm=None, rescale_gradient=None, clip_gradient=None, weight_decay=None, learning_rate=None, **kwargs)¶
Neural Topic Model (NTM) is
Estimatorused for unsupervised learning.
This Estimator may be fit via calls to
fit(). It requires Amazon
Recordprotobuf serialized data to be stored in S3. There is an utility
record_set()that can be used to upload data to S3 and creates
RecordSetto be passed to the fit call.
To learn more about the Amazon protobuf Record class and how to prepare bulk data in this format, please consult AWS technical documentation: https://docs.aws.amazon.com/sagemaker/latest/dg/cdf-training.html
After this Estimator is fit, model data is stored in S3. The model may be deployed to an Amazon SageMaker Endpoint by invoking
deploy(). As well as deploying an Endpoint, deploy returns a
NTMPredictorobject that can be used for inference calls using the trained model hosted in the SageMaker Endpoint.
NTM Estimators can be configured by setting hyperparameters. The available hyperparameters for NTM are documented below.
For further information on the AWS NTM algorithm, please consult AWS technical documentation: https://docs.aws.amazon.com/sagemaker/latest/dg/ntm.html
- role (str) – An AWS IAM role (either name or full ARN). The Amazon SageMaker training jobs and APIs that create Amazon SageMaker endpoints use this role to access training data and model artifacts. After the endpoint is created, the inference code might use the IAM role, if accessing AWS resource.
- train_instance_type (str) – Type of EC2 instance to use for training, for example, ‘ml.c4.xlarge’.
- num_topics (int) – Required. The number of topics for NTM to find within the data.
- encoder_layers (list) – Optional. Represents number of layers in the encoder and the output size of each layer.
- epochs (int) – Optional. Maximum number of passes over the training data.
- encoder_layers_activation (str) – Optional. Activation function to use in the encoder layers.
- optimizer (str) – Optional. Optimizer to use for training.
- tolerance (float) – Optional. Maximum relative change in the loss function within the last num_patience_epochs number of epochs below which early stopping is triggered.
- num_patience_epochs (int) – Optional. Number of successive epochs over which early stopping criterion is evaluated.
- batch_norm (bool) – Optional. Whether to use batch normalization during training.
- rescale_gradient (float) – Optional. Rescale factor for gradient.
- clip_gradient (float) – Optional. Maximum magnitude for each gradient component.
- weight_decay (float) – Optional. Weight decay coefficient. Adds L2 regularization.
- learning_rate (float) – Optional. Learning rate for the optimizer.
- **kwargs – base class keyword argument values.
Attach to an existing training job.
Create an Estimator bound to an existing training job, each subclass is responsible to implement
_prepare_init_params_from_job_description()as this method delegates the actual conversion of a training job description to the arguments that the class constructor expects. After attaching, if the training job has a Complete status, it can be
deploy()ed to create a SageMaker Endpoint and return a
If the training job is in progress, attach will block and display log messages from the training job, until the training job completes.
- training_job_name (str) – The name of the training job to attach to.
- sagemaker_session (sagemaker.session.Session) – Session object which manages interactions with Amazon SageMaker APIs and any other AWS services needed. If not specified, the estimator creates one using the default AWS configuration chain.
>>> my_estimator.fit(wait=False) >>> training_job_name = my_estimator.latest_training_job.name Later on: >>> attached_estimator = Estimator.attach(training_job_name) >>> attached_estimator.deploy()
Returns: Instance of the calling
EstimatorClass with the attached training job.
Delete an Amazon SageMaker
ValueError– If the endpoint does not exist.
deploy(initial_instance_count, instance_type, endpoint_name=None, **kwargs)¶
Deploy the trained model to an Amazon SageMaker endpoint and return a
More information: http://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works-training.html
- initial_instance_count (int) – Minimum number of EC2 instances to deploy to an endpoint for prediction.
- instance_type (str) – Type of EC2 instance to deploy to an endpoint for prediction, for example, ‘ml.c4.xlarge’.
- endpoint_name (str) – Name to use for creating an Amazon SageMaker endpoint. If not specified, the name of the training job is used.
- **kwargs – Passed to invocation of
create_model(). Implementations may customize
**kwargsto customize model creation during deploy. For more, see the implementation docs.
- A predictor that provides a
which can be used to send requests to the Amazon SageMaker endpoint and obtain inferences.
fit(records, mini_batch_size=None, wait=True, logs=True, job_name=None)¶
Fit this Estimator on serialized Record objects, stored in S3.
recordsshould be an instance of
RecordSet. This defines a collection of S3 data files to train this
Training data is expected to be encoded as dense or sparse vectors in the “values” feature on each Record. If the data is labeled, the label is expected to be encoded as a list of scalas in the “values” feature of the Record label.
More information on the Amazon Record format is available at: https://docs.aws.amazon.com/sagemaker/latest/dg/cdf-training.html
record_set()to construct a
- records (
RecordSet) – The records to train this
- mini_batch_size (int or None) – The size of each mini-batch to use when training. If
None, a default value will be used.
- wait (bool) – Whether the call should wait until the job completes (default: True).
- logs (bool) – Whether to show the logs produced by the job. Only meaningful when wait is True (default: True).
- job_name (str) – Training job name. If not specified, the estimator generates a default job name, based on the training image name and current timestamp.
- records (
Return the hyperparameters as a dictionary to use for training.
fit()method, which trains the model, calls this method to find the hyperparameters.
Returns: The hyperparameters. Return type: dict[str, str]
str – The model location in S3. Only set if Estimator has been
record_set(train, labels=None, channel='train')¶
RecordSetfrom a numpy
ndarraymatrix and label vector.
For the 2D
train, each row is converted to a
Recordobject. The vector is stored in the “values” entry of the
featuresproperty of each Record. If
labelsis not None, each corresponding label is assigned to the “values” entry of the
labelsproperty of each Record.
The collection of
Recordobjects are protobuf serialized and uploaded to new S3 locations. A manifest file is generated containing the list of objects created and also stored in S3.
The number of S3 objects created is controlled by the
train_instance_countproperty on this Estimator. One S3 object is created per training instance.
- train (numpy.ndarray) – A 2D numpy array of training data.
- labels (numpy.ndarray) – A 1D numpy array of labels. Its length must be equal to the
number of rows in
- channel (str) – The SageMaker TrainingJob channel this RecordSet should be assigned to.
A RecordSet referencing the encoded, uploading training and label data.
Return the Docker image to use for training.
fit()method, which does the model training, calls this method to find the image to use for model training.
Returns: The URI of the Docker image. Return type: str
TrainingJobAnalyticsobject for the current training job.
NTMModelreferencing the latest s3 model data produced by this Estimator.
NTMModel(model_data, role, sagemaker_session=None)¶
Reference NTM s3 model data. Calling
deploy()creates an Endpoint and return a Predictor that transforms vectors to a lower-dimensional representation.
Transforms input vectors to lower-dimesional representations.
The implementation of
predict()in this RealTimePredictor requires a numpy
ndarrayas input. The array should contain the same number of columns as the feature-dimension of the data used to fit the model this Predictor performs inference on.
predict()returns a list of
Recordobjects, one for each row in the input
ndarray. The lower dimension vector result is stored in the
projectionkey of the