anomed_challenge.challenge_server

This module provides means to create web applications that host challenges at and integrates well with the AnoMed competition platform.

Your first goal should be to make use of a web app factory, e.g. supervised_learning_MIA_challenge_server_factory, to avoid dealing with web programming issues. If this does not suit your use case, have a look at the more basic resource building block (e.g. StaticNumpyDataResource or UtilityResource.)

Module Contents

Classes

DataReconstructionPrivacyResource
DataReconstructionUtilityResource
DynamicNumpyDataResource	A dynamic NumPy dataset resource, i.e. the specific dataset content depends on GET request parameters.
StaticJSONResource	Any JSON serializable object, representing a “static” resource (i.e. a resource that does not depend on request parameters).
StaticNumpyDataResource	A static NumPy dataset resource, responding to GET requests.
UtilityResource	Evaluate the prediction of an estimator with respect to the ground truth and report the result statistics as a JSON string.

Functions

supervised_learning_MIA_challenge_server_factory	A factory to create a web application object which hosts a challenge.SupervisedLearningMIAChallenge, currently the most basic use case of challenges for the AnoMed competition platform.
tabular_data_reconstruction_challenge_server_factory	A factory to create a web application object which hosts a challenge.TabularDataReconstructionChallenge.

API

class anomed_challenge.challenge_server.DataReconstructionPrivacyResource(challenge_obj: anomed_challenge.challenge.TabularDataReconstructionChallenge, evaluation_submitter: Callable[[falcon.Request, dict[str, float]], None])

Initialization

on_post(req: falcon.Request, resp: falcon.Response) → None

class anomed_challenge.challenge_server.DataReconstructionUtilityResource(challenge_obj: anomed_challenge.challenge.TabularDataReconstructionChallenge, evaluation_submitter: Callable[[falcon.Request, dict[str, float]], None])

Initialization

on_post(req: falcon.Request, resp: falcon.Response) → None

class anomed_challenge.challenge_server.DynamicNumpyDataResource(individual_data_provider: Callable[[falcon.Request], anomed_challenge.challenge.NumpyDataset])

A dynamic NumPy dataset resource, i.e. the specific dataset content depends on GET request parameters.

The arrays (features and targets) will be serialized using anomed_utils.named_ndarrays_to_bytes and may be de-serialized using anomed_utils.bytes_to_named_ndarrays. The names of the arrays are ‘X’ for the features and ‘y’ for the targets.

Initialization

Parameters:

individual_data_provider (Callable[[falcon.Request], NumpyDataset]) – A function that, based on the GET request, returns a specific NumpyDataset.

on_get(req, resp: falcon.Response) → None

class anomed_challenge.challenge_server.StaticJSONResource(obj: Any)

Any JSON serializable object, representing a “static” resource (i.e. a resource that does not depend on request parameters).

The object will be represented as a plain JSON string, when a GET request is invoked.

Initialization

Parameters:

obj (Any) – A JSON serializable object, i.e. is should be compatible with json.dumps.

on_get(_, resp: falcon.Response)

class anomed_challenge.challenge_server.StaticNumpyDataResource(data: anomed_challenge.challenge.NumpyDataset)

A static NumPy dataset resource, responding to GET requests.

The arrays (features and targets) will be serialized using anomed_utils.named_ndarrays_to_bytes and may be de-serialized using anomed_utils.bytes_to_named_ndarrays. The names of the arrays are ‘X’ for the features and ‘y’ for the targets.

Initialization

on_get(_, resp: falcon.Response) → None

anomed_challenge.challenge_server.supervised_learning_MIA_challenge_server_factory(challenge_obj: anomed_challenge.challenge.SupervisedLearningMIAChallenge) → falcon.App

A factory to create a web application object which hosts a challenge.SupervisedLearningMIAChallenge, currently the most basic use case of challenges for the AnoMed competition platform.

By using this factory, you don’t have to worry any web-programming issues, as they are hidden from you. The generated web app will feature the following routes:

• [GET] /

• [GET] /data/anonymizer/training

• [GET] /data/anonymizer/tuning

• [GET] /data/anonymizer/validation

• [GET] /data/deanonymizer/members

• [GET] /data/deanonymizer/non-members

• [GET] /data/attack-success-evaluation

• [POST] /utility/anonymizer

• [POST] /utility/deanonymizer

Parameters:

challenge_obj (challenge.SupervisedLearningMIAChallenge) – A supervised learning, MIA threat model challenge object providing the necessary data and means of utility and privacy evaluation.

Returns:

A web application object based on the falcon web framework.

Return type:

falcon.App

anomed_challenge.challenge_server.tabular_data_reconstruction_challenge_server_factory(challenge_obj: anomed_challenge.challenge.TabularDataReconstructionChallenge) → falcon.App

A factory to create a web application object which hosts a challenge.TabularDataReconstructionChallenge.

By using this factory, you don’t have to worry any web-programming issues, as they are hidden from you. The generated web app will feature the following routes:

• [GET] /

• [GET] /data/anonymizer/leaky

• [GET] /data/anonymizer/background-knowledge

• [POST] /utility/anonymizer

• [POST] /utility/deanonymizer

Parameters:

challenge_obj (challenge.TabularDataReconstructionChallenge) – The data reconstruction challenge to lift to a web application.

Returns:

A web application object based on the falcon web framework, which features the supplied challenge_obj.

Return type:

falcon.App

class anomed_challenge.challenge_server.UtilityResource(target_data_provider: Callable[[falcon.Request], numpy.ndarray], evaluator: Callable[[numpy.ndarray, numpy.ndarray], dict[str, float]], submitter: Callable[[falcon.Request, dict[str, float]], None])

Evaluate the prediction of an estimator with respect to the ground truth and report the result statistics as a JSON string.

When receiving a POST request, extract a prediction array from the request’s body, compare it to the ground truth and report the evaluation results. The prediction array must be byte encoded compatible with anomed_utils.bytes_to_named_ndarrays and have the name ‘prediction’. The ground truth may depend on request parameters. The evaluation metrics used, are defined at runtime.

Initialization

Parameters:

• target_data_provider (Callable[[falcon.Request], np.ndarray]) – A callable returning a reference array (ground truth), depending on the request parameters. It is expected that this callable takes care of request (parameter) validation and raises a falcon.HTTPBadRequest in case there are

• evaluator (Callable[[np.ndarray, np.ndarray], dict[str, float]]) – A callable comparing the prediction (first argument, extracted from the POST request) with the ground truth (second argument, provided by target_data_provider), generating evaluation statistics (a dictionary of float values). That dictionary will be the JSON string response.

• submitter (Callable[falcon.Request, [dict[str, float]], None]) – A callable which will submit the statistics generated by evaluator to the AnoMed competition platform (where it will be persisted). It may obtain submission-critical information from the request. It is expected that the this callable raises an falcon.HTTPInternalServerError, if the submission fails.

on_post(req: falcon.Request, resp: falcon.Response) → None

Raises:

falcon.HTTPBadRequest – If extracting an array with name ‘prediction’ from the POST request body fails.