Skip to content

Latest commit

 

History

History
executable file
·
223 lines (131 loc) · 7.6 KB

Changes.md

File metadata and controls

executable file
·
223 lines (131 loc) · 7.6 KB

Changes in the BigML.io Java bindings

1.5.2 version

Improvements

  • Association Resource
  • AssociationSet Resource
  • TopicModel Resource
  • TopicDistribution Resource
  • BatchTopicDistribution Resource
  • Configuration Resource

Bugs

  • Fixed bug creating prediction from ensemble, input data was no being used for the prediction.

  • Fixed typos and updated document URLs.

  • Fixed test suite.

1.5.1 version

Improvements

  • Create evaluation from Logistic Regression.

1.5 version

Improvements

  • Create dataset from list of Datasets
  • Correlation Resource
  • StatisticalTest Resource
  • Logistic Regression Resource
  • Tests renaming and fixing
  • Made Local Ensemble and dependencies serializable
  • Preditcion/BatchPrediction: create from LogisticRegression
  • Create model from cluster
  • Whizzml Resources

1.4.2 version

Bugs

  • Fixed a LocalEnsemble prediction bug with the comparability option.

  • Fixed invalid samples and errors/typos.

  • Provided a workaround for JSON escaping forward slashes (/).

Improvements

  • Source didn't accept arguments other than name and source_parser. It now accepts all other arguments.

  • An initial version of BigML Java binding document is created and published at http://bigml-java.readthedocs.org.

1.4.1 version

Bugs

  • Fixed bug in the constructor of the LocalEnsemble. The JSON path used to access objects was wrong.

1.4 version

New Features

  • Added support for Local Anomalies

    A new LocalAnomaly class to score anomalies in a dataset locally or embedded into your application without needing to send requests to BigML.io.

    This class cannot only save you a few credits, but also enormously reduce the latency for each prediction and let you use your models offline.

    Example usage (assuming that you have previously set up the BIGML_USERNAME and BIGML_API_KEY environment variables and that you own the model/id below):

      // API client
  BigMLClient api = BigMLClient.getInstance();

  // Retrieve a remote anomaly by id
  JSONObject jsonAnomaly = api.getAnomaly("anomaly/551aa203af447f5484000ec0");

  // A lightweight wrapper around a Anomaly resurce
  LocalAnomaly localAnomaly = new LocalAnomaly(jsonAnomaly);

  // Input data
  JSONObject inputData = (JSONObject) JSONValue.parse("{\"src_bytes\": 350}");

  // Calculate score
  localAnomaly.score(inputData);

  • Added support for Samples resources.

    Full API documentation on the API can be found from BigML at: https://bigml.com/developers/samples

  • Added support for Project resources.

    Full API documentation on the API can be found from BigML at: https://bigml.com/developers/project

  • Added support for predictions using Median.

Improvements

  • We improve the predictions responses adding the more information: Distribution, DistributionUnit, Count and Median values associated to the prediction.

  • Added methods in LocalModel class to get information of the model. Added methods: Summarize, Data Distribution and Prediction Distribution.

  • Added BaseModel and ModelFields base classes to reuse logic in LocalCluster and LocalPredictiveModel

1.3 version

New Features

  • Added support for Anomaly detectors. Predictive model that can help identify the instances within a dataset that do not conform to a regular pattern.

    Now you can create new anomaly detectors, retrieve individual anomaly detectors, list your anomaly detectors, delete, and also update your anomaly detectors.

  • Added support for remote Anomaly Scores. This resources will automatically compute a score between 0 and 1. The closer the score is to 1, the more anomalous the instance being scored is. The Anomaly Score will also compute the relative importance for each field. That is, how much each value in the input data contributed to the score.

    Now you can create new anomaly scores, retrieve individual anomaly scores, list your anomaly scores, delete, and also update your anomaly scores.

  • Added support for remote Batch Anomaly Scores. A batch anomaly score provides an easy way to compute an anomaly score for each instance in a dataset in only one request. To create a new batch anomaly score you will need an anomaly/id and a dataset/id.

  • Local Ensembles improvements.

** Access to the Field Importance information

** New combiners supported in Local Ensembles:.

confidence weighted: uses prediction confidences as weight for the combined prediction.

probability weighted: uses the probability of the class in the distribution of classes in
            the leaf node as weight.

threshold-based: uses a given threshold k (by default 1) to predict a given class (by default the
                 minority class). You can set up both using the threshold argument. If there are less
                 than k models voting class, the most frequent of the remaining categories is chosen,
                 as in a plurality combination after removing the models that were voting for class.
                 The confidence of the prediction is computed as that of a plurality vote, excluding
                 votes for the majority class when it's not selected.

  • Local Predictions. Added support to "last prediction" and "proportional" missing strategies.

  • Local Clusters and Local Centroids. You can now instantiate a local version of a remote cluster that you can use to make local centroid predictions associated to an input data set.

    You must keep in mind, though, that to obtain a centroid prediction, input data must have values for all the numeric fields. No missing values for the numeric fields are allowed.

    As in the local model predictions, producing local centroids can be done independently of BigML servers, so no cost or connection latencies are involved.

  • Now is possible to create sources using the results of a batch anomaly score.

  • Now is possible to create sources using the results of a batch prediction.

  • Allow save and load batch predictions to/into local CSV files.

  • Added support for two different missing strategies for local predictions: last prediction and proportional.

  • Added support for download the content of a dataset.

  • Added support for download the result of a batch prediction.

  • We can change the BigML Domain programmatically (Ex. https://myprivatedeploy.com)

Bugs

  • Fix bug sending json with UTF-8 characters. We were using the DataOutputStream class to send the data to the server instead the OutputStreamWriter class.

  • Fix bug in multipart request. The connection was not using the MockTrustManagers.

1.2 version

New Features

  • Since version 1.2 we start to publish the release on the Maven Central Repository. Now, you can declare the library dependency in your project's pom file.

    If you were using the Apache HttpClient library in your own project, now you will need to explicitly declare this dependency in your own project's pom file, because it's not anymore a dependency of our library.

  • We have removed the unnecessary dependencies with third party libraries: Apache HttpClient, SLF4J-Log4j and SuperCsv.

  • Support for json files in UTF-8 encoding. Now we can send json objects with UTF-8 characters inside.

  • Added the system property BIGML_SEED to allow clients to generate deterministic models.

  • Added a Seed to the tests to make them deterministic

  • Contact support information has been updated

Bugs

  • Fixed bug in predictions when we were using fields in the input data that had periods in their names.

1.1 version

  • Since version 1.1 the name of the JAR file is bigml-binding.