JarCastingDo you need an API to export machine learning configurations and outcomes?
- LOG4MEX helps developers to manage ML metadata, improving interoperability and adding provenance information to the experiment.
- No need to create scripts to manage the outputs.
- No need to create wrappers to save outputs into SGBD's.
- Exporting clear information about your executions! Others do not need to guess what you did! (LOG4MEX is based on MEX Vocabulary)
Requirements
Java 8, Maven 3
License
LOG4MEX - a MEX Project - is Open Source and licensed under the GNU General Public License 3 (Copyright (c) 2014-2016, Diego Esteves).
Demo and Documentation
Bugs
Found a
Changelog
[v1.0.2]
- MEX vocabulary update
- Fixing serialization bug
[v1.0.1]
- MEX vocabulary update
- Adding support to more semantic formats
[v1.0.0]
- Initial version
Usage
1) The basics
Instantiating the main class for the experiment
MyMEX mex = new MyMEX();
Saving some basic provenance informations
mex.setAuthorName("James Walter");
mex.setAuthorEmail("[email protected]");
mex.setContext(EnumContexts.STOCK_MARKET_PREDICTIONS);
mex.setOrganization("Leipzig University");
mex.setExperimentTitle("My Experiment 2016");
mex.setExperimentDate(new Date());
...
2) Configurations
How many dimensions you want to have?
You can either represent an unique configuration and access the linked objects direct from it
...
mex.Configuration().
...
or group your executions by configurations, e.g.: 1) analysis for dataset A 2) analysis for dataset B, and access the objects by configuration
String c1 = mex.addConfiguration();
String c2 = mex.addConfiguration();
...
mex.Configuration(c1).
...
For each configuration, you define your environment
...
mex.Configuration().setDescription("here you can explain the logical group you've created");
mex.Configuration().setModel("model8745.jd", "svm model based on alpha=0.05 and C=1", "2016-05-10");
mex.Configuration().setHardwareConfiguration("ubuntu", EnumProcessors.INTEL_COREI7, EnumRAM.SIZE_16GB, "SSD", EnumCaches.CACHE_3MB);
mex.Configuration().setDataSet("http://www.bmfbovespa.com.br/shared/iframe.aspx?idioma=pt-br&url=http://www.bmfbovespa.com.br/pt-br/cotacoes-historicas/FormSeriesHistoricas.asp", "bovespads", "bovespa");
mex.Configuration().setSamplingMethod(EnumSamplingMethods.CROSS_VALIDATION, 10);
mex.Configuration().setTool(EnumTools.WEKA, "3.6.6");
mex.Configuration().addFeature(new String[]{"openp", "closep", "minp", "maxp"});
...
What are your algorithms? You can also refer to them in case you need to specify different hyperparameters values
...
String alg1 = mex.Configuration().addAlgorithm("svm", EnumAlgorithmsClasses.SupportVectorMachines);
String alg2 = mex.Configuration().addAlgorithm("svm", EnumAlgorithmsClasses.SupportVectorMachines);
mex.Configuration().addAlgorithm("nb", EnumAlgorithmsClasses.NaiveBayes);
...
mex.Configuration().Algorithm(alg1).addParameter(new String[]{"C", "10^3", "alpha", "0.2"});
mex.Configuration().Algorithm(alg2).addParameter(new String[]{"C", "10^3", "alpha", "0.4"});
...
Your runs: would like to represent every single execution (EnumExecutionsType.SINGLE) or a summary for a set of executions (EnumExecutionsType.OVERALL)? Are you training (EnumPhases.TRAIN) or testing (EnumPhases.TEST)) your model?
...
String ex1 = mex.Configuration().addExecution(EnumExecutionsType.OVERALL, EnumPhases.TRAIN);
mex.Configuration().Execution(ex1).setStartDate(new Date());
mex.Configuration().Execution(ex1).setAlgorithm(alg1);
mex.Configuration().Execution(ex1).setStartsAtPosition("1233");
mex.Configuration().Execution(ex1).setEndsAtPosition("1376");
...
mex.Configuration().Execution(ex1).setEndDate(new Date());
Adding performance measures
...
mex.Configuration().Execution(ex1).addPerformance(EnumMeasures.ACCURACY, .96);
mex.Configuration().Execution(ex1).addPerformance(EnumMeasures.TRUEPOSITIVERATE, .70);
...
Parsing and generating the mex file
MEXSerializer.getInstance().saveToDisk(<file path>, <uri>, mex, MEXConstant.EnumRDFFormats.JSON_LD);
...
A Simple Example
MyMEX_10 mex = new MyMEX_10();
//basic authoring provenance
mex.setAuthorName("D Esteves");
mex.setAuthorEmail("[email protected]");
//the algorithm and features
String[] features = {"min", "max", "ope clo"};
mex.Configuration().addFeature(features);
mex.Configuration().addAlgorithm(MEXEnum.EnumAlgorithm.NaiveBayes);
// -> your model's call here !
//setting the phase for the run
String executionId = mex.Configuration().addExecutionOverall(MEXEnum.EnumPhase.TEST);
//adding the performance(s)
mex.Configuration().ExecutionOverall(executionId).setAlgorithm(mex.Configuration().Algorithm(EnumAlgorithm.NaiveBayes));
mex.Configuration().ExecutionOverall(executionId).addPerformance(EnumMeasures.ACCURACY.toString(), .96);
try{
//parsing and saving the mex file
MEXSerializer_10.getInstance().parse(mex);
MEXSerializer_10.getInstance().saveToDisk("ex001.ttl","http://mex.aksw.org/examples/001/", mex);
}catch (Exception e){
System.out.println(e.toString());
}
More Examples
More basic examples are available here. You can also visit the project wiki in order to check recent updates and further use cases.