Multi Machine Learning (MML)

The goal of this project is to provide high-level facilities to perform machine learning tasks (e.g classification). The user "just" has to specify a configuration file (in e.g JSON) with information about the dataset, the predictive variables, etc. Then, her specification is compiled in different languages/libraries and can then be executed. In a sense, doing machine learning in a declarative way, without fighting with the cryptic details of off-the-self libraries.

Java implementation

The processing of configuration files and compilers are written in Java.

Python

Python 3 should be installed with the pandas and scikit-learn library: pip install -r requirements.txt

warning: on Windows system, the python command can well be py

R

R should be installed with rpart library

Julia

Julia should be installed and requires the following packages:

• DataFrames
• DecisionTree
• CSV
• StatsBase
• ScikitLearn
• DecisionTree
• Random
• MLMetrics

With the interactive mode of Julia, use using Pkg() then Pkg.add(<package name>) for each package to install them.

Maven

(Commands are for Linux)

To perform tests with JUnit use mvn test in the root folder of this project.

To package the project as an executable JAR use mvn package in the root folder of this project. Then get the JAR file (WITH DEPENDENCIES) in the /target folder. This JAR file can be executed with the command java -jar <filepath>.

How to use the application

• First launch the server. There are two ways to do this

  • use Eclipse or another IDE to open the project and run the class "ServerMain"
  • package the project using Maven as indicated above (or use the one in the root folder), then run the JAR file

• Then open the index.html file in your browser.

• Once you are on the webpage, you can get instructions, dowload an example JSON, and upload a JSON file which matches this kind of format (do not use it as it is, this is just an example, delete # lines first) :

CAUTION: the following parameters which are lists work serverside (they all the results) with several elements, however the web app only handles one result, so you should use one-element lists when it is recommended in order to use the web app correctly.

{
"dataset": {
"filename" : "", #path to your csv file
"separator" : "," #enter your separator here
},
"training": [0.7], #or any list of training ratios in ]0,1[, ONE ELEMENT ONLY IS RECOMMENDED
"target_variable": "variety", #you target variable
"metrics" : ['accuracy', 'confusion', 'macro_precision', 'macro_recall', 'macro_f1'], #you can select a sublist of this one
"max_depth" : [5], #or any list of integers above 1, ONE ELEMENT ONLY IS RECOMMENDED
"languages" : ["Python", "R", "Julia"], #you can take a sublist of this one, ONE ELEMENT ONLY IS RECOMMENDED
"repetition" : 1 #or any integer
}

If you lack inspiration, you can try file_to_upload.json

• Then click on the "Upload file" button and wait until "Success" is displayed.

• Finally, click on "Open result" to print the result.

• Click on the cross to close the window and delete the uploaded JSON.

• Shut the server down.

Contact

[email protected] [email protected] [email protected]