remisultan/java-ml

This repo intends to implement Machine Learning algorithms with Java and https://github.com/deeplearning4j/nd4j purely for understanding

java-ml

This repo intends to implement Machine Learning algorithms with Java and https://github.com/deeplearning4j/nd4j purely for understanding

You can check for examples in this package: src/main/java/org/rsultan/example

Requirements

• JDK16+

Getting started

Clone the repo and add the dependency according to your version

 $ git fetch --all --tags
 $ git checkout tags/<latest-release> -b <your-branch>  
 $ ./mvnw clean install

And then add the corresponding dependency

    <dependency>
      <groupId>org.rsultan</groupId>
      <artifactId>java-ml</artifactId>
      <version>[latest-version]</version>
    </dependency>

Once your are good with this, you can go read the wiki

You can also check the examples in the java-ml-example module.

Good luck !

java-ml's Issues

csv reader to take into account enclosures to remove to each field

If the field is contained within enclosures that requested (double quotes are default) we could remove them from the field

java-ml/src/main/java/org/rsultan/utils/CSVUtils.java

Line 42 in 8699d16

var typedValue = getValueWithType(lineArray[index]);

mean / median shift should accept double value as bandwidth

The constructor Meanshift and Medianshit was only accepting a long and not a double bandwidth as a parameter

dataframe.toMatrix() not casting negative numbers

Here is the regex which determines if a field is a number

java-ml/src/main/java/org/rsultan/dataframe/transform/matrix/MatrixDataframe.java

Line 22 in a961500

private static final String NUMBER_REGEX = "^\\d+(\\.\\d+)*$";

java-ml/src/main/java/org/rsultan/utils/CSVUtils.java

Line 18 in 8699d16

private static final Pattern DOUBLE_VALUE_REGEX = Pattern.compile("\\d+\\.\\d+");

java-ml/src/main/java/org/rsultan/utils/CSVUtils.java

Line 19 in 8699d16

private static final Pattern LONG_VALUE_REGEX = Pattern.compile("\\d+");

We need to take into account negative numbers

Implement AdaBoost

better logging for mean / median shift

Today System.out.println is the logger

give the possibility to save models

implement a visualization tool

redirect from the README --> Wiki page

Having a better README is mandatory

implement ALS recommendation algorithm

implement random forests

create toolset for model evaluation

This would mean to include

• Confusion Matrix and F1 score
• Cross validation
• and other techniques known to the world

implement DBSCAN clustering

implement Principal Component Analysis (PCA) algorithm

https://en.wikipedia.org/wiki/Principal_component_analysis

implement Gradient Boosted Trees

make LinearRegression omit intercept