cric96 / scala-machine-learning-libraries

Scala ML libraries

In this repository, there are some minimal examples of ML scala libraries. the main focus here is the classification and regression algorithms. I don't eval mathematical libraries. I'm interested in plot capability, so some tests are done even for plot API (if present). Each library is contained in an sbt subproject having the same name as the library.

smile

A good alternative to sklearn. It is written in Java but has a scala DSL facade that makes the development easier.

pro:

• a good API;
• a vast number of algorithm about classification, regression, feature extraction and so on;
• a good performance;
• a clean design;
• modest plot support (also with vega);

cons:

• currently don't have a deep learning support;

doddle-model

A functional library that leverages type classes to implement some relevant ml algorithm (e.g. classification, regression, clustering).

pro

• a functional design;
• decent performance (comparable with sklearn);
• a good api;

cons

• it hasn't a big activity on GitHub;
• it doesn't support a lot of algorithms;
• it hasn't a plot support;
• it doesn't support deep learning.

BIDData

🚫 In this repository there aren't examples about this framework.

This is a very interesting library from a performance point of view. Despite this, it is hosted in a bintray repository, that will be dropped soon..

scikit-learn, TensorFlow, and Keras leveraging ScalaPy

ScalaPy is an interesting project. It allows the usage of Python libraries via Scala with an ad-hoc syntax. I have integrated Scala code with Keras and Sklearn almost without problems. There are some facades (like NumPy and TensorFlow), but they seem to be deprecated or not updated.

pro

• you can access the python ecosystem quite agile;
• you can leverage scala.native to improve the performance;
• it is possible to use python libraries with scala typing support.

cons

• it is still quite immature;
• there aren't a good support from the community (few and poor typing implemented);
• the documentation is quite limited;
• you need to implement you own typing if there aren't implement;
• matplotlib seems not to work.

MLlib (Spark based)

MLlib is Spark’s machine learning (ML) library. Its goal is to make practical machine learning scalable and easy. At a high level, it provides tools such as:

• ML Algorithms: common learning algorithms such as classification, regression, clustering, and collaborative filtering
• Featurization: feature extraction, transformation, dimensionality reduction, and selection
• Pipelines: tools for constructing, evaluating, and tuning ML Pipelines
• Persistence: saving and load algorithms, models, and Pipelines
• Utilities: linear algebra, statistics, data handling, etc.

from https://spark.apache.org/docs/latest/ml-guide.html

The spark documentation is very detailed, please refer to that.

pro

• it has a lot of ml algorithms implemented;
• it has a beautiful documentation;
• it supports cluster computation;
• the API is quite simple.
• it is a well-supported framework;

cons

• it could be an overkill (because you don't interested in scale up);
• the performance is not good as other alternatives;
• it requires more code than other solution;
• it hasn't DL support (but DL-Lib seems a solution).

BigDL (Spark based)

🚫 In this repository there aren't examples about this framework.

BigDL is a distributed deep learning library for Apache Spark; with BigDL, users can write their deep learning applications as standard Spark programs, which can directly run on top of existing Spark or Hadoop clusters.

from https://bigdl-project.github.io/master/

pro

• supported by intel;
• deep learning in a distributed platform;
• excellent performance;

cons

• not easy to use;
• there aren't a lot of examples;
• it is an overkill for little projects;

ScalNet a facade for DeepLearning4J

ScalNet is a wrapper around Deeplearning4J emulating a Keras like API for deep learning.

pro

• easy enough to use;
• good performance;
• DeepLearning4J is well documented and supported;
• Spark could be used to run deep learning on a distributed framework;

cons

• ScalNet is in beta yet;
• it seems to be less supported than the main project;
• it is not famous as other framework;