Machine Learning in Kotlin: Overview of Libraries for Data Science and ML

Machine Learning in Kotlin: Modern Libraries and Tools

Kotlin is a statically typed programming language that runs on top of the JVM and is fully compatible with Java. Thanks to its concise syntax, null safety, and powerful ecosystem, Kotlin is increasingly used in Data Science and machine learning. In this article, we will review key libraries that allow solving ML tasks directly in Kotlin, without switching to Python.

1. KotlinDL — Deep Learning Library from JetBrains

KotlinDL is the official library from the creators of Kotlin (JetBrains), designed for building and training neural networks. It supports both a high-level API (Keras-like) and low-level tensor operations.

Key Features:

• Model building: Sequential, Functional API
• Layers: Dense, Conv2D, MaxPooling, Dropout, LSTM, etc.
• Optimizers: Adam, SGD, RMSprop
• Loss functions: CrossEntropy, MSE, MAE
• Loading pre-trained models (ONNX, TensorFlow SavedModel)
• Working with images, text, and tabular data

Example: Simple Neural Network for Classification

import org.jetbrains.kotlinx.dl.api.core.Sequentialimport org.jetbrains.kotlinx.dl.api.core.layer.core.Denseimport org.jetbrains.kotlinx.dl.api.core.layer.reshaping.Flattenimport org.jetbrains.kotlinx.dl.api.core.optimizer.Adamimport org.jetbrains.kotlinx.dl.dataset.mnist

fun main() {    val model = Sequential.of(        Flatten(inputShape = longArrayOf(28, 28)),        Dense(128, activation = "relu"),        Dense(10, activation = "softmax")    )

    model.use {        it.compile(optimizer = Adam(), loss = "categoricalCrossEntropy")        val (train, test) = mnist()        it.fit(train, epochs = 5, batchSize = 32)        val accuracy = it.evaluate(test)        println("Accuracy: $accuracy")    }}

2. Smile — Universal Library for Statistics and ML

Smile (Statistical Machine Intelligence & Learning Engine) is a powerful Java/Kotlin library covering virtually all classical machine learning algorithms. It includes implementations of over 200 algorithms.

Key Modules:

• Classification: SVM, Random Forest, Decision Trees, Naive Bayes
• Regression: Linear, Ridge, LASSO, Gaussian Process
• Clustering: K-Means, DBSCAN, Hierarchical, GMM
• Dimensionality Reduction: PCA, t-SNE, MDS
• Natural Language Processing (NLP): tokenization, stemming, N-grams
• Visualization: charts, scatter plots, heatmaps

Example: Linear Regression with Smile

import smile.regression.OLSimport smile.data.DataFrameimport smile.data.type.DataTypesimport smile.data.vector.DoubleVector

fun main() {    // Create a DataFrame with data    val x = doubleArrayOf(1.0, 2.0, 3.0, 4.0, 5.0)    val y = doubleArrayOf(2.1, 4.2, 6.0, 7.9, 10.1)    val df = DataFrame.of(DoubleVector.of("x", x), DoubleVector.of("y", y))

    // Train an OLS model (Ordinary Least Squares)    val model = OLS.fit(df, "y", "x")    println("Coefficients: ${model.coefficients()}")    println("Prediction for x=6: ${model.predict(doubleArrayOf(6.0))}")}

3. Koma — Matrix Computations and Scientific Calculations

Koma is a library for working with multidimensional arrays and matrices, inspired by NumPy from Python. It is ideal for numerical calculations, statistics, and data preprocessing before ML.

Features:

• Matrix creation: zeros, ones, eye, rand
• Operations: addition, multiplication, transposition, matrix inverse
• Linear algebra: SV