Multik v0.2.0 is out! 🎉 Now Multik is a multiplatform library, with native support for Android and Apple Silicon: https://blog.jetbrains.com/kotlin/2022/07/multik-0-2-multiplatform-with-support-for-android-and-apple-silicon Slack Conversation

@Pavel Gorgulov This seems to be a bug. The type should be DN, not D4

@Pavel Gorgulov I finished migrating KMath multik integration to MPP: https://github.com/mipt-npm/kmath/commit/a8182fad23417f4302ea77ccdeba3b931e50a471. The only major problem is that service provider mechanism does not work on JS (see https://youtrack.jetbrains.com/issue/KT-53056/Multiplatform-equivalent-of-JVMs-ServiceLoader). So getters like

Multik.stat

fail by default there. Engines must be loaded manually and it should be mentioned in the documentation.

I am thinking about replacing all @Ролан postfix notations in tensors like

tensor.lup()

with prefix notations like

Algebra::lup(tensor)

. It allows much more robust use of contexts and extensions and in my opinion is much more readable. In numpy/Torch postfix notation is used mostly because it could not be done in the other way. The only thing we loose is easy chaining (it could be brought back via scope functions). Does anybody know a case with real complicated chaining on tensors?

I'm experimenting with writing a DSL for AST matching (and hopefully also property extraction). I'm curious if anyone knows of prior art.

Hey there, why does

0.42.rem(0.14)

return

0.13999999999999996

and how can I fix it?

KMath-polynomial package by @Gleb Minaev was to heavy for our primary repository, so it has moved to a separate one: https://github.com/SciProgCentre/kmath-polynomial. Gleb is maintaining it now. The library is dedicated to operations on polynomials and rational functions.

Given x (values from 0..Int.MAX_VALUE), when

x in 0 until a

, return 0,

a until b

return 1,

b until c

return 2, greater than or equal to

d

, return 3?

A naïve classifier (proto convolutional network), using Commons Math and Plotly-kt: https://github.com/SciProgCentre/kmath/blob/dev/examples/notebooks/Naive%20classifier.ipynb

Have been playing around with Multik this weekend and have to say nice performance out of the box (I’m using the JVM version only). One thing I couldn’t figure out how to efficiently copy/set data. What I want to do in a matrix (D2Array<Float>) is to move the last x rows to the first x rows. I use a “hack”, but I feel this is not very elegantly:

data = data[rows-x until rows].cat(mk.zeros(rows-x, columns))

Does anyone knows a better way of doing this (performance is key) ?

Given a 2D square array, how do I get the row and column given a position? Can't quite get the modulos right. E.g. in a 5x5 array, how do I find the 13th element?

Is KMath going to have general purpose equation solving and simplification as sympy does?

Hey all! Are there any plans to release a new version of KMath with

iosSimulatorArm64

as target?

Hello! Is there KMath analog for org.apache.commons.math3.analysis.interpolation.PiecewiseBicubicSplineInterpolatingFunction?

They say "do not release on Friday". How about releases on Sunday? KMath 0.3.1 pre-release is out: https://github.com/SciProgCentre/kmath/releases/tag/v0.3.1-dev-RC and available on http://repo.kotlin.link. Documentation is still a bit of a mess. But on the other hand we now have Wasm support for core modules.

Hey there! We’re using Multik in production, and it’s working great - most of the time, because it crashes for some of our users with:

IllegalStateException: Fail to find engine

What’s weird about this is that it is happening only sometimes. Did anyone ran into this issue? Link: https://github.com/Kotlin/multik/issues/155

KMath 0.3.1 is released on mavencentral (without ios artefacts) and

repo.kotlin.link

(with ios artefacts). It contains a lot of minor and major changes and new features: https://github.com/SciProgCentre/kmath/releases. Additionally there is a huge new separate KMath-based project kmath-polynomials: https://github.com/SciProgCentre/kmath-polynomial with algebras for polynomials and rational functions maintained by @Gleb Minaev. A lot of work has been done. A lot more still awaits. Slack Conversation

dear community, I'm not a mathematician so please forgive me if I can't express my concern correctly. In my scenario, I have X and Y values that represent a connection. I would now like to use any X value and calculate the Y for it. I've already done this in a spreadsheet using a 4th degree polynomial. Now my question, can I solve this question with kmath and if so, how? Thank you for your time and effort. Daniel

Is there an official or widely used BigDecimal KMM equivalent, or do I need to

expect

these various implementations? https://discuss.kotlinlang.org/t/multiplatform-bigdecimal-implementation/5631/10

Hi! Do you know if

kotlin-statistics

has been moved somewhere else? I can't find it on GH anymore: https://github.com/thomasnield/kotlin-statistics

I am writing a program to calculate wages (usually doubles are a no-no, but I think it is ok in this case). Part of the company’s requirements are that the new salary must be expressed in terms of the old rate with a proportionate number of hours added to the total hours worked. This function is not giving me the answers that a colleague drafted on paper - are our equations correct?

/**
Given:
old rate = $30/hour
new rate = $37.5/hour
Hours in new rate = 36.72
Total hours = 52.21

Equations:
1/x := 30/37.5 == 37.5/30 = x; x=1.25, so 1 hour of the new rate is worth 1.25 hours in the old rate.

Hours in old rate = total hours - hours in new rate = 52.21 - 32.67 = 15.49
(1.25 * hours worked in new rate) * old rate = equivalent hours in new rate * old rate = salary for hours in new rate

Calculations:
1.25   * 36.72   = 45.9 upscaled hours
45.9   * 30      = $1,377  for hours in new rate
15.49  * 30      = $464.70 for hours in old rate
$1,377 + $464.70 = $1,841.70 total
Check: 
36.72  * $37.5   = $1,377  for hours in new rate
15.49  * $30     = $464.70 for hours in old rate
$1,377 + $464.70 = $1,841.70 total
 * */

fun calculateHoursInNewRate(oldRatePerHour: Double,newRatePerHour: Double,totalHoursWorked: Double,hoursInNewRateWorked: Double): Double {
    val equivalentOfOldHoursWorkedInNewRate = newRatePerHour / oldRatePerHour
    val hoursInOldRate = totalHoursWorked - hoursInNewRateWorked
    val upscaledHours = equivalentOfOldHoursWorkedInNewRate * hoursInNewRateWorked
    val salaryForHoursInNewRateWorked = upscaledHours * oldRatePerHour
    val salaryForHoursInOldRateWorked = hoursInOldRate * oldRatePerHour
    return salaryForHoursInNewRateWorked + salaryForHoursInOldRateWorked
}

Interesting fact. It seems like using newer Oracle GraalVM JIT significantly improves performance even for Multik

@Iaroslav Postovalov here are the results for Oracle GraalVM 21:

jvm summary:
Benchmark                            Mode  Cnt   Score   Error  Units
DotBenchmark.bufferedDot            thrpt    5   1.336 ± 0.026  ops/s
DotBenchmark.cmDot                  thrpt    5   0.836 ± 0.027  ops/s
DotBenchmark.cmDotWithConversion    thrpt    5   0.795 ± 0.344  ops/s
DotBenchmark.ejmlDot                thrpt    5   6.532 ± 0.066  ops/s
DotBenchmark.ejmlDotWithConversion  thrpt    5   6.051 ± 0.180  ops/s
DotBenchmark.multikDot              thrpt    5  40.559 ± 4.751  ops/s
DotBenchmark.parallelDot            thrpt    5  11.879 ± 1.117  ops/s
DotBenchmark.tensorDot              thrpt    5   1.183 ± 0.125  ops/s
DotBenchmark.tfDot                  thrpt    5   6.013 ± 0.323  ops/s

Has anyone here ever written a mathematics engine in Kotlin? If so, describe your implementation or provide a code snippet. I am using mathematical algorithms as references. I am currently building a mathematics engine to challenge Microsoft Math Solver but Open Source. How have you used KMath?

I'm trying to do matrix multiplication in multik (https://github.com/Kotlin/multik). It seems like I should get a 2 X 3 matrix, but I get an error message that says "Array shapes don't match: [2, 1] != [1, 3]". Can anyone say if this is a bug or am I doing something horribly wrong?

val a = mk.ndarray(floatArrayOf(1.0f, 2.0f), 2, 1)
val b = mk.ndarray(floatArrayOf(1.0f, 2.0f, 3.0f), 1, 3)

val result = a * b
println(result)

It would be interesting to see benchmarks comparing to other languages / tools that are being used for solving same problems.

Hey, I'm a little confused when it comes to Kotlin math libs. I'm looking for a good library to use for 3D graphics (matrix multiplications etc). There is kmath by romain guy, multik by Kotlin itself, korlibs-math, and some others. So which one fits best?

KMarth 0.4.1 is released: https://github.com/SciProgCentre/kmath/releases/tag/v0.4.1 • Moved to Kotlin 2.1 and enabled cross-compilation for iOS targets (now Maven central releases will be more frequent, but we can't run tests on iOS artefacts) • Added ojalgo wrapper. • Refactored Ejml and Commons-math linear spaces to match common pattern. • Small additions and refactoring.

KMath 0.4.2 is released. It fixes one critical bug in ojalgo and makes some changes to unstable API for matrix builders: https://github.com/SciProgCentre/kmath/releases/tag/v0.4.2