argamak 🐎

Hex Package Hex Docs License Build

A Gleam library for tensor maths.

“I admire the elegance of your method of computation; it must be nice to ride through these fields upon the horse of true mathematics while the like of us have to make our way laboriously on foot.”

—Albert Einstein, to Tullio Levi-Civita, circa 1915–1917

Argamak: A shiny steed.


As a dependency of your Gleam project

• Add argamak to gleam.toml from the command line

$ gleam add argamak

As a dependency of your Mix project

• Add argamak to mix.exs

defp deps do
   {:argamak, "~> 1.1"},

As a dependency of your Rebar3 project

• Add argamak to rebar.config

{deps, [
 {argamak, "1.1.0"}


The @tensorflow/tfjs package is a runtime requirement for argamak; however, its import path in the argamak_ffi.mjs module might need adjustment, depending on your use case. It can be used as is in your Node.js project after running npm install @tensorflow/tfjs-node or an equivalent command for your package manager of choice.


// derby.gleam
import gleam/function
import gleam/io
import gleam/list
import gleam/result
import gleam/string
import argamak/axis.{Axis, Infer}
import argamak/space
import argamak/tensor.{type TensorError, InvalidData}

pub fn announce_winner(
 from horses: List(String),
 with times: List(Float),
) -> Result(Nil, TensorError) {
 // Space records help maintain a clear understanding of a Tensor's data.
 // We begin by creating a two-dimensional Space with "Horse" and "Trial" Axes.
 // The "Trial" Axis size is two because horses always run twice in our derby.
 // The "Horse" Axis size will be inferred based on the data when a Tensor is
 // put into our Space (perhaps we won't always know how many horses will run).
 use d2 <- result.try(
   space.d2(Infer(name: "Horse"), Axis(name: "Trial", size: 2))
   |> result.map_error(with: tensor.SpaceErrors),

 // Every Tensor has a numerical Format, a Space, and some data.
 // A 2d Tensor can be visualized like a table or matrix.
 // Tensor(
 //   Format(Float32)
 //   Space(Axis("Horse", 5), Axis("Trial", 2))
 //                Trial
 //  H [[horse1_time1, horse1_time2],
 //  o  [horse2_time1, horse2_time2],
 //  r  [horse3_time1, horse3_time2],
 //  s  [horse4_time1, horse4_time2],
 //  e  [horse5_time1, horse5_time2]],
 // )
 // Next we create a Tensor from a List of times and put it into our 2d Space.
 use x <- result.try(tensor.from_floats(of: times, into: d2))

 let announce = function.compose(string.inspect, io.println)

 announce("Trial times per horse")

 // Axes can be referenced by name.
 // Here we reduce away the "Trial" Axis to get each horse's mean run time.
 announce("Mean time per horse")
 let mean_times =
   |> tensor.mean(with: fn(a) { == "Trial" })
   |> tensor.debug

 // This catch-all function will reduce away all Axes, although at this point
 // only the "Horse" Axis remains.
 let all_axes = fn(_) { True }

 // We get a String representation of the minimum mean time.
 announce("Fastest mean time")
 let time =
   |> tensor.min_over(with: all_axes)
   |> tensor.debug
   |> tensor.to_string(return: tensor.Data, wrap_at: 0)

 // And we get an index number, followed by the name of the winning horse.
 announce("Fastest horse")
 use horse <- result.try(
   |> tensor.arg_min(with: all_axes)
   |> tensor.debug
   |> tensor.to_int,
 use horse <- result.try(
   |> horse)
   |> result.replace_error(InvalidData),

 // Finally, we make our announcement!
 { horse <> " wins the day with a mean time of " <> time <> " minutes!" }
 |> announce
 |> Ok


> derby.announce_winner(
>   from: ["Pony Express", "Hay Girl", "Low Rider"],
>   with: [1.2, 1.3, 1.3, 1.0, 1.5, 0.9],
> )
"Trial times per horse"
 Space(Axis("Horse", 3), Axis("Trial", 2)),
 [[1.2, 1.3],
  [1.3, 1.0],
  [1.5, 0.9]],
"Mean time per horse"
 Space(Axis("Horse", 3)),
 [1.25, 1.15,  1.2],
"Fastest mean time"
"Fastest horse"
"Hay Girl wins the day with a mean time of 1.15 minutes!"
Search Document