C++ Implementation of the Largest Triangle Three Buckets (LTTB) Downsampling Algorithm

This is a straightforward C++ port of the reference implementation of the Largest Triangle Three Buckets (LTTB) downsampling algorithm described in the paper "Downsampling time series for visual representation" by Sveinn Steinarsson. It is a single header, with a single class template that allows using different structures and data types.

How To Install

This is a single-header library with no dependencies outside the C++ standard library, so there are many ways to incorporate it into your project:

Installing Without A Tool

Copy include/lttb.hpp into your project, make sure it's available from your compiler's include path, and include it from your source:

#include "lttb.hpp"

Installing With vcpkg

In classic mode, install the lttb package:

vcpkg install lttb

In manifest mode, add lttb to your project's vcpkg.json:

vcpkg add port lttb

Then use the CMake package from your project in either mode:

find_package(lttb CONFIG REQUIRED)
target_link_libraries(your-target PRIVATE lttb::lttb)

Include the header from your source:

#include "lttb.hpp"

Installing With CMake

Clone this repository into your source tree, for example as a subrepo under external/lttb-cpp, then add it from your top-level CMakeLists.txt:

add_subdirectory(external/lttb-cpp)
target_link_libraries(your-target PRIVATE lttb::lttb)

The lttb::lttb target adds the library's include directory and requires C++11.

Include the header from your source:

#include "lttb.hpp"

How To Use With LLMs

This repository includes an optional MCP server that helps AI coding tools generate and review lttb integrations. The server is dependency-free and runs over stdio.

Clone the upstream repository if you do not already have it locally:

git clone https://github.com/parkertomatoes/lttb-cpp.git

Registering with Codex:

codex mcp add lttb-helper -- python3 /path/to/lttb-cpp/mcp/lttb_mcp.py

Registering with Claude Code:

claude mcp add --transport stdio lttb-helper -- python3 /path/to/lttb-cpp/mcp/lttb_mcp.py

To share the Claude Code MCP configuration with other contributors, add it with project scope from the repository root:

claude mcp add --transport stdio --scope project lttb-helper -- python3 ./mcp/lttb_mcp.py

Other MCP Clients

Add a stdio server entry like this:

{
  "mcpServers": {
    "lttb-helper": {
      "command": "python3",
      "args": [
        "/path/to/lttb-cpp/mcp/lttb_mcp.py"
      ]
    }
  }
}

The MCP exposes helpers for API reference, CMake snippets, C++ usage snippets, and common integration checks.

How To Use

Create a typedef to specify your time series datapoint type

#include "lttb.hpp"

struct ExamplePoint {
    float x;
    float y;
};

using PointLttb = LargestTriangleThreeBuckets<ExamplePoint, float, &ExamplePoint::x, &ExamplePoint::y>

Then use the static method Downsample in the class. It can be used with iterators

std::vector<ExamplePoint> in = GetYourInputsFromSomewhere();
std::vector<ExamplePoint> out;
PointLttb::Downsample(in.begin(), in.size(), std::back_inserter(out), 50);

...or pointers:

ExamplePoint in[500];
GetYourInputsFromSomewhere(in);
ExamplePoint out[50];
PointLttb::Downsample(in, 500, out, 50);