Skip to content


Repository files navigation


Documentation Status

ulab is a numpy-like array manipulation library for micropython and CircuitPython. The module is written in C, defines compact containers (ndarrays) for numerical data of one to four dimensions, and is fast. The library is a software-only standard micropython user module, i.e., it has no hardware dependencies, and can be compiled for any platform. 8-, and 16-bit signed and unsigned integer dtypes, as well as float, and, optionally, complex are supported. The float implementation of micropython (32-bit float, or 64-bit double) is automatically detected and handled.

  1. Supported functions and methods
    1. ndarray methods
    2. numpy and scipy functions
    3. ulab utilities
    4. user module
  2. Usage
  3. Finding help
  4. Benchmarks
  5. Firmware
    1. Customising the firmware
    2. Platforms including ulab
    3. Compiling
      1. UNIX
      2. STM-based boards
      3. ESP32-based boards
      4. RP2-based boards
      5. Compiling for CircuitPython
  6. Issues, contributing, and testing
    1. Testing

Supported functions and methods

ndarray methods

ulab implements numpy's ndarray with the ==, !=, <, <=, >, >=, +, -, /, *, **, +=, -=, *=, /=, **= binary operators, and the len, ~, -, +, abs unary operators that operate element-wise. Type-aware ndarrays can be initialised from any micropython iterable, lists of iterables via the array constructor, or by means of the arange, concatenate, diag, eye, frombuffer, full, linspace, logspace, ones, or zeros functions.

ndarrays can be sliced, and iterated on, and have a number of their own methods, and properties, such as flatten(), itemsize, reshape(), shape, size, strides, tobytes(), tolist(), and transpose() and T. If the firmware is compiled with complex support, the imag, and real properties are automatically included.

numpy and scipy functions

In addition, ulab includes universal functions, many numpy functions, and functions from the numpy.fft, numpy.linalg, numpy.random, scipy.linalg, scipy.optimize, scipy.signal, and scipy.special modules. A complete list of available routines can be found under micropython-ulab.

ulab utilities

The utils module contains functions for interfacing with peripheral devices supporting the buffer protocol. These functions do not have an obvious numpy equivalent, but share a similar programming interface, and allow direct data input-output between numerical arrays and hardware components.

user module

User-defined functions operating on numerical data can easily be added via the user module. This allows for transparent extensions, without having to change anything in the core. Hints as to how to work with ndarrays at the C level can be found in the programming manual.


ulab sports a numpy/scipy-compatible interface, which makes porting of CPython code straightforward. The following snippet should run equally well in micropython, or on a PC.

    from ulab import numpy
    from ulab import scipy
except ImportError:
    import numpy
    import scipy.special

x = numpy.array([1, 2, 3])

Finding help

Documentation can be found on readthedocs under micropython-ulab, as well as at circuitpython-ulab. A number of practical examples are listed in Jeff Epler's excellent circuitpython-ulab overview. The tricks chapter of the user manual discusses methods by which RAM and speed can be leveraged in particular numerical problems.


Representative numbers on performance can be found under ulab samples.


Pre-built, and up-to-date firmware files for select platforms can be downloaded from micropython-builder.

Customising the firmware

If flash space is a concern, unnecessary functions can be excluded from the compiled firmware with pre-processor switches. In addition, ulab also has options for trading execution speed for firmware size. A thorough discussion on how the firmware can be customised can be found in the corresponding section of the user manual.

Platforms including ulab

ulab is also included in the following compiled micropython variants and derivatives:

  1. CircuitPython for SAMD51 and nRF microcontrollers
  2. MicroPython for K210
  3. MaixPy
  4. OpenMV
  5. pimoroni-pico
  6. pycom


If you want to try the latest version of ulab on micropython or one of its forks, the firmware can be compiled from the source by following these steps:

UNIX port

Simply clone the ulab repository with

git clone ulab

and then run

./ [matrix.dims] # Dimensions is 2 by default

This command will clone micropython, and build the unix port automatically, as well as run the test scripts. If you want an interactive unix session, you can launch it in


STM-based boards

First, you have to clone the micropython repository by running

git clone

on the command line. This will create a new repository with the name micropython. Staying there, clone the ulab repository with

git clone ulab

If you don't have the cross-compiler installed, your might want to do that now, for instance on Linux by executing

sudo apt-get install gcc-arm-none-eabi

If this step was successful, you can try to run the make command in the port's directory as

make BOARD=PYBV11 USER_C_MODULES=../../../ulab all

which will prepare the firmware for pyboard.v.11. Similarly,

make BOARD=PYBD_SF6 USER_C_MODULES=../../../ulab all

will compile for the SF6 member of the PYBD series. If your target is unix, you don't need to specify the BOARD parameter.

Provided that you managed to compile the firmware, you would upload that by running either

dfu-util --alt 0 -D firmware.dfu


python -u firmware.dfu

In case you got stuck somewhere in the process, a bit more detailed instructions can be found under, and

ESP32-based boards

ulab can be tested on the ESP32 in wokwi's micropython emulator without having to compile the C code. This utility also offers the possibility to save and share your micropython code.

Firmware for Espressif hardware can be built in two different ways, which are discussed in the next two paragraphs. A solution for issues with the firmware size is outlined in the last paragraph of this section.

Compiling with cmake

Beginning with version 1.15, micropython switched to cmake on the ESP32 port. If your operating system supports CMake > 3.12, you can either simply download, and run the single build script, or follow the step in this section. Otherwise, you should skip to the next one, where the old, make-based approach is discussed.

In case you encounter difficulties during the build process, you can consult the (general instructions for the ESP32)[].

First, clone the ulab, the micropython, as well as the espressif repositories:

export BUILD_DIR=$(pwd)

git clone ulab
git clone

cd $BUILD_DIR/micropython/

git clone -b v4.0.2 --recursive

Also later releases of esp-idf are possible (e.g. v4.2.1).

Then install the ESP-IDF tools:

cd esp-idf
. ./

Next, build the micropython cross-compiler, and the ESP sub-modules:

cd $BUILD_DIR/micropython/mpy-cross
cd $BUILD_DIR/micropython/ports/esp32
make submodules

At this point, all requirements are installed and built. We can now compile the firmware with ulab. In $BUILD_DIR/micropython/ports/esp32 create a makefile with the following content:

USER_C_MODULES = $(BUILD_DIR)/ulab/code/micropython.cmake

include Makefile

You specify with the BOARD variable, what you want to compile for, a generic board, or TINYPICO (for micropython version >1.1.5, use UM_TINYPICO), etc. Still in $BUILD_DIR/micropython/ports/esp32, you can now run make.

Compiling with make

If your operating system does not support a recent enough version of CMake, you have to stay with micropython version 1.14. The firmware can be compiled either by downloading and running the build script, or following the steps below:

First, clone ulab with

git clone ulab

and then, in the same directory, micropython

git clone

At this point, you should have ulab, and micropython side by side.

With version 1.14, micropython switched to cmake on the ESP32 port, thus breaking compatibility with user modules. ulab can, however, still be compiled with version 1.14. You can check out a particular version by pinning the release tag as

cd ./micropython/
git checkout tags/v1.14

Next, update the submodules,

git submodule update --init
cd ./mpy-cross && make # build cross-compiler (required)

and find the ESP commit hash

cd ./micropython/ports/esp32
make ESPIDF= # will display supported ESP-IDF commit hashes
# output should look like: """
# ...
# Supported git hash (v3.3): 9e70825d1e1cbf7988cf36981774300066580ea7
# Supported git hash (v4.0) (experimental): 4c81978a3e2220674a432a588292a4c860eef27b

Choose an ESPIDF version from one of the options printed by the previous command:


In the micropython directory, create a new directory with

mkdir esp32

Your micropython directory should now look like

ACKNOWLEDGEMENTS  esp32     lib      mpy-cross  docs             examples  LICENSE  ports      tests    drivers          extmod    logo     py         tools

In ./micropython/esp32, download the software development kit with

git clone esp-idf
cd ./esp-idf
git checkout $ESPIDF_VER
git submodule update --init --recursive # get idf submodules
pip install -r ./requirements.txt # install python reqs

Next, still staying in ./micropython/eps32/esd-idf/, install the ESP32 compiler. If using an ESP-IDF version >= 4.x (chosen by $ESPIDF_VER above), this can be done by running . $BUILD_DIR/esp-idf/ Otherwise, for version 3.x, run the following commands in in .micropython/esp32/esp-idf:

# for 64 bit linux
curl | tar xvz

# for 32 bit
# curl | tar xvz

# don't worry about adding to path; we'll specify that later

# also, see for more info

Finally, build the firmware:

cd ./micropython/ports/esp32
# temporarily add esp32 compiler to path
export PATH=../../esp32/esp-idf/xtensa-esp32-elf/bin:$PATH
export ESPIDF=../../esp32/esp-idf # req'd by Makefile
export BOARD=GENERIC # options are dirs in ./boards
export USER_C_MODULES=../../../ulab # include ulab in firmware

make submodules & make all

If it compiles without error, you can plug in your ESP32 via USB and then flash it with:

make erase && make deploy

What to do, if the firmware is too large?

When selecting BOARD=TINYPICO, the firmware is built but fails to deploy, because it is too large for the standard partitions. We can rectify the problem by creating a new partition table. In order to do so, in $BUILD_DIR/micropython/ports/esp32/, copy the following 8 lines to a file named partitions_ulab.cvs:

# Notes: the offset of the partition table itself is set in
# $ESPIDF/components/partition_table/Kconfig.projbuild and the
# offset of the factory/ota_0 partition is set in
# Name,   Type, SubType, Offset,  Size, Flags
nvs,      data, nvs,     0x9000,  0x6000,
phy_init, data, phy,     0xf000,  0x1000,
factory,  app,  factory, 0x10000, 0x200000,
vfs,      data, fat,     0x220000, 0x180000,

This expands the factory partition by 128 kB, and reduces the size of vfs by the same amount. Having defined the new partition table, we should extend sdkconfig.board by adding the following two lines:


This file can be found in $BUILD_DIR/micropython/ports/esp32/boards/TINYPICO/. Finally, run make clean, and make. The new firmware contains the modified partition table, and should fit on the microcontroller.

RP2-based boards

RP2 firmware can be compiled either by downloading and running the single build script/build script for Pico W, or executing the commands below.

First, clone micropython:

git clone

Then, setup the required submodules:

cd micropython
git submodule update --init lib/tinyusb
git submodule update --init lib/pico-sdk
cd lib/pico-sdk
git submodule update --init lib/tinyusb

You'll also need to compile mpy-cross:

cd ../../mpy-cross

That's all you need to do for the micropython repository. Now, let us clone ulab (in a directory outside the micropython repository):

cd ../../
git clone ulab

With this setup, we can now build the firmware. Back in the micropython repository, use these commands:

cd ports/rp2
make USER_C_MODULES=/path/to/ulab/code/micropython.cmake

If micropython and ulab were in the same folder on the computer, you can set USER_C_MODULES=../../../ulab/code/micropython.cmake. The compiled firmware will be placed in micropython/ports/rp2/build.

Compiling for CircuitPython

Adafruit Industries always include a relatively recent version of ulab in their nightly builds. However, if you really need the bleeding edge, you can easily compile the firmware from the source. Simply clone circuitpython, and move the commit pointer to the latest version of ulab (ulab will automatically be cloned with circuitpython):

git clone

cd circuitpyton/extmod/ulab

# update ulab here
git checkout master
git pull

You might have to check, whether the CIRCUITPY_ULAB variable is set to 1 for the port that you want to compile for. You find this piece of information in the make fragment:


After this, you would run make with the single BOARD argument, e.g.:

make BOARD=mini_sam_m4

Issues, contributing, and testing

If you find a problem with the code, please, raise an issue! An issue should address a single problem, and should contain a minimal code snippet that demonstrates the difference from the expected behaviour. Reducing a problem to the bare minimum significantly increases the chances of a quick fix.

Feature requests (porting a particular function from numpy or scipy) should also be posted at ulab issue.

Contributions of any kind are always welcome. If you feel like adding to the code, you can simply issue a pull request. If you do so, please, try to adhere to micropython's coding conventions.

However, you can also contribute to the documentation (preferably via the jupyter notebooks, or improve the tests.


If you decide to lend a hand with testing, here are the steps:

  1. Write a test script that checks a particular function, or a set of related functions!
  2. Drop this script in one of the folders in ulab tests!
  3. Run the ./ script in the root directory of ulab! This will clone the latest micropython, compile the firmware for unix, execute all scripts in the ulab/tests, and compare the results to those in the expected results files, which are also in ulab/tests, and have an extension .exp. In case you have a new snippet, i.e., you have no expected results file, or if the results differ from those in the expected file, a new expected file will be generated in the root directory. You should inspect the contents of this file, and if they are satisfactory, then the file can be moved to the ulab/tests folder, alongside your snippet.