Installation#

Quick install

In Python3.9 environment, in a virtualenv:

pip install -U pip wheel
pip install "dnikit[complete]"

The preceding command will install the main DNIKit package, TensorFlow2 and PyTorch compatibility, and requirements to run the notebook examples.

Note: For dev installation or to install from a specific branch, refer to the Contributor’s Guide.

Python Support#

DNIKit currently supports Python version 3.7 or greater for macOS or Linux. Python 3.9 is recommended. Note: to run TensorFlow 1, install Python 3.7.

To install Python version 3.9 (recommended):

MacOS: The Python installer package can be downloaded from the Python.org website. During installation, deselecting GUI applications, UNIX command-line tools, and Python documentation will reduce the size of what is installed.

Ubuntu:

sudo apt install -y python3.9-dev python3.9-venv python3.9-tk
sudo apt-get install -y libsm6 libxext6 libxrender-dev libgl1-mesa-glx

Virtualenv Creation#

It’s recommended to use a virtual environment to manage all dependencies:

python3.9 -m venv .venv39
source .venv37/bin/activate

And update pip and wheel:

pip install --upgrade pip wheel

Installation with pip#

The base DNIKit is installed with pip as follows:

pip install dnikit

DNIKit has additional installation options, which are installed in square brackets, using quotes:

pip install "dnikit[dataset-report,tensorflow,...]"

Here are the options currently available:

Module	Description
dnikit	Always installed, base DNIKit, with Familiarity, PFA, INA, and DimensionReduction.
-> [notebook]	Installs dependencies to run and visualize the jupyter notebook tutorials, including jupyter, matplotlib, pandas, …
-> [image]	Installs opencv (headless) and Pillow to enable image processing capabilities.
-> [dimreduction]	Installs umap_learn and pacmap for dimensionality reduction.
-> [dataset-report]	Installs all requirements to run the Dataset Report.
-> [tensorflow]	Installs dnikit_tensorflow and TF2 to load & run TF models within DNIKit.
-> [tensorflow1]	Installs dnikit_tensorflow and TF1 to load & run TF models within DNIKit. Must have Python <=3.7 due to TF 1.
-> [tensorflow1-gpu]	Same as preceding row, but with TensorFlow GPU. Must have Python <=3.7 due to TF 1 constraints.
-> [torch]	Installs `dnikit_pytorch`: convert between PyTorch Dataset and DNIKit Producer.
-> [complete]	Installs `notebook`, `image`, `dimreduction`, `dataset-report`, `tensorflow`, & `torch` options.

Running the Jupyter Notebooks Examples#

First, install the notebook dependencies:

pip install "dnikit[notebook]"

Next, download the DNIKit notebooks directly or use them via cloning the dnikit repository.

Finally, launch jupyter to open the notebooks:

jupyter notebook

Installation for developers#

Check out the Development Installation page to install DNIKit for development.

Issues with installation?#

Please file an issue in the GitHub repository.