DEVELOPMENT.md

Notes for TensorFlow.js-Converter Developers

Python Development

There are some Python libraries, binary and tests in the python/ directory.

It is recommended to do your Python development and testing in a virtualenv or pipenv.

As a prerequisite, install the following dependencies for python testing

cd python
pip install -r requirements.txt

For Python linter, install pylint, e.g.,

• apt-get install -y pylint

To run the Python linter:

cd python
pylint tensorflowjs

To run the python unit tests, there are two options. You can choose the one that you prefer.

1. Run the tests using the run-python-tests.sh script:

   cd python
   ./run-python-tests.sh

2. Run the tests using Bazel. See bazel installation guide here. Once bazel is installed, do:

   cd python
   bazel test tensorflowjs/...

Be sure to run the tests under both Python 2 and Python 3.

Building and testing the tensorflowjs pip package

cd python

# You need to specify a folder where the pip wheel file will be stored, e.g.,
./build-pip-package.sh /tmp/my_tensorflowjs_pip

# If the script succeeds, you can use `pip install` to install the pip package:

pip install --force-reinstall \
  /tmp/my_tensorflowjs_pip/tensorflowjs-0.0.1-py2-none-any.whl

build-pip-package.sh provides a flag (--test) with which you can run a test-on-install after building the pip package. Make sure you are using a virutalenv or pipenv to avoid changing your base environmnet.

./build-pip-package.sh --test /tmp/my_tensorflowjs_pip