Skip to content

Latest commit

 

History

History
87 lines (61 loc) · 4.17 KB

README.md

File metadata and controls

87 lines (61 loc) · 4.17 KB
Raw

Jacinle

Jacinle is a personal python toolbox. It contains a range of utility functions for python development, including project configuration, file IO, image processing, inter-process communication, etc.

[Website] [Examples] [Jacinle References] [JacLearn References] [JacTorch References]

Installation

Clone the Jacinle package (be sure to clone all submodules), and add the bin path to your PATH environment.

git clone https://github.com/vacancy/Jacinle --recursive
export PATH=<path_to_jacinle>/bin:$PATH

Optionally, you may need to install third-party packages specified in requirements.txt

Command Line

  1. jac-run xxx.py

    Jacinle comes with a command line to replace the python command: jac-run. In short, this command will automatically add the Jacinle packages into PYTHONPATH, as well as adding a few vendor Python packages into PYTHONPATH (for example, JacMLDash). Using this command to replace python xxx.py is the best practice to manage dependencies.

    Furthremore, this command also supports a configuration file specific to projects. The command will search for a configuration file named jacinle.yml in the current working directory and its parent directories. This file specifies additional environmental variables to add, for example.

    project_root: true  # tell the script that the folder containing this file is the root of a project. The directory will be added to PYTHONPATH.
system:
    envs:
        CUDA_HOME: /usr/local/cuda-10.0  # set needed environment variables here.
path:
    bin:  # will be prepended to $PATH
        /usr/local/bin
    python:  # will be prepended to $PYTHONPATH
        /Users/jiayuanm/opt/my_python_lib
vendors:  # load additional Python packages (root paths will be added to PYTHONPATH)
    pybullet_tools:
        root: /Users/jiayuanm/opt/pybullet/utils
    alfred:
        root: /Users/jiayuanm/opt/alfred

  2. jac-crun <gpu_ids> xxx.py

    The same as jac-run, but takes an additional argument, which is a comma-separated list of gpu ids, following the convension of CUDA_VISIBLE_DEVICES.

  3. jac-debug xxx.py

    The same as jac-run, but sets the environment variable JAC_DEBUG=1 before running the command. By default, in the debug mode, an ipdb interface will be started when an exception is raised.

  4. jac-cdebug <gpu_ids> xxx.py

    The combined jac-debug and jac-crun.

  5. jac-update

    Update the Jacinle package (and all dependencies inside vendors/).

  6. jac-inspect-file xxx.json yyy.pkl

    Start an IPython interface and loads all files in the argument list. The content of the files can be accessed via f1, f2, ...

Python Libraries

Jacinle contains a collection of useful packages. Here is a list of commonly used packages, with links to the documentation.

  • jacinle.*: frequently used utility functions, such as jacinle.JacArgumentParser, jacinle.TQDMPool, jacinle.get_logger, jacinle.cond_with, etc.
  • jacinle.io.*: IO functions. Two of the mostly used ones are: jacinle.io.load(filename) and jacinle.io.dump(filename, obj)
  • jacinle.random.*: almost the same as numpy.random.*, but with a few additional utility functions and RNG state management functions.
  • jacinle.web.*: the old jacweb package, which is a customized wrapper around the tornado web server.
  • jaclearn.*: machine learning modules.
  • jactorch.*: a collection of PyTorch functions in addition to the torch.* functions.