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]
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
-
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 intoPYTHONPATH
, as well as adding a few vendor Python packages intoPYTHONPATH
(for example, JacMLDash). Using this command to replacepython 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
-
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 ofCUDA_VISIBLE_DEVICES
. -
jac-debug xxx.py
The same as
jac-run
, but sets the environment variableJAC_DEBUG=1
before running the command. By default, in the debug mode, anipdb
interface will be started when an exception is raised. -
jac-cdebug <gpu_ids> xxx.py
The combined
jac-debug
andjac-crun
. -
jac-update
Update the Jacinle package (and all dependencies inside
vendors/
). -
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
, ...
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 asjacinle.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)
andjacinle.io.dump(filename, obj)
jacinle.random.*
: almost the same asnumpy.random.*
, but with a few additional utility functions and RNG state management functions.jacinle.web.*
: the oldjacweb
package, which is a customized wrapper around the tornado web server.jaclearn.*
: machine learning modules.jactorch.*
: a collection of PyTorch functions in addition to thetorch.*
functions.