Skip to main content

Some tools for Luigi to cut down the length of your pipelines and work in interactive environments such as Jupyter notebooks.

Project description

pipecutter

PyPI version Python versions build status coverage

pipecutter provides a few tools for luigi such that it works better with data science libraries and environments such as pandas, scikit-learn, and Jupyter notebooks.

Installation

pip install pipecutter

Python 3.6+ is required. pipecutter follows semantic versioning.

Usage

pipecutter currently provides

  • a more convenient way to run and debug luigi tasks in interactive environments such as Jupyter notebooks
  • some luigi targets for saving pandas dataframes to parquet, scikit-learn models with joblib, ...

Debug in an interactive environment

With luigi, you can already run tasks in a Python script/Jupyter notebook/Python console by using the luigi.build function (probably with local_scheduler=True as arugment). However, if the tasks throws an exception this will be caught by luigi and you are not able to drop into a post mortem debugging session. pipecutter.run is a light wrapper around luigi.build which disables this exception handling.

In [1]: import luigi
In [2]: import pipecutter

In [3]: class TaskWhichFails(luigi.Task):
   ...:     def run(self):
   ...:         raise Exception("Something is wrong")

# Traceback below is shortened for readability
In [4]: pipecutter.run(TaskWhichFails())
---------------------------------------------------------------------------
Exception                                 Traceback (most recent call last)
<ipython-input-5-a970d52d810a> in <module>
----> 1 pipecutter.run(TaskWhichFails())

...

<ipython-input-3-4e27674090fa> in run(self)
      1 class TaskWhichFails(luigi.Task):
      2     def run(self):
----> 3         raise Exception

Exception: Something is wrong

# Drop straight into the debugger
In [5]: %debug
> <ipython-input-6-e7528a27d82e>(3)run()
      1 class TaskWhichFails(luigi.Task):
      2     def run(self):
----> 3         raise Exception
      4
ipdb>

This should reduce the barrier for already using luigi tasks while developing a model and thereby making it easier to move into production later on.

Additionally, you can print the dependencies of tasks with pipecutter.print_tree (wrapper around luigi.tools.deps_tree.print_tree).

Targets

In pipecutter.targets you find a few targets which build on luigi's LocalTarget but additionally have a load and a dump method. A convenient way to name the targets is hereby to use the task_id in the name, which is unique with respect to the task name and its passed in parameters.

import luigi
import pipecutter
from pipecutter.targets import JoblibTarget
from sklearn.ensemble import RandomForestClassifier

class TrainModel(luigi.Task):
    n_estimators = luigi.Parameter()

    def output(self):
        return JoblibTarget(self.task_id + ".pkl")

    def run(self):
        model = RandomForestClassifier(n_estimators=self.n_estimators)
        self.output().dump(model)

pipecutter.run(TrainModel(n_estimators=100))
# -> Produces a file called TrainModel_100_0b0ec0cdea.pkl

Other interesting libraries

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for pipecutter, version 1.1.0
Filename, size File type Python version Upload date Hashes
Filename, size pipecutter-1.1.0-py3-none-any.whl (5.4 kB) File type Wheel Python version py3 Upload date Hashes View hashes
Filename, size pipecutter-1.1.0.tar.gz (4.3 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page