Skip to content
Pipelining framework for python. Designed for data analysis but can be useful to other applications
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

A parallel pipelining framework for Python. Developers can create nodes and chain them together to create pipelines.

Classes that extend Node must implement run method that will be called whenever new data is available.

Table of Contents


pip install pypiper

Example Usage

from pyPiper import Node, Pipeline

class Generate(Node):
    def setup(self, size):
        self.size = size
        self.pos = 0

    def run(self, data):
        if self.pos < self.size:
            self.pos = self.pos + 1

class Square(Node):
    def run(self, data):

pipeline = Pipeline(Generate("gen", size=10) | Square("square"))

Nodes can also specify a batch size that dictates how much data should be pushed to the node. For example, building on the previous example. In this case batch_size is specified in the nodes setup method. Alternatively, it can be set when creating the node (ex. Printer("print", batch_size=5))

class Printer(Node):
    def setup(self):
        self.batch_size = Node.BATCH_SIZE_ALL

    def run(self, data):

pipeline = Pipeline(Generate("gen", size=10) | Square("square") | Printer("print"))

Parallel Execution

To process pipelines in parallel, pass n_threads > 1 when creating the pipeline. Parallel execution is done using multiprocessing and is well suited to CPU intensive tasks such as audio processing and feature extraction. For example:

class Generate(Node):
    def setup(self, size):
        self.pos = 0

    def run(self, data):
        if self.pos < self.size:
            self.pos = self.pos + 1

pipeline = Pipeline(Generate("gen", size=10) | Square("square") | Printer("print"), n_threads=2)

Stream Names

You can also name input and output streams. For example:

gen = EvenOddGenerate("gen", size=20, out_streams=["even", "odd"])
double = Double("double", out_streams="num", in_streams="even")
square = Square("square", out_streams="num", in_streams="odd")

printer1 = Printer("p1", in_streams="num", batch_size=Node.BATCH_SIZE_ALL)
printer2 = Printer("p2", in_streams="num", batch_size=Node.BATCH_SIZE_ALL)

p = Pipeline(gen | [double | printer1, square | printer2], quiet=False)

EvenOddGenerate generates a pair of numbers. using the out_streams parameter, we name the first number even and second number odd. When initializing the double and square nodes, we tell double to take the even number and square to take the odd number.

If multiple output streams are passed into a node, by default, they will be come into the node as a list. For example,

gen = EvenOddGenerate("gen", size=10, out_streams=["even", "odd"])
printer = Printer("p1", batch_size=1)

p = Pipeline(gen | printer, quiet=False)

Will output


However, you can split the streams by specifying their names in the input streams in_streams parameter. So,

gen = EvenOddGenerate("gen", size=20, out_streams=["even", "odd"])

printer = Printer("p1", in_streams=["even", "odd"], batch_size=1)

p = Pipeline(gen | printer, quiet=False)

Will generate:


Progress Updates

When calling, you can provide a callback function for progress updates. Whenever the pipelines makes progress, it calls this function with the number of items that have been processed so far and the total number of items that need to be processed. For example, if you were using a tqdm progress bar, you could use the following code:

from tqdm import tqdm

class TqdmUpdate(tqdm):
    def update(self, done, total_size=None):
        if total_size is not None:
   = total_size
        self.n = done

if __name__ == '__main__':
    gen = Generate("gen", size=10)
    double = Double("double")
    sleeper = Sleep("sleep")

    p = Pipeline(gen | [double, sleeper], n_threads=4, quiet=True)
    with TqdmUpdate(desc="Progress") as pbar:

Projects Using PyPiper

  • COVFEFE: A feature extraction tool focusing on lexical, syntactic and pragmatic features from text and audio features from audio.
You can’t perform that action at this time.