sspipe 0.1.10

Simple Smart Pipe Operator

Simple Smart Pipe

SSPipe is a python productivity-tool for rapid data manipulation in python.

It helps you break up any complicated expression into a sequence of simple transformations, increasing human-readability and decreasing the need for matching parentheses!

If you're familiar with | operator of Unix, or %>% operator of R's magrittr, or DataFrame.pipe method of pandas library, sspipe provides the same functionality for any object in python.

Installation and Usage

Install sspipe using pip:

pip install --upgrade sspipe

Then import it in your scripts.

from sspipe import p, px

The whole functionality of this library is exposed by two objects p (as a wrapper for functions to be called on the piped object) and px (as a placeholder for piped object).

Examples

Description	Python expression using p and px	Equivalent python code
Simple function call	"hello world!" | p(print)	X = "hello world!" print(X)
Function call with extra args	"hello" | p(print, "world", end='!')	X = "hello" print(X, "world", end='!')
Explicitly positioning piped argument with px placeholder	"world" | p(print, "hello", px, "!")	X = "world" print("hello", X, "!")
Chaining pipes	5 | px + 2 | px ** 5 + px | p(print)	X = 5 X = X + 2 X = X ** 5 + X print(X)
Tailored behavior for builtin map and filter	( range(5) | p(filter, px % 2 == 0) | p(map, px + 10) | p(list) | p(print) )	X = range(5) X = filter((lambda x:x%2==0),X) X = map((lambda x: x + 10), X) print(list(X))
NumPy expressions	range(10) | np.sin(px)+1 | p(plt.plot)	X = range(10) X = np.sin(X) + 1 plt.plot(X)
Pandas support	people_df | px.loc[px.age > 10, 'name']	X = people_df X.loc[X.age > 10, 'name']
Assignment	people_df['name'] |= px.str.upper()	X = people_df['name'] X = X.str.upper() people_df['name'] = X
Builtin Data Structures	2 | p({px-1: p([px, p((px+1, 4))])})	X = 2 X = {X-1: [X, (X+1, 4)]}

Introduction

Suppose we want to generate a dict, mapping names of 5 biggest files in current directory to their size in bytes, like below:

{'README.md': 3732, 'setup.py': 1642, '.gitignore': 1203, 'LICENSE': 1068, 'deploy.sh': 89}

One approach is to use os.listdir() to list files and directories in current working directory, filter those which are file, map each to a tuple of (name, size), sort them by size, take first 5 items, make adict and print it.

Although it is not a good practice to write the whole script in single expression without introducing intermediary variables, it is an exaggerated example, doing it in a single expression for demonstration purpose:

import os

print(
    dict(
        sorted(
            map(
                lambda x: [x, os.path.getsize(x)],
                filter(os.path.isfile, os.listdir('.'))
            ), key=lambda x: x[1], reverse=True
        )[:5]
    )
)

Using sspipe's p operator, the same single expression can be written in a more human-readable flow of sequential transformations:

import os
from sspipe import p

(
    os.listdir('.')
    | p(filter, os.path.isfile)
    | p(map, lambda x: [x, os.path.getsize(x)])
    | p(sorted, key=lambda x: x[1], reverse=True)[:5]
    | p(dict)
    | p(print)
)

As you see, the expression is decomposed into a sequence starting with initial data, os.list('.'), followed by multiple | p(...) stages.

Each | p(...) stage describes a transformation that is applied to to left-hand-side of |.

First argument of p() defines the function that is applied on data. For example, x | p(f1) | p(f2) | p(f3) is equivalent to f3(f2(f1(x))).

Rest of arguments of p() are passed to the transforming function of each stage. For example, x | p(f1, y) | p(f2, k=z) is equivalent to f2(f1(x, y), k=z)

Advanced Guide

The px helper

TODO: explain.

• px is implemented by: px = p(lambda x: x)
• px is similar to, but not same as, magrittr's dot(.) placeholder
  • x | p(f, px+1, y, px+2) is equivalent to f(x+1, y, x+2)
• A+1 | f(px, px[2](px.y)) is equivalent to f(A+1, (A+1)[2]((A+1).y)
• px can be used to prevent adding parentheses
  • x+1 | px * 2 | np.log(px)+3 is equivalent to: np.log((x+1) * 2) + 3

Integration with Numpy, Pandas, Pytorch

TODO: explain.

• p and px are compatible with Numpy, Pandas, Pytorch.
• [1,2] | p(pd.Series) | px[px ** 2 < np.log(px) + 1] is equivalent to x=pd.Series([1, 2]); x[x**2 < np.log(x)+1]

Compatibility with JulienPalard/Pipe

This library is inspired by, and depends on, the intelligent and concise work of JulienPalard/Pipe. If you want a single pipe.py script or a lightweight library that implements core functionality and logic of SSPipe, Pipe is perfect.

SSPipe is focused on facilitating usage of pipes, by integration with popular libraries and introducing px concept and overriding python operators to make pipe a first-class citizen.

Every existing pipe implemented by JulienPalard/Pipe library is accessible through p.<original_name> and is compatible with SSPipe. SSPipe does not implement any specific pipe function and delegates implementation and naming of pipe functions to JulienPalard/Pipe.

For example, JulienPalard/Pipe's example for solving "Find the sum of all the even-valued terms in Fibonacci which do not exceed four million." can be re-written using sspipe:

def fib():
    a, b = 0, 1
    while True:
        yield a
        a, b = b, a + b

from sspipe import p, px

euler2 = (fib() | p.where(lambda x: x % 2 == 0)
                | p.take_while(lambda x: x < 4000000)
                | p.add())

You can also pass px shorthands to JulienPalard/Pipe API:

euler2 = (fib() | p.where(px % 2 == 0)
                | p.take_while(px < 4000000)
                | p.add())

Internals

TODO: explain.

• p is a class that overrides __ror__ (|) operator to apply the function to operand.