kojo 0.28.1

A library to transform data through a pipeline

kojo is a tool to help to transform, validate, and filter data. It has two main components:

1. Process allows to filter and map data items in a chainable interface for better readability.
2. Item allows to add log entries and meta data to each data item individually.

The rest of kojo is “nice to have”.

Process

Motivation

If we want to transform a number of dictionary items we can e.g. use a generator comprehension.

def onedigit():
    for i in range(10):
        yield i

iterator = onedigit()
iterator = (i * i for i in iterator)
print(list(iterator))
# > [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]

Let's pretend we need all square numbers, that do not contain the digit 6:

def onedigit():
    for i in range(10):
        yield i

iterator = onedigit()
iterator = (i * i for i in iterator)
iterator = (i for i in iterator if "6" not in str(i))
print(list(iterator))
# > [0, 1, 4, 9, 25, 49, 81]

This approach works fine when there are only some easy steps. When migrating data however there are often a lot of steps and maybe even several input files.

Let's presume, we have two inputs and the transformations are too complex to use comprehensions. We use a transformation function and to keep the code dry and call range directly:

def square(i):
    return i * i

def no6(i):
    return "6" not in str(i)

def transform(iterator):
    iterator = (square(i) for i in iterator)
    iterator = (i for i in iterator if no6(i))
    return iterator

iterator = transform(range(10))
print(list(iterator))
# > [0, 1, 4, 9, 25, 49, 81]

iterator = transform(range(10, 20))
print(list(iterator))
# > [100, 121, 144, 225, 289, 324]

This code has two issues:

1. The transformation process needs data as input and is therefore not completely decoupled
2. Readability is bad when there are a lot of transformation steps

Basic usage

from kojo import Process, apply

def square(i):
    return i * i

def no6(i):
    return "6" not in str(i)

process = Process().map(square).filter(no6)

iterator = apply(range(10), process)
print(list(iterator))
# all square numbers of 0-9 that do not contain the digit 6
# > [0, 1, 4, 9, 25, 49, 81]

iterator = apply(range(10, 20), process)
print(list(iterator))
# all square numbers of 10-20 that do not contain the digit 6
# > [100, 121, 144, 225, 289, 324]

def digit_sum(i):
    return sum(int(digit) for digit in str(i))

post_process = Process().map(digit_sum)

iterator = apply(range(10, 20), process, post_process)
print(list(iterator))
# the digit sum of all square numbers of 10-20 that do not contain the digit 6
# > [1, 4, 9, 9, 19, 9]

The Process class is used to define a new process. It has two methods: filter and map. These methods can be chained.

The apply function takes an Iterable and 1…n processes as arguments. It returns an Iterator that generates a list of transformed and mapped items.

Item

Motivation

When migrating data it happens, that input data has issues. To handle these issues one need to know, which item has which issues.

Instead of using logging that contains a reference to the item the Item class follows another approach by having an own log. Item extends Pythons dict class and provides a log system, which is independent from the dictionary.

Basic usage

from kojo import Item

item = Item(meaningOfLife=54)
print(item)
# > {'meaningOfLife': 54}

if item["meaningOfLife"] != 42:
    item.log.warning("That's not the meaning of life")

print(item)
# > {'meaningOfLife': 54}

print(item.log.level)
# > 30 # (logging.WARNING)

print(len(item.log))
# > 1

print(item.log[0].message)
# > That's not the meaning of life

item.log.info("Please review")

print(len(item.log))
# > 2