PyFunctional

PyFunctional

Features

PyFunctional makes creating data pipelines easy by using chained functional operators. Here are a few examples of what it can do:

• Chained operators: seq(1, 2, 3).map(lambda x: x * 2).reduce(lambda x, y: x + y)
• Expressive and feature complete API
• Read and write text, csv, json, jsonl, sqlite, gzip, bz2, and lzma/xz files
• Parallelize "embarrassingly parallel" operations like map easily
• Complete documentation, rigorous unit test suite, 100% test coverage, and CI which provide robustness

PyFunctional's API takes inspiration from Scala collections, Apache Spark RDDs, and Microsoft LINQ.

Table of Contents

1. Installation
2. Examples
   1. Simple Example
   2. Aggregates and Joins
   3. Reading and Writing SQLite3
   4. Data Interchange with Pandas
3. Writing to Files
4. Parallel Execution
5. Github Shortform Documentation
   1. Streams, Transformations, and Actions
   2. Streams API
   3. Transformations and Actions APIs
   4. Lazy Execution
6. Contributing and Bug Fixes
7. Changelog

Installation

PyFunctional is available on pypi and can be installed by running:

# Install from command line
$ pip install pyfunctional

Then in python run: from functional import seq

Examples

PyFunctional is useful for many tasks, and can natively open several common file types. Here are a few examples of what you can do.

Simple Example

from functional import seq

seq(1, 2, 3, 4)\
    .map(lambda x: x * 2)\
    .filter(lambda x: x > 4)\
    .reduce(lambda x, y: x + y)
# 14

# or if you don't like backslash continuation
(seq(1, 2, 3, 4)
    .map(lambda x: x * 2)
    .filter(lambda x: x > 4)
    .reduce(lambda x, y: x + y)
)
# 14

Streams, Transformations and Actions

PyFunctional has three types of functions:

1. Streams: read data for use by the collections API.
2. Transformations: transform data from streams with functions such as map, flat_map, and filter
3. Actions: These cause a series of transformations to evaluate to a concrete value. to_list, reduce, and to_dict are examples of actions.

In the expression seq(1, 2, 3).map(lambda x: x * 2).reduce(lambda x, y: x + y), seq is the stream, map is the transformation, and reduce is the action.

Filtering a list of account transactions

…

Aggregates and Joins

The account transactions example could be done easily in pure python using list comprehensions. To show some of the things PyFunctional excels at, take a look at a couple of word count examples.

words = 'I dont want to believe I want to know'.split(' ')
seq(words).map(lambda word: (word, 1)).reduce_by_key(lambda x, y: x + y)
# [('dont', 1), ('I', 2), ('to', 2), ('know', 1), ('want', 2), ('believe', 1)]

In the next example we have chat logs formatted in json lines (jsonl) which contain messages and metadata. A typical jsonl file will have one valid json on each line of a file. Below are a few lines out of examples/chat_logs.jsonl.

{"message":"hello anyone there?","date":"10/09","user":"bob"}
{"message":"need some help with a program","date":"10/09","user":"bob"}
{"message":"sure thing. What do you need help with?","date":"10/09","user":"dave"}

from operator import add
import re
messages = seq.jsonl('examples/chat_logs.jsonl')

# Split words on space and normalize before doing word count
def extract_words(message):
    return re.sub('[^0-9a-z ]+', '', message.lower()).split(' ')

word_counts = messages\
    .map(lambda log: extract_words(log['message']))\
    .flatten().map(lambda word: (word, 1))\
    .reduce_by_key(add).order_by(lambda x: x[1])

Next, let's continue that example but introduce a json database of users from examples/users.json. In the previous example we showed how PyFunctional can do word counts, in the next example let's show how PyFunctional can join different data sources.

…

CSV, Aggregate Functions, and Set functions

In examples/camping_purchases.csv there is a list of camping purchases. Let's do some cost analysis and compare it to the required camping gear list stored in examples/gear_list.txt.

purchases = seq.csv('examples/camping_purchases.csv')
total_cost = purchases.select(lambda row: int(row[2])).sum()
# 1275

most_expensive_item = purchases.max_by(lambda row: int(row[2]))
# ['4', 'sleeping bag', ' 350']

purchased_list = purchases.select(lambda row: row[1])
gear_list = seq.open('examples/gear_list.txt').map(lambda row: row.strip())
missing_gear = gear_list.difference(purchased_list)
# ['water bottle','gas','toilet paper','lighter','spoons','sleeping pad',...]

In addition to the aggregate functions shown above (sum and max_by) there are many more. Similarly, there are several more set like functions in addition to difference.

Reading/Writing SQLite3

PyFunctional can read and write to SQLite3 database files. In the example below, users are read from examples/users.db which stores them as rows with columns id:Int and name:String.

db_path = 'examples/users.db'
users = seq.sqlite3(db_path, 'select * from user').to_list()
# [(1, 'Tom'), (2, 'Jack'), (3, 'Jane'), (4, 'Stephan')]]

sorted_users = seq.sqlite3(db_path, 'select * from user order by name').to_list()
# [(2, 'Jack'), (3, 'Jane'), (4, 'Stephan'), (1, 'Tom')]

Writing to a SQLite3 database is similarly easy

…

Writing to files

Just as PyFunctional can read from csv, json, jsonl, sqlite3, and text files, it can also write them. For complete API documentation see the collections API table or the official docs.

Compressed Files

PyFunctional will auto-detect files compressed with gzip, lzma/xz, and bz2. This is done by examining the first several bytes of the file to determine if it is compressed so therefore requires no code changes to work.

To write compressed files, every to_ function has a parameter compression which can be set to the default None for no compression, gzip or gz for gzip compression, lzma or xz for lzma compression, and bz2 for bz2 compression.

Parallel Execution

The only change required to enable parallelism is to import from functional import pseq instead of from functional import seq and use pseq where you would use seq. The following operations are run in parallel with more to be implemented in a future release:

• map/select
• filter/filter_not/where
• flat_map

Parallelization uses python multiprocessing and squashes chains of embarrassingly parallel operations to reduce overhead costs. For example, a sequence of maps and filters would be executed all at once rather than in multiple loops using multiprocessing.

Documentation

Shortform documentation is below and full documentation is at docs.pyfunctional.pedro.ai.

Streams API

All of PyFunctional streams can be accessed through the seq object. The primary way to create a stream is by calling seq with an iterable. The seq callable is smart and is able to accept multiple types of parameters as shown in the examples below.

# Passing a list
seq([1, 1, 2, 3]).to_set()
# [1, 2, 3]

# Passing direct arguments
seq(1, 1, 2, 3).map(lambda x: x).to_list()
# [1, 1, 2, 3]

# Passing a single value
seq(1).map(lambda x: -x).to_list()
# [-1]

seq also provides entry to other streams as attribute functions as shown below.

# number range
seq.range(10)

# text file
seq.open('filepath')

# json file
seq.json('filepath')

# jsonl file
seq.jsonl('filepath')

# csv file
seq.csv('filepath')
seq.csv_dict_reader('filepath')

# sqlite3 db and sql query
seq.sqlite3('filepath', 'select * from data')

For more information on the parameters that these functions can take, reference the streams documentation

Transformations and Actions APIs

Below is the complete list of functions which can be called on a stream object from seq. For complete documentation reference transformation and actions API.

Function Description Type map(func)/select(func) Maps func onto elements of sequence transformation starmap(func)/smap(func) Applies func to sequence with itertools.starmap transformation filter(func)/where(func) Filters elements of sequence to only those where func(element) is True transformation filter_not(func) Filters elements of sequence to only those where func(element) is False transformation flatten() Flattens sequence of lists to a single sequence transformation flat_map(func) Maps func to each element, then merges the result to one flat sequence. func must return an iterable transformation group_by(func) Groups sequence into (key, value) pairs where key=func(element) and value is from the original sequence transformation group_by_key() Groups sequence of (key, value) pairs by key transformation reduce_by_key(func) Reduces list of (key, value) pairs using func