lazyops 0.0.7

lazyops

Lazy Op(eration)s or lazyops

A collection Python modules and submodules focused on balancing functionality, performance, utility and Lazyiness.

---

Installation

Install directory from pypi

pip install --upgrade lazyops

Install from Github

pip install --upgrade git+https://github.com/trisongz/lazyops

---

The LazyOps Philosophy

Minimalistic: Minimal External Libraries/Dependencies. Minimal imports.

Automatic and JIT: Use a Just-in-Time approach for loading dependencies. Auto-Install external libraries as needed for submodules.

Minimal Lines of Code to Accomplish Goal: Tries to handle as many use-cases as possible with minimal limitations on end-user in how its defined.

High Performant: Carefully selected, battle-tested libraries providing the highest level of performance, while consuming as little resources as needed (RAM) or scale to as many resources allowed (Threads)

(Mostly) Filesystem Agnostic: Cloud Object (GCS atm) or Local files are supported automatically.

Reduce Redudancy: Handle redundant user logic automatically unless explicitly defined.

Object Oriented: Clean Classes and Objects allow for out-of-the-box usage or for customizability.

Builtin Async: If it needs to be done fast or many times, async should probably be available.

Counter-intuitively, building lazyops was anything-but-lazy. To (try) to accomplish all the above, every class, object, or function had to be designed to be written once with the idea that they can stand on their own for as many use cases as possible.

This project is a work in process and will continue to evolve over time.

---

The Minimal Dependencies (and rationale)

• Python 3.7+
• pysimdjson: simdjson's C++ JSON parser can handle 2 GB/s+ of throughput, allowing millions of JSON docs to be parsed per sec/per core.
• fileio: My personal file lib that wraps many useful I/O ops, allowing you to work with cloud objects faster, and in less lines of code.
• aiohttp: Enables async http calls, maximizing throughput for web requests.
• dateparser: Makes working with natural query dates easier 5 days ago.
• six: Enables functionality in retryable decorator.

---

LazyOps API Usage

• LazyOps Core
• LazyClassses
• LazyAPI
• LazyDatabase - WIP
• LazyRPC - WIP
• LazyIO - WIP
• LazySources - WIP
• LazyBloom - WIP

---

LazyOps Core

LazyOps' Core APIs that can be used to handle most common functionalities that can often be redundant to constantly define.

Utilities

Useful functions that speed up development.

from lazyops import lazy_init

# Lazily initialize Tensorflow and ensure Tensorflow == 2.5.0 
# before importing
# tf can now be used as you would if you called `import tensorflow as tf`

tf = lazy_init('tensorflow==2.5.0', 'tensorflow')
physical_devices = tf.config.list_physical_devices('GPU')
print("Num GPUs:", len(physical_devices))

from lazyops import get_logger

# This will create a logger that handles stderr / stdout for Notebook envs
# This logger is also threadsafe as it creates a threadlock to prevent duplication

logger = get_logger('MyLib', submodule = __file__)

from lazyops import timed_cache, retryable, lazymultiproc, lazyproc

def get_dataset(*args, **kwargs):
    # do stuff
    return dataset

# default num_processes = CPU Cores
@lazymultiproc(dataset_callable=get_dataset, num_procs=10) 
def process_dataset(item, *args, **kwargs):
    # do stuff
    return item

# LRU Cache with timed expiration
@timed_cache(seconds=60, maxsize=600) 
def load_dataset(*args, **kwargs):
    # do stuff
    return dataset

# turns this function into a thread.Thread function, returning thread obj.
@lazyproc(name = 'monitoring_x', start = True, daemon = False) 
def monitor_x(*args, **kwargs):
    # do stuff
    return

# Retries this function call 
@retryable(stop_max_attempt_number = 5) 
def call_api(*args, **kwargs):
    # do stuff
    return response

# Utils for running shell commands and cloning repo projects
from lazyops import run_cmd, clone_repo

# clones lazyops to relative /cwd/lops. 
# if add_to_syspath = True, will append repo path to $PATH
clone_repo(repo='https://github.com/trisongz/lazyops', path='lops', absl=False, add_to_syspath=True)

# returns system response of `df -h` as string
res = run_cmd('df -h') 

Time/Timer/Date

Handle time, timers and datetime mostly automatically.

import time
# as easy to remember aliases
from lazyops import timer, ttime, dtime

# or direct imports
from lazyops import LazyTimer, LazyTime, LazyDate

# Initialize a new Time
x = start_time(short=False) # Shorthand, secs = s, mins = m, etc.
time.sleep(5)

print(x.ablstime) # absolute time, which will return
# This allows you to access both a string value (aka no more formatting) and the actual value
# LazyData(string='5.0 secs', value={'secs': 5.046847105026245}, dtype='time')

# You can also call any of the properties 
print(x.secs)
# 5.046847105026245

# The time will continue unless x.stop() is called, which will stop the time from incrementing.
x.stop()

# Timer allows you to initialize and manage a collection of timers

# since it's not initialized yet, calling it will create a new timer and return a LazyTime.
timer('benchmark_1')

# Calling it again will return the same LazyTime object that's been initialized
timer('benchmark_1')

# This will now stop the timer, as the same method as x.stop()
timer.stop_timer('benchmark_1')

# You can access all the timers through the class's global properties
timer.timers()

---

LazyClasses

LazyClasses are an extension of dataclasses, building on top of fastclasses-json, and inspired by heavy usage of dataclasses-json. LazyClass utilizes simdjson as the default JSON serializer for faster performance.

Example Usage:

from dataclass import dataclass
from lazyops import lazyclass

from typing import Optional, Dict

@lazyclass
@dataclass
class HeroPower:
    desc: str
    strength: float


@lazyclass
@dataclass
class NewHero:
    name: str
    desc: str
    powers: Optional[Dict[str, HeroPower]]

    @classmethod
    def load(cls, data):
        if isinstance(data, dict):
            return NewHero.from_dict(data)
        return NewHero.from_json(data)


hero_config = {
    'name': 'Batman',
    'desc': 'Gothams Dark Knight',
    'powers': {
        'Rich': {
            'desc': 'Money enables Batman to do many things',
            'strength': 1000.0
        },
        'Ingenuity': {
            'desc': 'Batman invents lots of things',
            'strength': 500.0
        }
    }
}

Batman = NewHero.load(hero_config)
# Or directly
Batman = NewHero(**hero_config)

Roadmap

• Implement many=True found in dataclasses_json

---

LazyAPI

Define and Create API Endpoints as a Class with just a Dict with builtin async support. API Models are compiled JIT with endpoints set as the LazyAPI class attribute, enabling them to be called as a function.

Example Usage

from lazyops import LazyAPI

api_config = {
    'url': 'https://model.api.domain.com',
    'token': 'supersecuretoken',
    'default_params': {'model': 'nlpmodel'},
    'params_key': 'params',
    'data_key': 'inputs',
    'default_fetch': 'predict',
    'default_async': None,
    'route_config': {
        'predict': {
            'method': 'POST',
            'path': '/predict/',
            'params': {'model': 'nlpmodel'},
            'params_key': 'params',
            'data_key': 'inputs',
            'is_async': False,
            'prefix_payload': 'payload',
            'decode_json': True,
        },
        'status': {
            'method': 'GET',
            'path': '/status/',
            'params': None,
            'params_key': None,
            'data_key': None,
            'decode_json': True,
        },
    }
}

api = LazyAPI.build(api_config)
res = api.predict(data='api model task data ...')
print(res)

---

LazyRPC

WIP

---

LazyDatabase

WIP

---

LazyIO

WIP

---

LazySources

WIP

---

LazyBloom

WIP

---

Credits & Acknowledgements

LazyOps heavily builds & borrows on the work of others in many instances. In order to maintain minimalism and not have unneeded libraries added to dependencies, some parts of the library have been added into LazyOps. The original works often inspired new ideas from studying the code and are definitely worth checking out.

In no particular order:

• retrying
• fastapi-jsonrpc
• fastclasses-json
• dataclasses-json
• gdelt-doc-api

TBC