palashio / nylon Goto Github PK

View Code? Open in Web Editor NEW

84.0 7.0 8.0 1.94 MB

An intelligent, flexible grammar of machine learning.

Home Page: http://docs.paraglide.ai

License: MIT License

Python 100.00%

machine-learning auto-ml grammar data-science

nylon's Introduction

An english representation of machine learning. Modify what you want, let us handle the rest.

Overview

Nylon is a python library that lets you customize automated machine learning workflows through a concise, JSON syntax. It provides a built in grammar, in which you can access different operations in ML with the english language.

Installation

Install latest release version:

pip install -U nylon-ai

Install directory from github:

git clone https://github.com/Palashio/nylon.git
cd nylon-ai
pip install .

Usage: the basics

A new Polymer object should be created everytime you're working with a new dataset. When initializing an object, a dataset in the form of a .csv or .xs file should be passed to it by path:

nylon_object = Polymer('housing.csv')

Now, it's time to create a specifications file using the nylon grammar. Here's a basic one, that lets Nylon handle most of the work. Nylon currently has four major parts in it's grammar: the data reader, preprocessor, modeler, and analysis modules. In the example below, you can see that we're specifying the target column under data (which is always required), and manually specifying the type of preprocessing we'd like. Everything we haven't specified will be handled for us.

{
  "data": {
    "target": "ocean_proximity"
  },
  "preprocessor": {
    "fill": "ALL",
    "label-encode": "ocean_proximity"
  }
}

Now, we can override more components to take advantage of the built in ensembling of SVM's, and nearest neighbors modeling in nylon.

 json_file = {
    "data": {
        "target": "ocean_proximity"
    },
    "preprocessor": {
        "fill": "ALL",
        "label-encode": "ocean_proximity"
    },
    "modeling": {
        "type": ["svms", "neighbors"]
    }
}

Now we can call,

nylon_object.run(json_file)

This will return a fully trained nylon object. You can access all information about this particular iteration in the .results field of the object.

Demos

Asking for help

Welcome to the Nylon community!

If you have any questions, feel free to:

Contact

Shoot me an email at [email protected] if you'd like to get in touch!

Follow me on twitter for updates and my insights about modern AI!

nylon's People

Contributors

Stargazers

Watchers

Forkers

adbmd schots rhythm99 akshayakula ruch-code sarda-devesh admariner bryan2022

nylon's Issues

uuid only generates when you run the model once, doesn't generate everytime there's a new run

look into + implement PCA

move analysis code into analysis.py

overhaul documentation

check comment on reddit where lots of people agreed:

"It seems to be something useful. I wish the documentation was better. Before going into "how" it should first explain the "what". I've spent full 5 minutes skimming the github and the docs and I still don't know what this is.

"Nylon offers an easy to-use Python library to help users build complex machine learning models with the english language"

Is it a natural language processing framework? Or is it a framework with which I can build "complex machine learning models" just by describing them in plain English?

What is this?"

default modeling should try all the models without any parameters, and then find best parameters for that individual model

passing dictionary instead of json file

create design for correlations

things that users choose should affect the automatic choices in different parts of the pipeline. for example, if you add a trimming statement, more models should be tried out or a certain preprocessing mechanic should trigger certain models etc.

should we get into graphs for statistics

should we get into confusion matrices etc. maybe possible integration with vega-lite

y is being saved as None in models.history

TypeError: svm_stroke() got an unexpected keyword argument 'json_file'

nylon_object = Polymer('housing.csv')

spec = {
    "data": {
        "target": "ocean_proximity"
    },
    "modeling": {
        "type": "svms"
    }
}

nylon_object.run(spec)