scikit_rest 0.1.1

Automatically serve ML model as a REST API

scikit-rest

Automatically deploy your ML model as a REST API

Often times, deploying your favorite Scikit-learn / XGBoost / Pytorch / Tensorflow model as a REST API might take a lot of time. There are a lot of boilerplate codes to be written. scikit-rest is a package designed to alleviate most of the pain points within this process.

Prerequisites

This package officially supports Python 3

Installing

pip install scikit_rest

Usage

The main function offered in this package is serve, with the following syntax:

    serve(
        col_list: List[str],
        col_types: Dict[str, Union[List, type]],
        transform_fn: Callable,
        predict_fn: Union[Callable, sklearn.base.BaseEstimator],
        port: int,
        is_nullable: bool ,
        name: str,
    )

col_list

List of Column names, where the order of the values will dictate the order within the pandas DataFrame

col_list = ['class', 'sex', 'age', 'embarked', 'date', 'is_englishman']

col_types

Dictionary of Column Names and the type of the variable, used for input Validation. If the values of the dictionary is instead a list, We assume that any input for the variable can only be any of the ones listed within the list

col_types = {
    'class' : int,
    'sex' : str,
    'age' : float,
    'embarked': ['C', 'S', 'Q'],
    'date': datetime.datetime,
    'is_englishman': bool
}

transform_fn

Function which convert the input dataframe into test dataframe, we can call model.predict upon to get the final result

def transform_fn(input_df):
    df = input_df.copy()
    df['sex'] = df['sex'].apply(lambda x : transform_sex(x))
    df['embarked'] = df['embarked'].apply(lambda x : transform_embarked(x))
    df['date'] = df['date'].dt.year
    df = df.fillna(0.)
    return df

predict_fn

Function which convert the test dataframe into result. If a ML model instance is passed in, we will instead try to call model.predict_proba / model.predict to get the result

def predict_fn(input_df):
    df = input_df.copy()
    return model.predict(df).item()

port

Port Number where the REST API should be served upon

is_nullable

Whether input API can be nullable

name

Name of the program

Example

Example of Usage can be found at example folder

Contributing

Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.

Authors

[Aditya Kelvianto Sidharta][https://adityasidharta.com]

License

This project is licensed under the MIT License - see the LICENSE file for details