1. Lightweight and versatile

MLJET is written entirely in Python with few dependencies. This means that once you get interested in MLJET, we can quickly move to a practical example.

Simple Sklearn Local Example

MLJET provides a simple interface to create project and deploy a model.

• mljet.contrib.entrypoint.cook() is a function that takes a model and a strategy and deploys it.

In this example, we simply create a project with scikit-learn.

Firstly, import mljet.

import mljet

Import sklearn.ensemble.RandomForestClassifier as classifier and sklearn.datasets to load the iris dataset.

from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier

Let’s load dataset and create and train a simple model.

X, y = load_iris(return_X_y=True)
clf = RandomForestClassifier()
clf.fit(X, y)

Now, we can deploy the model with mljet.contrib.entrypoint.cook(). Main arguments are model and strategy.

The strategy can be either local or docker. The local strategy will deploy the model locally. The docker strategy will deploy the model in a docker container. The mljet.contrib.entrypoint.cook() function will return a bool or container name.

Now we make only a project without running it. After calling the mljet.contrib.entrypoint.cook() function You can see build folder in the current directory.

It contains:

• Dockerfile - Dockerfile for the model

• requirements.txt - requirements for the model

• models directory - directory with the dumped model

• data directory - directory with the example for the model

• server.py - main file for the model

mljet.contrib.cook(strategy="local", model=clf)

Let’s see on mljet.contrib.entrypoint.cook() signature. This function accepts a lot of parameters, but we see only the most important ones.

• model - model to deploy

• strategy - strategy to use

• backend - backend to use

• need_run - run service after build or not

• scan_path - path to scan for requirements

• silent - silent mode

• verbose - verbose mode

Model parameter is the most important one. It can be any model that implements the predict and other methods.

Note

The model must be picklable.

Note

Now is supported sklearn, xgboost.

Strategy parameter determines the strategy to use.

Backend parameter determines the backend to use. Now is implemented sanic and flask backends.