phanthanhtrung / mlchain-python Goto Github PK

MLChain: Auto-Magical Deploy AI model at large scale, high performance, and easy to use

Explore the docs »
Our Website · Examples in Python

MLChain is a simple, easy to use library that allows you to deploy your Machine Learning model to hosting server easily and efficiently, drastically reducing the time required to build API that support an end-to-end AI product.

• Fast: MLChain prioritize speed above other criteria.

• Fast to code: With a finished Machine Learning model, it takes 4 minutes on average to deploy a fully functioning API with MLChain.

• Flexible: The nature of ML-Chain allows developing end-to-end adaptive, with varied serializer and framework hosting at your choice.

• Less debugging : We get it. Humans make mistakes. With MLChain, its configuration makes debugging a lot easier and almost unnecessary.

• Easy to code: as a piece of cake!

• Standards-based: Based on the open standards for APIs: OpenAPI (previously known as Swagger), along with JSON Schema and other options.

MLChain required Python 3.6 and above

To install latest stable version of MLChain, simply run:

pip install mlchain

If you can't wait for the next release, install the most up to date code with from master branch by running the following command:

git clone https://github.com/Techainer/mlchain-python
cd mlchain-python
pip install -r requirements.txt
python setup.py install

Or simply install using git:

pip install git+https://github.com/Techainer/mlchain-python@master --upgrade

Read ours documentation here

Here's a minimal example of serving a dummy python class

Create a server.py file:

import cv2
import numpy as np
from mlchain.base import ServeModel


class Model():
  """ Just a dummy model """

  def predict(self, image: np.ndarray):
      """
      Resize input to 100 by 100.
      Args:
          image (numpy.ndarray): An input image.
      Returns:
          The image (np.ndarray) at 100 by 100.
      """
      image = cv2.resize(image, (100, 100))
      return image


# Define model
model = Model()

# Serve model
serve_model = ServeModel(model)

# Deploy model
if __name__ == '__main__':
    from mlchain.server import FlaskServer
    # Run flask model with upto 12 threads
    FlaskServer(serve_model).run(port=5000, threads=12)

Now run:

python server.py

And you should see something like this:

[mlchain-logger]:[7895] [2020-08-18 09:53:02 +0700]-[INFO]-[flask_server.py:424]---------------------------------------------------------------------------------
[mlchain-logger]:[7895] [2020-08-18 09:53:02 +0700]-[INFO]-[flask_server.py:425]-Served model with Flask at host=127.0.0.1, port=5000
[mlchain-logger]:[7895] [2020-08-18 09:53:02 +0700]-[INFO]-[flask_server.py:426]-Debug = False
[mlchain-logger]:[7895] [2020-08-18 09:53:02 +0700]-[INFO]-[flask_server.py:427]---------------------------------------------------------------------------------

Now you can access your API at http://localhost:5000

You can open Swagger UI at http://localhost:5000/swagger and try your API out right away

After explore all your API endpoint over there, create a client.py file:

import numpy as np
from mlchain.client import Client

model = Client(api_address='http://localhost:5000').model()
# Create a dummy input with shape (200, 200)
input_image = np.ones((200, 200), dtype=np.uint8)
# Then pass it through our client just like normal Python
result_image = model.predict(input_image)
print(result_image.shape)  # And the result should be (100, 100)

Now you have a supper simple Client to work with. Sooo easy :D

• Serving MNIST using MLchain: https://github.com/Techainer/mnist-mlchain-examples

Welcome to the MLChain community!

If you have any questions, please feel free to:

1. Read the docs
2. Open an issues

We are happy to help