跳转至

将 ML/DL 模型公开为 REST API

原文:https://www.geesforgeks.org/expering-ml-dl-models-as-rest-API/

在本文中,我们将学习如何将 ML/DL 模型公开为烧瓶 API。

我们将使用的框架: Keras 是一个深度学习库,构建在后端之上,如 Tensorflow、茶诺或 CNTK。它提供了抽象,并允许快速开发 ML/DL 模型。

Flask 是 Python 中的一个微 web 框架,用来快速旋转服务器来服务页面。参见烧瓶介绍

由于本文的重点是使用一个应用编程接口服务于一个 ML/DL 模型,我们不会更深入地研究卷积模型的制作。我们将使用 ResNet50 卷积神经网络。

安装 Tensorflow 和 Keras

步骤 Linux/MAC 窗户
更新 zip 画中画安装 python -m pip 安装-U pip
--- --- ---
Install virtualenv sudo pip install virtualenv pip install virtualenv
--- --- ---
创建一个新文件夹 mkdir ml API & & CD ml API mkdir ml API & & CD ml API
--- --- ---

。\ venv \脚本\激活安装 TensorFlowpip 安装 – 升级张量流pip 安装 – 升级张量流硬安装点安装硬点安装硬安装其他依赖项pip 安装烧瓶 gevent 请求枕头pip install flask gevent requests pillow

例:

创建 REST 应用编程接口:

# keras_server.py

# Python program to expose a ML model as flask REST API

# import the necessary modules
from keras.applications import ResNet50 # pre-built CNN Model
from keras.preprocessing.image import img_to_array 
from keras.applications import imagenet_utils
import tensorflow as tf
from PIL import Image
import numpy as np
import flask
import io

# Create Flask application and initialize Keras model
app = flask.Flask(__name__)
model = None

# Function to Load the model 
def load_model():

    # global variables, to be used in another function
    global model     
    model = ResNet50(weights ="imagenet")
    global graph 
    graph = tf.get_default_graph()

# Every ML/DL model has a specific format
# of taking input. Before we can predict on
# the input image, we first need to preprocess it.
def prepare_image(image, target):
    if image.mode != "RGB":
        image = image.convert("RGB")

    # Resize the image to the target dimensions
    image = image.resize(target) 

    # PIL Image to Numpy array
    image = img_to_array(image) 

    # Expand the shape of an array,
    # as required by the Model
    image = np.expand_dims(image, axis = 0) 

    # preprocess_input function is meant to
    # adequate your image to the format the model requires
    image = imagenet_utils.preprocess_input(image) 

    # return the processed image
    return image

# Now, we can predict the results.
@app.route("/predict", methods =["POST"])
def predict():
    data = {} # dictionary to store result
    data["success"] = False

    # Check if image was properly sent to our endpoint
    if flask.request.method == "POST":
        if flask.request.files.get("image"):
            image = flask.request.files["image"].read()
            image = Image.open(io.BytesIO(image))

            # Resize it to 224x224 pixels 
            # (required input dimensions for ResNet)
            image = prepare_image(image, target =(224, 224))

        # Predict ! global preds, results
            with graph.as_default():
                preds = model.predict(image)
                results = imagenet_utils.decode_predictions(preds)
                data["predictions"] = []

            for (ID, label, probability) in results[0]:
                r = {"label": label, "probability": float(probability)}
                data["predictions"].append(r)

            data["success"] = True

    # return JSON response
    return flask.jsonify(data)

if __name__ == "__main__":
    print(("* Loading Keras model and Flask starting server..."
        "please wait until server has fully started"))
    load_model()
    app.run()

运行烧瓶服务器

python keras_server.py 

注#1: 第一次运行,需要时间下载模型的权重。在随后的运行中,情况不会如此。 注 2: 将图像保存在同一目录(ml-API)–dog.jpg、cat.jpg、lion.jpg。

向服务器发出请求:

方法#1 : 使用 cURL (从这里下载 Windows 的 cURL。)

$ curl -X POST -F image=@dog.jpg "http://localhost:5000/predict"

方法 2 : (如果做一个全栈网络应用程序,用合适的用户界面就合适了) 创建一个简单的 HTML 表单

<!-- index.html -->
<form action = "http://localhost:5000/predict" method = "POST" 
         enctype = "multipart/form-data">
         <input type = "file" name = "image" />
         <input type = "submit"/>
</form>

方法#3 : 创建一个简单的 python 脚本,向 flask 服务器发出 HTTP 请求。

# simple_request.py
import requests
import sys

URL = "http://localhost:5000/predict"

# provide image name as command line argument
IMAGE_PATH = sys.argv[1] 

image = open(IMAGE_PATH, "rb").read()
payload = {"image": image}

# make request to the API
request = requests.post(URL, files = payload).json()

if request["success"]:
    # Print formatted Result
    print("% s % 15s % s"%("Rank", "Label", "Probability"))
    for (i, result) in enumerate(request["predictions"]):
        print("% d.    % 17s %.4f"%(i + 1, result["label"],
            result["probability"]))

    else:
        print("Request failed")

运行为:

python simple_request.py dog.jpg

输出:

完成后要停用(Windows 和 Linux):

> deactivate

因此,一般来说,任何模型的工作流程都是

  • 建立模型并保存。
  • 创建一个烧瓶应用程序。
  • 加载模型
  • 为每种类型的请求定义应用编程接口的端点。
  • 根据模型的体系结构对输入进行预处理,并将其提供给模型
  • 处理输入并将输出发送回客户端。


回到顶部