Simple ML Model on AWS lambda Service

The purpose of this project is to show how to create a ML model and make it available through a AWS lambda service.

The main requirements for this projects are:

• scikit-learn==0.19.0
• scipy==0.19.1
• Flask==0.12.2
• numpy==1.13.1
• pandas==0.20.3
• requests==2.18.4
• zappa==0.43.2
• awscli==1.11.153

You also need to have a valid AWS account from amazon.

Preparing the environment

1. Creating the virtual environment:

virtualenv ml_cloud_aws

2. Activate the virtual environment:

source ml_cloud_aws/bin/activate

3. Clone the repository:

   cd ml_cloud_aws
   git clone https://github.com/dasoto/ml_cloud_aws.git
   

Create your ML model

On this example we are using the dataset of breast cancer included in our scikit-learn.datasets. Run the script to create the machine learning model model.pkl:

python ml_cloud_src/model_creation.py

Now you have your model.pkl file that can be loaded as described in test_model.py to test your odel locally or app.py

Load your model into a local web server using Flask

Now that you have your model, the app.py file contains a very basic example of a webservice that can answer POST requests. You can test it locally running:

1. Running the Flask server

python app.py

The code is:

from flask import Flask, jsonify, request
from sklearn.externals import joblib
import pandas as pd
import numpy as np

app = Flask(__name__)

print('Loading Model...')
clf = joblib.load('model.pkl')
print(clf)
print('Model loaded.')
@app.route('/predict', methods=['POST'])
def predict():
    print('Receiving data of prediction')
    data = request.get_json(force = True)
    data = dict(data[0])
    print(data)
    print('Printing data 0')
    print(data['1'])
    predict_request = [data['1'],data['2'],data['3'],data['4'],data['5'],data['6'],data['7'],\
                       data['8'],data['9'],data['10'],data['11'],data['12'],data['13'],\
                       data['14'],data['15'],data['16'],data['17'],data['18'],data['19'],\
                       data['20'],data['21'],data['22'],data['23'],data['24'],data['25'],\
                       data['26'],data['27'],data['28'],data['29'],data['30']]
    print('After...')
    print(predict_request)
    predict_request = np.array(predict_request).reshape(1,-1)
    prediction = str(clf.predict(predict_request).astype(int))
    print('Predicted value is:')
    print(prediction)
    print(type(prediction))
    return jsonify({'prediction': list(prediction)})
    #return jsonify(result = prediction)
if __name__ == '__main__':
    print('Passing main:')
    app.run(port=8080,debug=True)

2. Testing locally using curl: TOu can test locally using the curl command in linux or mac:

curl -d '[
      {  "1":1.62500000e+01,   "2":1.95100000e+01,   "3":1.09800000e+02,
         "4":8.15800000e+02,   "5":1.02600000e-01,   "6":1.89300000e-01,
         "7":2.23600000e-01,   "8":9.19400000e-02,   "9":2.15100000e-01,
         "10":6.57800000e-02,  "11":3.14700000e-01,   "12":9.85700000e-01,
         "13":3.07000000e+00,   "14":3.31200000e+01,   "15":9.19700000e-03,
         "16":5.47000000e-02,   "17":8.07900000e-02,   "18":2.21500000e-02,
         "19":2.77300000e-02,   "20":6.35500000e-03,   "21":1.73900000e+01,
         "22":2.30500000e+01,   "23":1.22100000e+02,   "24":9.39700000e+02,
         "25":1.37700000e-01,   "26":4.46200000e-01,   "27":5.89700000e-01,
         "28":1.77500000e-01,   "29":3.31800000e-01,   "30":9.13600000e-02}
]' -H "Content-Type: application/json" \
     -X POST http://127.0.0.1/predict && \
    echo -e "\n -> predict OK"

3. Test using a python script:

import requests
import json

url = 'http://127.0.0.1:8080/predict'
data = json.dumps([{  "1":1.62500000e+01,   "2":1.95100000e+01,   "3":1.09800000e+02,
         "4":8.15800000e+02,   "5":1.02600000e-01,   "6":1.89300000e-01,
         "7":2.23600000e-01,   "8":9.19400000e-02,   "9":2.15100000e-01,
         "10":6.57800000e-02,  "11":3.14700000e-01,   "12":9.85700000e-01,
         "13":3.07000000e+00,   "14":3.31200000e+01,   "15":9.19700000e-03,
         "16":5.47000000e-02,   "17":8.07900000e-02,   "18":2.21500000e-02,
         "19":2.77300000e-02,   "20":6.35500000e-03,   "21":1.73900000e+01,
         "22":2.30500000e+01,   "23":1.22100000e+02,   "24":9.39700000e+02,
         "25":1.37700000e-01,   "26":4.46200000e-01,   "27":5.89700000e-01,
         "28":1.77500000e-01,   "29":3.31800000e-01,   "30":9.13600000e-02}])
r = requests.post(url,data)
print(r.json())

Upload to Amazon AWS Lambda service

To complete this step we will use awscli and zappa.

1. Install awscli:

pip install awscli

2. Install zappa:

pip install zappa

3. Configure amazon credentials and region:

aws configure

4. Init zappa:

zappa init

5. Edit your zappa settings file zappa_settings.json and add the line slim_handler:

{
    "dev": {
        "app_function": "app.app",
        "aws_region": "us-east-1",
        "profile_name": "default",
        "s3_bucket": "zappa-ml-model-dev",
        "slim_handler": true
    }
}

6. Deploy your webservice:

zappa deploy

7. With the address that will provide the previous step you can modify the python script or curl command to test your service!!!