Amazon SageMaker XGBoost Bring Your Own Model

*Hosting a Pre-Trained scikit-learn Model in Amazon SageMaker XGBoost Algorithm Container*

---

---

Contents

1. Background

2. Setup

3. Optionally, train a scikit learn XGBoost model

4. Upload the pre-trained model to S3

5. Set up hosting for the model

6. Validate the model for use

Setup

Let’s start by specifying:

• AWS region.

• The IAM role arn used to give learning and hosting access to your data. See the documentation for how to specify these.

• The S3 bucket that you want to use for training and model data.

%%time

import os
import boto3
import re
import json
import sagemaker
from sagemaker import get_execution_role

region = boto3.Session().region_name

role = get_execution_role()

bucket = sagemaker.Session().default_bucket()

prefix = 'sagemaker/DEMO-xgboost-byo'
bucket_path = 'https://s3-{}.amazonaws.com/{}'.format(region, bucket)
# customize to your bucket where you have stored the data

Optionally, train a scikit learn XGBoost model

These steps are optional and are needed to generate the scikit-learn model that will eventually be hosted using the SageMaker Algorithm contained.

Install XGboost

Note that for conda based installation, you’ll need to change the Notebook kernel to the environment with conda and Python3.

!conda install -y -c conda-forge xgboost==0.90

Fetch the dataset

%%time
import pickle, gzip, numpy, urllib.request, json

# Load the dataset
urllib.request.urlretrieve("http://deeplearning.net/data/mnist/mnist.pkl.gz", "mnist.pkl.gz")
f = gzip.open('mnist.pkl.gz', 'rb')
train_set, valid_set, test_set = pickle.load(f, encoding='latin1')
f.close()

Prepare the dataset for training

%%time

import struct
import io
import boto3

def get_dataset():
  import pickle
  import gzip
  with gzip.open('mnist.pkl.gz', 'rb') as f:
      u = pickle._Unpickler(f)
      u.encoding = 'latin1'
      return u.load()

train_set, valid_set, test_set = get_dataset()

train_X = train_set[0]
train_y = train_set[1]

valid_X = valid_set[0]
valid_y = valid_set[1]

test_X = test_set[0]
test_y = test_set[1]

Train the XGBClassifier

import xgboost as xgb
import sklearn as sk

bt = xgb.XGBClassifier(max_depth=5,
                       learning_rate=0.2,
                       n_estimators=10,
                       objective='multi:softmax')   # Setup xgboost model
bt.fit(train_X, train_y, # Train it to our data
       eval_set=[(valid_X, valid_y)],
       verbose=False)

Save the trained model file

Note that the model file name must satisfy the regular expression pattern: ^[a-zA-Z0-9](-*[a-zA-Z0-9])*;. The model file also need to tar-zipped.

model_file_name = "DEMO-local-xgboost-model"
bt._Booster.save_model(model_file_name)

!tar czvf model.tar.gz $model_file_name

Upload the pre-trained model to S3

fObj = open("model.tar.gz", 'rb')
key= os.path.join(prefix, model_file_name, 'model.tar.gz')
boto3.Session().resource('s3').Bucket(bucket).Object(key).upload_fileobj(fObj)

Set up hosting for the model

Import model into hosting

This involves creating a SageMaker model from the model file previously uploaded to S3.

from sagemaker.amazon.amazon_estimator import get_image_uri
container = get_image_uri(boto3.Session().region_name, 'xgboost', '0.90-2')

%%time
from time import gmtime, strftime

model_name = model_file_name + strftime("%Y-%m-%d-%H-%M-%S", gmtime())
model_url = 'https://s3-{}.amazonaws.com/{}/{}'.format(region,bucket,key)
sm_client = boto3.client('sagemaker')

print (model_url)

primary_container = {
    'Image': container,
    'ModelDataUrl': model_url,
}

create_model_response2 = sm_client.create_model(
    ModelName = model_name,
    ExecutionRoleArn = role,
    PrimaryContainer = primary_container)

print(create_model_response2['ModelArn'])

Create endpoint configuration

SageMaker supports configuring REST endpoints in hosting with multiple models, e.g. for A/B testing purposes. In order to support this, you can create an endpoint configuration, that describes the distribution of traffic across the models, whether split, shadowed, or sampled in some way. In addition, the endpoint configuration describes the instance type required for model deployment.

from time import gmtime, strftime

endpoint_config_name = 'DEMO-XGBoostEndpointConfig-' + strftime("%Y-%m-%d-%H-%M-%S", gmtime())
print(endpoint_config_name)
create_endpoint_config_response = sm_client.create_endpoint_config(
    EndpointConfigName = endpoint_config_name,
    ProductionVariants=[{
        'InstanceType':'ml.m4.xlarge',
        'InitialInstanceCount':1,
        'InitialVariantWeight':1,
        'ModelName':model_name,
        'VariantName':'AllTraffic'}])

print("Endpoint Config Arn: " + create_endpoint_config_response['EndpointConfigArn'])

Create endpoint

Lastly, you create the endpoint that serves up the model, through specifying the name and configuration defined above. The end result is an endpoint that can be validated and incorporated into production applications. This takes 9-11 minutes to complete.

%%time
import time

endpoint_name = 'DEMO-XGBoostEndpoint-' + strftime("%Y-%m-%d-%H-%M-%S", gmtime())
print(endpoint_name)
create_endpoint_response = sm_client.create_endpoint(
    EndpointName=endpoint_name,
    EndpointConfigName=endpoint_config_name)
print(create_endpoint_response['EndpointArn'])

resp = sm_client.describe_endpoint(EndpointName=endpoint_name)
status = resp['EndpointStatus']
print("Status: " + status)

while status=='Creating':
    time.sleep(60)
    resp = sm_client.describe_endpoint(EndpointName=endpoint_name)
    status = resp['EndpointStatus']
    print("Status: " + status)

print("Arn: " + resp['EndpointArn'])
print("Status: " + status)

Validate the model for use

Now you can obtain the endpoint from the client library using the result from previous operations and generate classifications from the model using that endpoint.

runtime_client = boto3.client('runtime.sagemaker')

Lets generate the prediction for a single datapoint. We’ll pick one from the test data generated earlier.

import numpy as np
point_X = test_X[0]
point_X = np.expand_dims(point_X, axis=0)
point_y = test_y[0]
np.savetxt("test_point.csv", point_X, delimiter=",")

%%time
import json


file_name = 'test_point.csv' #customize to your test file, will be 'mnist.single.test' if use data above

with open(file_name, 'r') as f:
    payload = f.read().strip()

response = runtime_client.invoke_endpoint(EndpointName=endpoint_name,
                                   ContentType='text/csv',
                                   Body=payload)
result = response['Body'].read().decode('ascii')
print('Predicted Class Probabilities: {}.'.format(result))

Post process the output

Since the result is a string, let’s process it to determine the the output class label.

floatArr = np.array(json.loads(result))
predictedLabel = np.argmax(floatArr)
print('Predicted Class Label: {}.'.format(predictedLabel))
print('Actual Class Label: {}.'.format(point_y))

(Optional) Delete the Endpoint

If you’re ready to be done with this notebook, please run the delete_endpoint line in the cell below. This will remove the hosted endpoint you created and avoid any charges from a stray instance being left on.

sm_client.delete_endpoint(EndpointName=endpoint_name)