Regression with Amazon SageMaker XGBoost algorithm
This notebook’s CI test result for us-west-2 is as follows. CI test results in other regions can be found at the end of the notebook.
Single machine training for regression with Amazon SageMaker XGBoost algorithm
Introduction
This notebook demonstrates the use of Amazon SageMaker’s implementation of the XGBoost algorithm to train and host a regression model. We use the Abalone data originally from the UCI data repository [1]. More details about the original dataset can be found here. In the libsvm converted version, the nominal feature (Male/Female/Infant) has been converted into a real valued feature. Age of abalone is to be predicted from eight physical measurements. Dataset is already processed and stored on S3. Scripts used for processing the data can be found in the Appendix. These include downloading the data, splitting into train, validation and test, and uploading to S3 bucket.
[1] Dua, D. and Graff, C. (2019). UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA: University of California, School of Information and Computer Science.
Setup
This notebook was tested in Amazon SageMaker Studio on a ml.t3.medium instance with Python 3 (Data Science) kernel.
Let’s start by specifying: 1. The S3 buckets and prefixes that you want to use for saving the model and where training data is located. This should be within the same region as the Notebook Instance, training, and hosting. 1. The IAM role arn used to give training and hosting access to your data. See the documentation for how to create these. Note, if more than one role is required for notebook instances, training, and/or hosting, please replace the boto regexp with a the appropriate full IAM role arn string(s).
[ ]:
!pip3 install -U sagemaker
[ ]:
%%time
import os
import boto3
import re
import sagemaker
role = sagemaker.get_execution_role()
region = boto3.Session().region_name
s3_client = boto3.client("s3")
# S3 bucket where the training data is located.
data_bucket = f"sagemaker-sample-files"
data_prefix = "datasets/tabular/uci_abalone"
data_bucket_path = f"s3://{data_bucket}"
# S3 bucket for saving code and model artifacts.
# Feel free to specify a different bucket and prefix
output_bucket = sagemaker.Session().default_bucket()
output_prefix = "sagemaker/DEMO-xgboost-abalone-default"
output_bucket_path = f"s3://{output_bucket}"
for data_category in ["train", "test", "validation"]:
data_key = "{0}/{1}/abalone.{1}".format(data_prefix, data_category)
output_key = "{0}/{1}/abalone.{1}".format(output_prefix, data_category)
data_filename = "abalone.{}".format(data_category)
s3_client.download_file(data_bucket, data_key, data_filename)
s3_client.upload_file(data_filename, output_bucket, output_key)
Training the XGBoost model
After setting training parameters, we kick off training, and poll for status until training is completed, which in this example, takes between 5 and 6 minutes.
Training can be done by either calling SageMaker Training with a set of hyperparameters values to train with, or by leveraging SageMaker Automatic Model Tuning (AMT). AMT, also known as hyperparameter tuning (HPO), finds the best version of a model by running many training jobs on your dataset using the algorithm and ranges of hyperparameters that you specify. It then chooses the hyperparameter values that result in a model that performs the best, as measured by a metric that you choose.
In this notebook, both methods are used for demonstration purposes, but the model that the HPO job creates is the one that is eventually hosted. You can instead choose to deploy the model created by the standalone training job by changing the below variable deploy_amt_model
to False.
Initiliazing common variables
[ ]:
container = sagemaker.image_uris.retrieve("xgboost", region, "1.7-1")
client = boto3.client("sagemaker", region_name=region)
deploy_amt_model = True
Training with SageMaker Training
[ ]:
%%time
import boto3
from time import gmtime, strftime
import time
training_job_name = f"DEMO-xgboost-regression-{strftime('%Y-%m-%d-%H-%M-%S', gmtime())}"
# Ensure that the training and validation data folders generated above are reflected in the "InputDataConfig" parameter below.
create_training_params = {
"AlgorithmSpecification": {"TrainingImage": container, "TrainingInputMode": "File"},
"RoleArn": role,
"OutputDataConfig": {"S3OutputPath": f"{output_bucket_path}/{output_prefix}/single-xgboost"},
"ResourceConfig": {"InstanceCount": 1, "InstanceType": "ml.m5.2xlarge", "VolumeSizeInGB": 5},
"TrainingJobName": training_job_name,
"HyperParameters": {
"max_depth": "5",
"eta": "0.2",
"gamma": "4",
"min_child_weight": "6",
"subsample": "0.7",
"objective": "reg:linear",
"num_round": "50",
"verbosity": "2",
},
"StoppingCondition": {"MaxRuntimeInSeconds": 3600},
"InputDataConfig": [
{
"ChannelName": "train",
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri": f"{output_bucket_path}/{output_prefix}/train",
"S3DataDistributionType": "FullyReplicated",
}
},
"ContentType": "libsvm",
"CompressionType": "None",
},
{
"ChannelName": "validation",
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri": f"{output_bucket_path}/{output_prefix}/validation",
"S3DataDistributionType": "FullyReplicated",
}
},
"ContentType": "libsvm",
"CompressionType": "None",
},
],
}
print(
f"Creating a training job with name: {training_job_name}. It will take between 5 and 6 minutes to complete."
)
client.create_training_job(**create_training_params)
status = client.describe_training_job(TrainingJobName=training_job_name)["TrainingJobStatus"]
print(status)
while status != "Completed" and status != "Failed":
time.sleep(60)
status = client.describe_training_job(TrainingJobName=training_job_name)["TrainingJobStatus"]
print(status)
Note that the “validation” channel has been initialized too. The SageMaker XGBoost algorithm actually calculates RMSE and writes it to the CloudWatch logs on the data passed to the “validation” channel.
Tuning with SageMaker Automatic Model Tuning
To create a tuning job using the AWS SageMaker Automatic Model Tuning API, you need to define 3 attributes.
the tuning job name (string)
the tuning job config (to specify settings for the hyperparameter tuning job - JSON object)
training job definition (to configure the training jobs that the tuning job launches - JSON object).
To learn more about that, refer to the Configure and Launch a Hyperparameter Tuning Job documentation.
Note that the tuning job will 12-17 minutes to complete.
[ ]:
from time import gmtime, strftime, sleep
tuning_job_name = "DEMO-xgboost-reg-" + strftime("%d-%H-%M-%S", gmtime())
tuning_job_config = {
"ParameterRanges": {
"CategoricalParameterRanges": [],
"ContinuousParameterRanges": [
{
"MaxValue": "0.5",
"MinValue": "0.1",
"Name": "eta",
},
{
"MaxValue": "5",
"MinValue": "0",
"Name": "gamma",
},
{
"MaxValue": "120",
"MinValue": "0",
"Name": "min_child_weight",
},
{
"MaxValue": "1",
"MinValue": "0.5",
"Name": "subsample",
},
{
"MaxValue": "2",
"MinValue": "0",
"Name": "alpha",
},
],
"IntegerParameterRanges": [
{
"MaxValue": "10",
"MinValue": "0",
"Name": "max_depth",
},
{
"MaxValue": "4000",
"MinValue": "1",
"Name": "num_round",
},
],
},
# SageMaker sets the following default limits for resources used by automatic model tuning:
# https://docs.aws.amazon.com/sagemaker/latest/dg/automatic-model-tuning-limits.html
"ResourceLimits": {
# Increase the max number of training jobs for increased accuracy (and training time).
"MaxNumberOfTrainingJobs": 6,
# Change parallel training jobs run by AMT to reduce total training time. Constrained by your account limits.
# if max_jobs=max_parallel_jobs then Bayesian search turns to Random.
"MaxParallelTrainingJobs": 2,
},
"Strategy": "Bayesian",
"HyperParameterTuningJobObjective": {"MetricName": "validation:rmse", "Type": "Minimize"},
}
training_job_definition = {
"AlgorithmSpecification": {"TrainingImage": container, "TrainingInputMode": "File"},
"InputDataConfig": [
{
"ChannelName": "train",
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri": f"{output_bucket_path}/{output_prefix}/train",
"S3DataDistributionType": "FullyReplicated",
}
},
"ContentType": "libsvm",
"CompressionType": "None",
},
{
"ChannelName": "validation",
"DataSource": {
"S3DataSource": {
"S3DataType": "S3Prefix",
"S3Uri": f"{output_bucket_path}/{output_prefix}/validation",
"S3DataDistributionType": "FullyReplicated",
}
},
"ContentType": "libsvm",
"CompressionType": "None",
},
],
"OutputDataConfig": {"S3OutputPath": f"{output_bucket_path}/{output_prefix}/single-xgboost"},
"ResourceConfig": {"InstanceCount": 1, "InstanceType": "ml.m5.2xlarge", "VolumeSizeInGB": 5},
"RoleArn": role,
"StaticHyperParameters": {
"objective": "reg:linear",
"verbosity": "2",
},
"StoppingCondition": {"MaxRuntimeInSeconds": 43200},
}
print(
f"Creating a tuning job with name: {tuning_job_name}. It will take between 12 and 17 minutes to complete."
)
client.create_hyper_parameter_tuning_job(
HyperParameterTuningJobName=tuning_job_name,
HyperParameterTuningJobConfig=tuning_job_config,
TrainingJobDefinition=training_job_definition,
)
status = client.describe_hyper_parameter_tuning_job(HyperParameterTuningJobName=tuning_job_name)[
"HyperParameterTuningJobStatus"
]
print(status)
while status != "Completed" and status != "Failed":
time.sleep(60)
status = client.describe_hyper_parameter_tuning_job(
HyperParameterTuningJobName=tuning_job_name
)["HyperParameterTuningJobStatus"]
print(status)
Set up hosting for the model
In order to set up hosting, we have to import the model from training to hosting.
Import model into hosting
Register the model with hosting. This allows the flexibility of importing models trained elsewhere.
[ ]:
%%time
import boto3
from time import gmtime, strftime
if deploy_amt_model == True:
training_of_model_to_be_hosted = client.describe_hyper_parameter_tuning_job(
HyperParameterTuningJobName=tuning_job_name
)["BestTrainingJob"]["TrainingJobName"]
else:
training_of_model_to_be_hosted = training_job_name
model_name = f"{training_of_model_to_be_hosted}-model"
print(model_name)
info = client.describe_training_job(TrainingJobName=training_of_model_to_be_hosted)
model_data = info["ModelArtifacts"]["S3ModelArtifacts"]
print(model_data)
primary_container = {"Image": container, "ModelDataUrl": model_data}
create_model_response = client.create_model(
ModelName=model_name, ExecutionRoleArn=role, PrimaryContainer=primary_container
)
print(create_model_response["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, customers 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 = f"DEMO-XGBoostEndpointConfig-{strftime('%Y-%m-%d-%H-%M-%S', gmtime())}"
print(f"Creating endpoint config with name: {endpoint_config_name}.")
create_endpoint_config_response = client.create_endpoint_config(
EndpointConfigName=endpoint_config_name,
ProductionVariants=[
{
"InstanceType": "ml.m5.xlarge",
"InitialVariantWeight": 1,
"InitialInstanceCount": 1,
"ModelName": model_name,
"VariantName": "AllTraffic",
}
],
)
print(f"Endpoint Config Arn: {create_endpoint_config_response['EndpointConfigArn']}")
Create endpoint
Lastly, the customer creates 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 = f'DEMO-XGBoostEndpoint-{strftime("%Y-%m-%d-%H-%M-%S", gmtime())}'
print(
f"Creating endpoint with name: {endpoint_name}. This will take between 9 and 11 minutes to complete."
)
create_endpoint_response = client.create_endpoint(
EndpointName=endpoint_name, EndpointConfigName=endpoint_config_name
)
print(create_endpoint_response["EndpointArn"])
resp = client.describe_endpoint(EndpointName=endpoint_name)
status = resp["EndpointStatus"]
while status == "Creating":
print(f"Status: {status}")
time.sleep(60)
resp = client.describe_endpoint(EndpointName=endpoint_name)
status = resp["EndpointStatus"]
print(f"Arn: {resp['EndpointArn']}")
print(f"Status: {status}")
Validate the model for use
Finally, the customer can now validate the model for use. They can obtain the endpoint from the client library using the result from previous operations, and generate classifications from the trained model using that endpoint.
[ ]:
runtime_client = boto3.client("runtime.sagemaker", region_name=region)
Download test data
[ ]:
FILE_TEST = "abalone.test"
s3 = boto3.client("s3")
s3.download_file(data_bucket, f"{data_prefix}/test/{FILE_TEST}", FILE_TEST)
Start with a single prediction.
[ ]:
!head -1 abalone.test > abalone.single.test
[ ]:
%%time
import json
from itertools import islice
import math
import struct
file_name = "abalone.single.test" # customize to your test file
with open(file_name, "r") as f:
payload = f.read().strip()
response = runtime_client.invoke_endpoint(
EndpointName=endpoint_name, ContentType="text/x-libsvm", Body=payload
)
result = response["Body"].read()
result = result.decode("utf-8")
result = result.split(",")
result = [math.ceil(float(i)) for i in result]
label = payload.strip(" ").split()[0]
print(f"Label: {label}\nPrediction: {result[0]}")
OK, a single prediction works. Let’s do a whole batch to see how good is the predictions accuracy.
[ ]:
import sys
import math
def do_predict(data, endpoint_name, content_type):
payload = "\n".join(data)
response = runtime_client.invoke_endpoint(
EndpointName=endpoint_name, ContentType=content_type, Body=payload
)
result = response["Body"].read()
result = result.decode("utf-8")
result = result.strip("\n").split("\n")
preds = [float(num) for num in result]
preds = [math.ceil(num) for num in preds]
return preds
def batch_predict(data, batch_size, endpoint_name, content_type):
items = len(data)
arrs = []
for offset in range(0, items, batch_size):
if offset + batch_size < items:
results = do_predict(data[offset : (offset + batch_size)], endpoint_name, content_type)
arrs.extend(results)
else:
arrs.extend(do_predict(data[offset:items], endpoint_name, content_type))
sys.stdout.write(".")
return arrs
The following helps us calculate the Median Absolute Percent Error (MdAPE) on the batch dataset.
[ ]:
%%time
import json
import numpy as np
with open(FILE_TEST, "r") as f:
payload = f.read().strip()
labels = [int(line.split(" ")[0]) for line in payload.split("\n")]
test_data = [line for line in payload.split("\n")]
preds = batch_predict(test_data, 100, endpoint_name, "text/x-libsvm")
print(
"\n Median Absolute Percent Error (MdAPE) = ",
np.median(np.abs(np.array(labels) - np.array(preds)) / np.array(labels)),
)
Delete Endpoint
Once you are done using the endpoint, you can use the following to delete it.
[ ]:
client.delete_endpoint(EndpointName=endpoint_name)
Appendix
Data split and upload
Following methods split the data into train/test/validation datasets and upload files to S3.
[ ]:
import io
import boto3
import random
def data_split(
FILE_DATA,
FILE_TRAIN,
FILE_VALIDATION,
FILE_TEST,
PERCENT_TRAIN,
PERCENT_VALIDATION,
PERCENT_TEST,
):
data = [l for l in open(FILE_DATA, "r")]
train_file = open(FILE_TRAIN, "w")
valid_file = open(FILE_VALIDATION, "w")
tests_file = open(FILE_TEST, "w")
num_of_data = len(data)
num_train = int((PERCENT_TRAIN / 100.0) * num_of_data)
num_valid = int((PERCENT_VALIDATION / 100.0) * num_of_data)
num_tests = int((PERCENT_TEST / 100.0) * num_of_data)
data_fractions = [num_train, num_valid, num_tests]
split_data = [[], [], []]
rand_data_ind = 0
for split_ind, fraction in enumerate(data_fractions):
for i in range(fraction):
rand_data_ind = random.randint(0, len(data) - 1)
split_data[split_ind].append(data[rand_data_ind])
data.pop(rand_data_ind)
for l in split_data[0]:
train_file.write(l)
for l in split_data[1]:
valid_file.write(l)
for l in split_data[2]:
tests_file.write(l)
train_file.close()
valid_file.close()
tests_file.close()
def write_to_s3(fobj, bucket, key):
return (
boto3.Session(region_name=region)
.resource("s3")
.Bucket(bucket)
.Object(key)
.upload_fileobj(fobj)
)
def upload_to_s3(bucket, channel, filename):
fobj = open(filename, "rb")
key = f"{prefix}/{channel}"
url = f"s3://{bucket}/{key}/{filename}"
print(f"Writing to {url}")
write_to_s3(fobj, bucket, key)
Data ingestion
Next, we read the dataset from the existing repository into memory, for preprocessing prior to training. This processing could be done in situ by Amazon Athena, Apache Spark in Amazon EMR, Amazon Redshift, etc., assuming the dataset is present in the appropriate location. Then, the next step would be to transfer the data to S3 for use in training. For small datasets, such as this one, reading into memory isn’t onerous, though it would be for larger datasets.
[ ]:
%%time
s3 = boto3.client("s3")
bucket = sagemaker.Session().default_bucket()
prefix = "sagemaker/DEMO-xgboost-abalone-default"
# Load the dataset
FILE_DATA = "abalone"
s3.download_file(
f"sagemaker-example-files-prod-{region}",
f"datasets/tabular/uci_abalone/abalone.libsvm",
FILE_DATA,
)
# split the downloaded data into train/test/validation files
FILE_TRAIN = "abalone.train"
FILE_VALIDATION = "abalone.validation"
FILE_TEST = "abalone.test"
PERCENT_TRAIN = 70
PERCENT_VALIDATION = 15
PERCENT_TEST = 15
data_split(
FILE_DATA,
FILE_TRAIN,
FILE_VALIDATION,
FILE_TEST,
PERCENT_TRAIN,
PERCENT_VALIDATION,
PERCENT_TEST,
)
# upload the files to the S3 bucket
upload_to_s3(bucket, "train", FILE_TRAIN)
upload_to_s3(bucket, "validation", FILE_VALIDATION)
upload_to_s3(bucket, "test", FILE_TEST)
Notebook CI Test Results
This notebook was tested in multiple regions. The test results are as follows, except for us-west-2 which is shown at the top of the notebook.