sagemaker-lab-demo-712567

MySageMakerLabInstance

import pandas as pd
import numpy as np
import sagemaker
import boto3
import io

# Get the SageMaker session, execution role, and S3 bucket
sagemaker_session = sagemaker.Session()
role = sagemaker.get_execution_role()
bucket_name = '' # Replace with your S3 bucket name from Step 1
region = sagemaker_session.boto_region_name
s3_client = boto3.client('s3', region_name=region)

print(f"SageMaker Role ARN: {role}")
print(f"S3 Bucket: {bucket_name}")
print(f"Region: {region}")

# Generate sample data: Predicting house price based on size
# Feature: Size (in square feet), Target: Price (in $1000s)
np.random.seed(42) # for reproducibility
num_samples = 100
house_size = np.random.randint(500, 3500, num_samples) # Size between 500 and 3500 sq ft
# Price = 50 + 0.1 * Size + some noise
# Convert to integer to have whole numbers for price
house_price = (50 + 0.1 * house_size + np.random.normal(0, 20, num_samples)).astype(int)

# Create a Pandas DataFrame
# Ensure 'Price' is the first column for SageMaker Linear Learner CSV input
df = pd.DataFrame({'Price': house_price, 'Size': house_size})

# Display the first few rows
print("\nSample Data (Price as whole numbers):")
print(df.head())

# Convert DataFrame to CSV string
csv_buffer = io.StringIO()
# No header, no index for SageMaker built-in algos like Linear Learner
df.to_csv(csv_buffer, header=False, index=False)
csv_content = csv_buffer.getvalue()

# Define S3 paths
prefix = 'linear-learner-demo' # A prefix (folder) in your S3 bucket
train_data_s3_path = f's3://{bucket_name}/{prefix}/train/train.csv'

# Upload the CSV data to S3
s3_client.put_object(Bucket=bucket_name, Key=f'{prefix}/train/train.csv', Body=csv_content)
print(f"\nTraining data uploaded to: {train_data_s3_path}")

# Define output path for model artifacts
s3_output_location = f's3://{bucket_name}/{prefix}/output'
print(f"Model artifacts will be stored in: {s3_output_location}")

sagemaker-lab-demo-712567

# Get the Docker image URI for SageMaker's Linear Learner algorithm
container = sagemaker.image_uris.retrieve(
    framework='linear-learner',
    region=region,
    version='1' # Use 'latest' or a specific version
)
print(f"Linear Learner container URI: {container}")

# Configure the Linear Learner estimator
linear_estimator = sagemaker.estimator.Estimator(
    image_uri=container,
    role=role,
    instance_count=1,
    instance_type='ml.m5.large', # Choose a suitable instance type for training
    output_path=s3_output_location,
    sagemaker_session=sagemaker_session,
    # Hyperparameters for Linear Learner
    hyperparameters={
        'feature_dim': 1,          # Number of features (just 'Size' in our case)
        'predictor_type': 'regressor', # For regression problems
        'mini_batch_size': 32
    }
)

# Define the input data channels
# The content_type 'text/csv;label_size=1' indicates CSV format where the first column is the target label.
train_input = sagemaker.inputs.TrainingInput(
    s3_data=train_data_s3_path,
    distribution='FullyReplicated',
    content_type='text/csv;label_size=1', # label_size=1 means the first column is the target
    s3_data_type='S3Prefix'
)

data_channels = {'train': train_input}

# Start the training job
print("\nStarting model training...")
linear_estimator.fit(inputs=data_channels, logs='All') # Set logs to 'All' to see training progress
print("Training job completed.")

# Deploy the trained model to an endpoint
print("\nDeploying model to a SageMaker endpoint...")
linear_predictor = linear_estimator.deploy(
    initial_instance_count=1,
    instance_type='ml.m5.large', # Choose a suitable instance type for hosting
    serializer=sagemaker.serializers.CSVSerializer(), # Input format for predictions
    deserializer=sagemaker.deserializers.JSONDeserializer() # Output format from predictions
)
print(f"Endpoint '{linear_predictor.endpoint_name}' deployed successfully.")

# Test the endpoint with a sample house size
sample_house_size = "1500"
# A house of 1500 sq ft (as a string for CSVSerializer)

try:
    response = linear_predictor.predict(sample_house_size)
    # The Linear Learner regressor returns predictions in a JSON
    # structure. For a single prediction, it might look like:
    # {'predictions': [{'score': 200.5}]}
    predicted_price = response['predictions'][0]['score']

    print(f"\nPredicted price for a house of size "
          f"{sample_house_size} sq ft: ${predicted_price*1000:.2f}")

    # Let's try another one
    sample_house_size_2 = "2500"
    response_2 = linear_predictor.predict(sample_house_size_2)
    predicted_price_2 = response_2['predictions'][0]['score']
    print(f"Predicted price for a house of size "
          f"{sample_house_size_2} sq ft: ${predicted_price_2*1000:.2f}")

except Exception as e:
    print(f"Error making prediction: {e}")

# Delete the SageMaker endpoint
try:
    linear_predictor.delete_endpoint()
    print(f"Endpoint '{linear_predictor.endpoint_name}'"
          f" deleted successfully.")
except Exception as e:
    print(f"Error deleting endpoint (it might already be "
          f"deleted or not exist): {e}")

    endpoint_name_to_log = (
        linear_predictor.endpoint_name
        if 'linear_predictor' in locals()
        else 'Unknown'
    )
    print(f"You may need to manually delete the endpoint "
          f"from the SageMaker console: {endpoint_name_to_log}")

permanently delete

sagemaker-lab-demo-712567