Create a Full Machine Learning Pipeline Without Coding in AWS SageMaker Canvas

---

Discover how to set up your SageMaker domain and user profile in AWS Academy, explore and clean data effortlessly using AWS SageMaker Canvas, and build a complete machine-learning pipeline without coding. Perfect for those who want to harness the power of machine learning without programming!

Data

Download the house price data from https://github.com/xbwei/machine_learning_in_python/blob/master/house_price_full.csv.

Steps

• Download the house price data and upload it to a S3 bucket.  Create a bucket if needed.

• Open the SageMaker service and create a SageMaker Domain by clicking Create domain:

  • Choose set up for organizations

  • Provide a domain name

  • Use the existing LabRole for step 2

  • In Step 3, choose the existing execution role as follows

arn:aws:iam::ACCOUNT-ID:role/LabRole

• In Step 5, choose the public internet access and provide two subnets.

• Set up a SageMaker user profile when the domain is ready

  • Open the created domain and select User Profiles

  • Choose Add User and provide a username

  • In step 4, choose the existing execution role as follows

arn:aws:iam::ACCOUNT-ID:role/LabRole

• Start Canvas with the created user profile.

• In Data Wrangler, clean the house price data and split it into training and test datasets.

  • Import the house price data from S3

  • Convert ID and zip to string

  • Filter out house type that is lot or land

  • Drop columns with many missing values or columns you think will not contribute to the prediction.

  • Remove rows with missing values.

  • Celan the lot_size, and you need this Python code to clean the lot_size: 

df.loc[df['lot_size'] < 10, 'lot_size'] = df['lot_size']*43560

• Remove outliers

• Run the quality report to drop the columns with low-prediction power.

• Split data for training and testing datasets, copy the S3 location and import the datasets to Canvas.

• In My Models, train a model to predict the house price.

  • Use the training dataset

  • Uncheck the ID column

  • Use the Auto training configuration and Quick Build

  • Check model performance

• Predict data

  • Generate a manual prediction with the test dataset

  • Upload the prediction result to Canvas

  • Build a new data flow to join the prediction result and the test dataset.

  • Calculate the Feature Correlation and create a scatter plot in Data analysis to evaluate the model performance.

• Optional: Deploy the Model to use it outside of Canvas or Register it to share it with other users.

• Log out of Canvas to save the credits.