Machine learning services: choosing the right one
What each of the eight SAA-C03 machine learning services does, when to use a pretrained AI service and when to build a model with SageMaker AI, common pipelines that combine them, and how to go from the clues in a question to the right service.
3 min read · Last reviewed · SAA-C03
Sign in to bookmark tutorials and track what you have read.
On this page
The services at a glance
The ML (Machine learning)Software that learns patterns from example data to make predictions or decisions, instead of following rules written by hand. A model is first trained on data, then used to make predictions on new data, which is called inference.Glossary entry category of the exam covers eight services. Seven are pretrained AI services: you call an API (Application programming interface)A defined way for one program to ask another for data or actions, such as a set of HTTPS requests and responses. Every AWS service is controlled through its API, which the console, the CLI, and the SDKs call for you.Glossary entry and get a result with no ML expertise. One, SageMaker AI, is for building, training, and hosting your own models. Most questions describe the input (images, documents, text, speech) and the output wanted, and the answer is the service that does that conversion. Each service name links to its own tutorial.
| Service | Input to output | Typical use |
|---|---|---|
| Amazon SageMaker AI | Your data to your own model | Build, train, and deploy custom ML models |
| Amazon Rekognition | Images and video to labels, faces, text, moderation | Content moderation, face verification, object detection |
| Amazon Textract | Documents to text, forms, tables, fields | Replace manual data entry from scanned documents |
| Amazon Comprehend | Text to entities, key phrases, sentiment, language, PII (Personally identifiable information)Information that can identify a person, alone or combined with other data, such as a name, email address, or passport number. Amazon Macie finds PII in Amazon S3.Glossary entry | Analyze reviews, support tickets, social posts |
| Amazon Transcribe | Speech to text | Subtitles, call transcripts, meeting notes |
| Amazon Polly | Text to lifelike speech | Voice for apps, audio versions of articles |
| Amazon Translate | Text in one language to another | Multilingual apps and content |
| Amazon Lex | Voice and text conversations | Chatbots and voice assistants |
Two choices to make
- Pretrained service or custom model? If a pretrained service already does the task, use it: it is faster, cheaper, and needs no ML skills. Use SageMaker AI when the problem is specific to your data, such as predicting churn or fraud from your own records, or when you need full control of the model.
- Which conversion? Match the input and output: images to Rekognition, documents to Textract, text meaning to Comprehend, speech to text with Transcribe, text to speech with Polly, language to language with Translate, conversation with Lex.
Services are often chained. A call center might Transcribe a call, Translate the transcript, and run Comprehend for sentiment. A document workflow might run Textract and then Comprehend to find entities and personal data.
Reverse mindmap: from the question to the service
- Build your own model
- Custom model on our data leads to SageMaker AI
- No-code predictions for analysts leads to SageMaker Canvas
- Images and documents
- Objects, faces, unsafe content leads to Rekognition
- Text, forms, tables from scans leads to Textract
- Text
- Sentiment, entities, key phrases, PII leads to Comprehend
- Translate to another language leads to Translate
- Speech and conversation
- Speech to text leads to Transcribe
- Text to speech leads to Polly
- Chatbot or voice assistant leads to Lex
Commonly confused pairs
| If the choice is between | Choose the first when | Choose the second when |
|---|---|---|
| Rekognition or Textract | Analyzing photos or video, including text in scenes | Extracting text and structure from documents |
| Transcribe or Polly | Speech to text | Text to speech |
| Comprehend or Translate | Understanding what text means | Changing the language of text |
| Lex or Polly | A two-way conversation that understands intent | One-way speech output |
| AI service or SageMaker AI | A pretrained API already does the task | You need a custom model trained on your data |
Key takeaways
- Seven pretrained AI services cover vision, documents, text, speech, and conversation; SageMaker AI builds custom models.
- Match the input and output: images to Rekognition, documents to Textract, text to Comprehend or Translate, speech to text with Transcribe, text to speech with Polly, chatbots with Lex.
- Prefer a pretrained service when it solves the problem; choose SageMaker AI for custom models.
Sources
- AWS Certification: SAA-C03 in-scope AWS services
- Amazon SageMaker AI Developer Guide: What is Amazon SageMaker AI?
- Amazon Rekognition Developer Guide: What is Amazon Rekognition?
- Amazon Textract Developer Guide: What is Amazon Textract?
- Amazon Comprehend Developer Guide: What is Amazon Comprehend?
- Amazon Transcribe Developer Guide: What is Amazon Transcribe?
- Amazon Polly Developer Guide: What is Amazon Polly?
- Amazon Translate Developer Guide: What is Amazon Translate?
- Amazon Lex V2 Developer Guide: What is Amazon Lex V2?