AWS Machine Learning flowchart a day: Part 2
In part1 of this blogpost series, I covered flowchart for text dataset on AWS ML platform. This blogpost focuses on speech and Image/Video datasets and also structured/unstructured dataset algorithms on SageMaker platform. Let’s get started!
Images and Videos:
Following flowchart presents ML service options for Images and Video datasets. There are 2 main service options here, AWS Rekognition (application level service) and AWS SageMaker (i.e. AWS’ managed ML platform). As a general guiding principle if you are leveraging AWS ML application level services and need more control, use SageMaker’s out-of-the-box or custom algorithm. As shown in the following chart, you can use ResNet for image classification and object detection if AWS Rekognition doesn’t give you the control you desire.
Speech:
Following flowchart focuses on options for Speech dataset. Similar to above you can fallback on the SageMakers Seq2Seq algo if AWS Transcribe is inadequate for the target use case.
SageMaker ML Algorithms Map:
Finally, I decided to add a mindmap chart for the OOB machine learning algorithms (for both supervised and unsupervised) available on AWS SageMaker platform. Algos are listed as leaf nodes of the tree and are classified as supervised/unsupervised followed by the use case. Also one more key thing to note is you can either leverage OOB supervised or unsupervised algos or bring your own model (BYOM) and train+deploy it on SageMaker.
Summary:
This along with part1 should now cover the full spectrum of options for both types of datasets, unstructured and structured on the AWS ML platform.
What’s next?
I will be putting together a single snapshot picture of all supported ML services on AWS platform across application, platform and infrastructure layers. Stay tuned for part 3!
