CODEWITHSUNDEEP
amazon-web-servicesamazon-sagemaker
DGS
DGS

Reputation: 535

Information regarding Amazon Sagemaker groundtruth

I'm trying to run a simple GroundTruth labeling job with a private workforce for text classification. Since I'm new to AWS GroundTruth, I have some questions:

  1. If I use private workforce what is the maximum number of persons that I can allocate to the labeling job? Does the pricing cost depend on number of persons used in private workforce.

  2. I have a labeled dataset (text classication), and I upload it to S3 bucket, if I upload another unlabeled datas to it, will AutoML label the provided raw data? If not, how can I use already labelled dataset to label new raw datas/

  3. Groundtruth documentation says that it needs atleast 1000 objects to be labeled by humans. Does it mean 1000 objects of all classes, or 1000 objects for individual class? If I manually label 1000+ objects, how many more objects will AutoML label or what is the maximum number of objects can AutoML label?

Upvotes: 0

Views: 1155

Answers (1)

vikmadan
vikmadan

Reputation: 116

I'm the product manager for Amazon SageMaker Ground Truth, and I would be happy to answer your query. Here are my responses:

[1] Your private labeling workforce can be as large or small as you would like it to be. The pricing does not depend on this size of your labeling workforce.

[2] You learn more about how to bring a "partially" labeled dataset here: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-reusing-data.html#sms-reusing-data-newdata

You can also use the ML model trained from a previous labeling job. Learn more here; https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-ground-truth-using-a-pre-trained-model-for-faster-data-labeling/

[3] To clarify, you need 1,000 dataset objects to start an auto-labeling job, but some of these 1,000 objects can be auto-labeled (the % depends on your data and use case). It is 1,000 objects across your classes - i.e. there is no additional requirement beyond having 1,000 text dataset objects.

You can learn more about the mechanics of auto-labeling from this blog post: https://aws.amazon.com/blogs/machine-learning/annotate-data-for-less-with-amazon-sagemaker-ground-truth-and-automated-data-labeling/

Upvotes: 3

Related Questions