AWS developers who want to extract meaning and forecasts out of their data now no longer need other hosted machine learning platforms.
Public cloud giant Amazon Web Services (AWS) on Thursday unveiled its first product for machine learning - simply called Amazon Machine Learning - to make it easier for AWS developers to extract value from the troves of transactional and operational data their hosted systems collect.
The move by Amazon follows IBM's recent launch of hosted Watson Analytics and Microsoft's Azure Machine Learning, with all three now looking for ways to help developers embed machine learning intelligence into their apps. Google's own machine learning offering, Prediction API, was launched in 2012.
Much like AWS itself, Amazon Machine Learning emerged from technology that Amazon's internal data scientists used to create machine learning models that help it ferret out useful patterns from its data and ultimately make sharper predictions. However, AWS wants Amazon Machine Learning to be useful even for those without a degree in statistics.
Using AWS Machine Leaning APIs, developers can create new models derived from data stored in Amazon S3, Amazon Redshift, or MySQL databases in Amazon RDS.
Charges for the service come on top of existing services used in AWS and are available as two prediction types: batch predictions for applications that need a bunch of predictions all at one; and real-time predictions that allow apps to request predictions on demand.
For batch predictions, its charging 10 cents per 1,000 predictions, while real-time predictions, based on memory required, are priced at $0.0001 per prediction.
The batch prediction pricing example AWS offers is includes compute fees and prediction fees. So at 10 cents per 1,000 predictions, if a developer wanted to run 890,000 predictions in a month that would cost $89. They'd pay additional compute fees based on the number of hours required.
Similarly, a real-time prediction bill includes compute fees, monthly prediction fees based on $0.0001 per prediction but it also comes with a reserve capacity charge at $0.001 per 10MB per hour.
Amazon says that its platform supports three types of predictions to answer certain types of questions; however, all require a properly 'trained' model that's been fed with existing data:
- Binary classification is used to predict one of two possible outcomes. Is this transaction legitimate, will the customer buy this product, or is the shipping address an apartment complex?
- Multiclass classification is used to predict one of three or more possible outcomes and the likelihood of each one. Is this product a book, a movie, or an article of clothing? Is this movie a comedy, a documentary, or a thriller? Which category of products is of most interest to this customer?"
- Regression is used to predict a number. How many 27″ monitors should we place in inventory? How much should we charge for them? What percent of them are likely to be sold as gifts?
Read more on machine learning