AI and ML Week - AWS
Transcript of AI and ML Week - AWS
![Page 1: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/1.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Paolo Emilio Barbano
Sr. Data Scientist
Artificial intelligence and machine learning, AWS
AI and ML WeekForecasting service demand and planning capacity
![Page 2: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/2.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Agenda
• Statistical modeling of time series
• Traditional vs. machine learning-based models
• AWS services including Amazon Forecast
• Demo of Amazon Forecast
![Page 3: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/3.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Applications of forecasting in the public sector
Student demand and
drop-outs
Citizen transportation, web,
and benefits usage
Budget and staffing
loads
Donations and donor
activitiesPopulation health data
and research results
Financial analysis
![Page 4: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/4.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Key challenges with statistical modeling of time series
How do we describe the important features of the time series pattern?
How is the past behavior affecting the future?
How accurately can we predict future values of the series?
![Page 5: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/5.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Three Approaches
Traditional
methods
Deep
learning
(DL)
methods
Combination:
Amazon
Forecast
![Page 6: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/6.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Traditional time-series models
Advantages
• Independent forecasts
• Strong structural assumptions
• De facto industry standard
• Well-understood,
>50 yrs. research
• High data efficiency
Keep in mind…
• Data must match the structural assumptions
• You cannot identify patterns across a time series
Algorithm types
• Nonparametric time series
model
• Exponential smoothing (ETS)
• (Auto-) ARIMA
• Prophet
Limitations
• Metadata can’t be naturally
embedded in the analysis
• External factors are difficult to
include
• Historical data is required
![Page 7: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/7.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Deep learning methods
Deep learning tends to do
well on highly complex,
interpret-able data
Deep learning outperforms
traditional methods in a
variety of cases
Deep learning may be able
to handle cold-start
situations
![Page 8: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/8.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Amazon Forecast
Fully managed
service
Highly accurate Easy to use Your data,
your models
Automatically sets up
data pipeline, training
and prediction
50% improvement in
accuracy over
traditional methods
No deep learning
experience
required
Encrypted with customer
keys through Amazon Key
Management Service
Automated machine learning service for accurate forecasting
![Page 9: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/9.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Amazon Forecast – How it works
Historical DataSales, Inventory, Pricing, etc.
Related DataWeather, Competitive Promotions etc.
Mata-DataColor, City, Country, Author etc.
1. Load Data
2. Inspect Data
3. Identity Features
4. Algorithm Selection
5. Hyper-parameters Selection
6. Model Training
7. Optimize Models
8. Model Deployment and Hosting
Amazon Forecast
Private
Customized
Forecasting
API
![Page 10: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/10.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
There are three types of datasets in Amazon Forecast
Related time-seriesRelated time-series such as
price, web hits, etc.
Target time-seriesHistoric time series data of
items to forecast
Item metadataAttributes of the item such
as category, genre, and
brand.
![Page 11: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/11.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
How Amazon Forecast works
![Page 12: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/12.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Amazon Forecast options
Model Training
Auto-regressive integrated moving average (ARIMA)
Exponential smoothing (ETS)
Non-parametric time series (NPTS)
Prophet
Deep auto-regressive plus (DeepAR+)
Application across multiple Domains
Set your domain from console or via the API
Upload datasets with different schemas based on
the domain
Metric Comparison
Choose hyper parameter tuning
Compare results with different Metrics
![Page 13: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/13.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Amazon Forecast demo
![Page 14: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/14.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Amazon Forecast demo
![Page 15: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/15.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Useful Resources for Amazon Forecast
A description of the functionalities and capabilities is here:
https://docs.aws.amazon.com/forecast/latest/dg/what-is-forecast.html
Also, you can find a full setup and detailed documentation of Forecast here:
https://docs.aws.amazon.com/forecast/latest/dg/forecast.dg.pdf
More advanced users will be able to find fully documented notebooks on this link:
https://github.com/aws-samples/amazon-forecast-samples
![Page 16: AI and ML Week - AWS](https://reader033.fdocuments.net/reader033/viewer/2022041413/6251e6ede0a9dc520d1b92d1/html5/thumbnails/16.jpg)
© 2020, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Thank You!