Downloaded 12 times
![Building a Smarter Application • Q: What is a smarter application? • A: An app that learns from data [From rules-based to model-based]](https://image.slidesharecdn.com/pojowebapptk20150917-150918184306-lva1-app6891/75/Design-Patterns-for-Smart-Applications-3-2048.jpg)
Uploaded bySri Ambati
Design Patterns for Smart Applications
The document discusses building a smarter web application that incorporates machine learning using an H2O-generated POJO model. It outlines the steps for model development and integration, including question selection, data utilization, model exporting, and the use of a REST API for predictions. It also highlights resource links and addresses challenges such as scoring, storing predictions, and model update frequency.
Design Patterns for Smart Applications
- 1. Building a SmarterWebApp with a Generated POJO Model H2O.ai Tom Kraljevic September 17, 2015
- 2. Target Audience • JavaScriptWebApp developer who wants to incorporate machine learning • More code, less PowerPoint
- 3. Building a SmarterApplication • Q: What is a smarter application? • A: An app that learns from data [From rules-based to model-based]
- 4. Software Pieces • Front-end –Web browser – JavaScript application (run in the browser) • Back-end – Jetty servlet container – H2O-generated model POJO (hosted by servlet container) • Out-of-band – H2O
- 5. Overview Step 1: Pickingthe question you want your model to answer Step 2: Using your data to build a model Step 3: Exporting the generated model as a Java POJO Step 4: Compiling the model Step 5: Hosting the model in a servlet container Step 6: Running the JavaScript app in a browser Step 7: Using a REST API to make predictions Step 8: Incorporating the prediction into your application
- 6. Resources • This meetup –https://github.com/h2oai/h2o- meetups/tree/master/2015_09_17_PojoWebApp • H2O-3 Slater Release – http://h2o-release.s3.amazonaws.com/h2o/rel- slater/1/index.html • Generated POJO Model Javadoc – http://h2o-release.s3.amazonaws.com/h2o/rel- slater/1/docs-website/h2o-genmodel/javadoc/index.html • POJO REST API Example – https://github.com/h2oai/h2o-world-2015- training/tree/master/tutorials/pojo_webapp
- 7.
- 8. Next Steps fora Real Use Case • Scoring (judging how good the predictions really are) – Need to get the correct answers from somewhere • Storing predictions (and the correct answers) – Often Hadoop. Seems easy, but this can be real work to organize • Model update frequency – From the stored data above – Need depends on the problem – Hourly, daily, monthly? – Cost of training the model becomes a factor • Hot swapping the model – Separating front-end and back-end makes this easier – Java reflection for in-process hot-swap – Load balancer for servlet container hot-swap
Editor's Notes
- #4 An app that makes predictions using a model to affect program behavior (and subsequently affect user behavior!)