Webinar - Comparative Analysis of Cloud based Machine Learning Platforms
•
2 gostaram•1,060 visualizações
This webinar discusses cloud based Machine Learning platforms in detail while identifying suitable business use cases for each of them: Microsoft Azure ML, Amazon Machine Learning DataBricks Cloud
Recomendados
Recomendados
Mais conteúdo relacionado
Mais procurados
Mais procurados (20)
Destaque
Destaque (20)
Semelhante a Webinar - Comparative Analysis of Cloud based Machine Learning Platforms
Semelhante a Webinar - Comparative Analysis of Cloud based Machine Learning Platforms (20)
Mais de BigDataCloud
Mais de BigDataCloud (20)
Último
Último (20)
Webinar - Comparative Analysis of Cloud based Machine Learning Platforms
- 3. Disclaimer • Third Eye is a direct vendor to Microsoft, Amazon & Google. • Third Eye has implemented numerous Big Data projects for them over last 3 years. • Third Eye is NOT a reseller of the cloud services of these companies. • Third Eye does NOT financially benefit for making any of the following recommendations. • This work is purely meant for a technical evaluation of the ML platforms and should not be construed for any other purposes.
- 4. ComparisonApproach: What do we look for in an online Machine Learning Platform? Data Preparation • Data Ingestion (out of the box support of data sources) & Data Export • Data Cleaning, Transformation, Visualization Data Selection • Feature selection/engineering Algorithms • Which algorithms are supported out the box? Modify or create new ones? • Saving/comparing results Optimize • E.g. Identify the optimal parameter settings for algorithms Knowledge
- 5. Meet the Contestants • Amazon ML • Azure ML • Databricks Cloud
- 7. Amazon ML • Supported capabilities are described in Use Cases terminology – as opposed to names of algorithms: – Fraud detection – Content Personalization – Document Classification – Customer Churn Prediction – Relevancy modeling for marketing – Recommendations
- 8. Amazon ML • Available algorithms: – Binary and Multiclass Classification – Regression • Limited or no customizability: the algorithms are already implemented and chosen for you :e.g. Binary Regression is implemented via Logistic Regression
- 9. Amazon ML Available Performance Metrics • BinaryAUC: The binary MLModel uses the Area Under the Curve (AUC) technique to measure performance. • RegressionRMSE: The regression MLModel uses the Root Mean Square Error (RMSE) technique to measure performance. RMSE measures the difference between predicted and actual values for a single variable. • MulticlassAvgFScore: The multiclass MLModel uses the F1 score technique to measure performance.
- 10. Amazon ML • Data Ingestion / integration – This is their strongest use case: easy integration with AWS storage media • S 3, EBS • RedShift • RDS
- 11. Azure ML • Introduced February this year • But do not let its relative youthfulness be a distraction: this is a feature rich offering • Has a different approach: a more serious/rich set of algorithms and configurations are made available. • Default / canned algorithms are still available for those newer to Machine Learning
- 13. Azure ML • First class usability: – Tutorials – Walkthroughs – Videos – Integrated Development Environment • Azure ML Studio – Documentation
- 14. Azure ML • Process Tools – Select the data processing, modeling, or prediction activity manually
- 25. Azure ML: More choices • Regression: – Linear, Bayesian, Neural Network , Decision Forest, Boosted Decision Tree, Poisson • Binary Classification – SVM, Percepton, LR, Bayes, NN's, Decision forest • Multiclass : – LR, NN, Decision Forest/Jungle, One vs All • Anomaly Detection: – SVM, PCA • Clustering: Kmeans
- 27. Databricks Cloud • Spark: has joined Hadoop as de-facto industry standards for distributed computing • Rapidly approaching popularity of hadoop – And supplanting it if/when organizations can make the switch • Databricks is the spin-off of Berkeley Amplab –the original creators of Spark • Databricks staff include a large fraction of the Spark core committers • And an even larger proportion of the key decision makers / "shepherds" – Including the spark.ml/mllib shepherds • Cloud based availability of Spark including Spark SQL and spark.ml • Access to capabilities of Spark Mllib, Spark Dataframes/SQL, Streaming, and Resilient Distributed Datasets • Notebooks approach: Scala, Python, Java, and R
- 28. Spark Ecosystem
- 29. Databricks Cloud • The online offering was announced July 2014 at Spark Summit • Purpose statement - Ease of workflow mangement for Data Scientists:
- 31. Databricks Cloud: Notebooks • Notebooks are Data Scientists''' friends • A standard / typically preferred approaching to doing their work – Experiment with data – Perform ad-hoc visualizations – Communicate / share results with colleagues – or even publish them • Wide spectrum of sophistication levels available: – simply use existing libraries – develop new algorithms from scratch
- 33. Wrap-up/Summary Three general types of approaches (not mutually exclusive) Point and Click (as well as backend API's) Amazon ML Azure ML APIs-Only Google Prediction API Notebooks Azure ML Databricks Cloud
- 34. Wrap-up/Summary Amazon ML may be sufficient for: - customers that already have data residing in those providers - simpler/fewer options are acceptable AzureML has a strong usability and workflow approch and provides a reasonable cross section of algorithms available for casual & intermediate users Databricks Cloud has the most comprehensive offering – Variety, performance, configurability of Algorithms – Richness of the capabilities of the Notebooks – Options /configurability of the hosting clusters/environment