Build, Scale, and Deploy Deep Learning Pipelines with Ease
1 like1,392 views
The document discusses the process of building, scaling, and deploying deep learning pipelines using Databricks and Apache Spark, emphasizing ease of integration and performance. It covers topics such as the definition of deep learning, successful applications, and a typical workflow, while highlighting the advantages of using Spark for distributed computation. Additionally, it provides insights into the development of deep learning models with transfer learning and model deployment in SQL.
Build, Scale, and Deploy Deep Learning Pipelines with Ease
- 1. Build, Scale, and Deploy Deep Learning Pipelines with Ease Tim Hunter (Software Engineer) Sue Ann Hong (Software Engineer) Jules S. Damji (Spark Community Evangelist) July 27, 2017
- 2. Agenda • Logistics • Databricks Overview • Build, Scale and Deploy Deep Learning Pipelines with Ease • Q & A
- 3. Logistics • We can’t hear you… • Recording will be available... • Slides will be available... • Queue up Questions …. • Orange Button for Tech Support difficulties...
- 4. TEAM About Databricks Started Spark project (now Apache Spark) at UC Berkeleyin 2009 PRODUCT Unified Analytics Platform MISSION Making Big Data Simple
- 5. Accelerate innovation by unifying data science, engineering and business. Unified Analytics Platform UNIFIED INFRASTRUCTURE UNIFIED EXPERIENCE ACROSS TEAMS UNIFIED ANALYTIC WORKFLOWS
- 6. The Unified Analytics Platform
- 7. About Us • Sue Ann Hong • Software engineer @ Databricks • Ph.D. from CMU in Machine Learning • Contributor to MLlib • Author of Deep Learning Pipelines
- 8. About Us • Tim Hunter • Software engineer @ Databricks • Ph.D. from UC Berkeley in Machine Learning • Very early Spark user • Contributor to MLlib • Author of Deep Learning Pipelines, TensorFrames and GraphFrames
- 9. Build, Scale, and Deploy Deep Learning Pipelines with Ease Tim Hunter (Software Engineer) Sue Ann Hong (Software Engineer) July 27, 2017
- 10. Today • Deep Learning at scale made easy: the vision • Processing images with DL Pipelines • Building simple Deep Learning models with transfer learning • Model deployment via SQL Further advanced topics will be covered in our next webinar.
- 11. Deep Learning with ease
- 12. What is Deep Learning? • A set of machine learning techniques that use layers that transform numerical inputs • Classification • Regression • Arbitrary mapping • Popular in the 80’s as Neural Networks • Recently came back thanks to advances in data collection, computation techniques, and hardware.
- 13. Success of Deep Learning • Tremendous success for applications with complex data • AlphaGo • Image interpretation • Automatictranslation • Speech recognition
- 14. But still requires a lot of effort • Low level APIs with steep learning curve • Tedious to distribute computations • Not well integrated with other enterprise tools • No exact science around deep learning • Success requires many engineer-hours
- 15. Deep Learning in industry • Currently limited adoption • Huge potential beyond the industrial giants • How do we accelerate the road to massive availability?
- 16. A typical Deep Learning workflow • Load data (images, text, time series, …) • Interactive work • Train • Select an architecture for a neural network • Optimize the weights of the NN • Evaluateresults, potentially re-train • Apply: • Pass the data through the NN to produce new features or output
- 17. How can Spark help? • A lot of libraries available for Deep Learning in Spark • TensorFlowOnSpark, BigDL, … • Goes from simple to very advanced • See our previous webinar for more detail • Spark is great at scaling out computations • Distribute the transforms • Manage the trainingcomputation • Spark MLlib Pipelines • Simple, concise APIto capture the ML workflow
- 18. Deep Learning Pipelines: Deep Learning with Simplicity • Open-source Databricks library: https://github.com/databricks/spark-deep-learning • Focuses on easeof useand integration,without sacrificing performance • Scales out common tasks • Integrates with Spark APIs • Primary language: Python
- 19. Deep Learning Pipelines • Load data • Interactive work • Train • Evaluate model • Apply • Image loading in Spark • Deploying models in SQL • Transfer learning • Distributed tuning • Distributed prediction • Pre-trained models This webinar: ✓ ✓ ✓ ✓
- 20. Image processing with DL Pipelines and Databricks
- 21. Adds support for images in Spark • ImageSchema, reader, conversion functions to/from numpy arrays • Most of the tools we’ll describe work on ImageSchema columns from sparkdl import readImages image_df = readImages(sample_img_dir)
- 22. Applying popular models • Popular pre-trained models accessible through MLlib Transformers predictor = DeepImagePredictor(inputCol="image", outputCol="predicted_labels", modelName="InceptionV3") predictions_df = predictor.transform(image_df)
- 23. Applying popular models predictor = DeepImagePredictor(inputCol="image", outputCol="predicted_labels", modelName="InceptionV3") predictions_df = predictor.transform(image_df)
- 24. Fast model training via transfer learning
- 25. Example: Identify the James Bond cars
- 26. DEMO
- 32. SoftMax GIANT PANDA 0.9 RED PANDA 0.05 RACCOON 0.01 … Classifier Transfer Learning DeepImageFeaturizer
- 33. MLlib primer • MLlib: the machine learning library included with Spark • Transformer • Transforms the data: takes a Spark dataframe and appends a new column • Estimator • Produces a model (fit) • Pipeline: sequence of transformers and estimators
- 34. Transfer Learning as a Pipeline MLlib Pipeline Image Loading Preprocessing Logistic Regression DeepImageFeaturizer
- 35. DEMO
- 36. Sharing and exporting Deep Learning models
- 37. Classifier Deep Learning Model Model Export and Sharing
- 38. Shipping predictors in SQL Take a trained model / Pipeline, register a SQL UDF usable by anyone in the organization In Spark SQL: registerKerasUDF(”my_object_recognition_function", keras_model_file="/mymodels/007model.h5") select image, my_object_recognition_function(image) as objects from traffic_imgs
- 39. Conclusion
- 40. Deep Learning without Deep Pockets • Simple API for Deep Learning, integrated with MLlib • Scales common tasks with transformers and estimators • Embeds Deep Learning models in MLlib and SparkSQL • Early release of Deep Learning Pipelines https://github.com/databricks/spark-deep-learning
- 41. Deep Learning Pipelines - future In progress • Hyper-parameter tuning for Keras models • Official image support in Spark Potential future work • Scala API • Text models • Support for more backends, e.g. MXNet, PyTorch, BigDL
- 42. Resources Blog posts & webinars — http://databricks.com/blog • Deep Learning Pipelines • GPU acceleration in Databricks • BigDL on Databricks • Deep Learning and Apache Spark Docs for Deep Learning on Databricks — http://docs.databricks.com • Getting started • Deep Learning Pipelines Example • Spark integration
- 43. Thank You! Questions? Happy Sparking & Deep Learning!
- 44. UNIFIED ANALYTICS PLATFORM Try Apache Spark in Databricks! • Collaborative cloud environment • Free version (community edition) DATABRICKS RUNTIME 3.0 • Apache Spark - optimized for the cloud • Caching and optimization layer - DBIO • Enterprise security - DBES Try for free today databricks.com