Taking Models to the Next Level With Azure Machine Learning

Artificial Intelligence and Machine Learning can be used in many ways to increase productivity of business processes and gather meaningful insights, by analyzing images, texts and trends within unstructured flows of data. While many tasks can be solved using existing models, in some cases it is also required to train your own model for more specific tasks, or for increased accuracy. In this session, we will explore the complete path of integrating text analysis intelligent services into the business processes of Tailwind Traders, starting from pre-build models available as cognitive services, up to training a third-party neural custom model for Aspect-Based Sentiment Analysis available as part of Intel NLP Architect using Azure Machine Learning Service. We will talk about cases when one needs a custom model and show how to fine-tune model hyperparameters using HyperDrive

Transcript

1. aka.ms/AIML40 Henk Boelman (@hboelman)
2. aka.ms/AIML40 Henk Boelman (@hboelman) In this session
3. aka.ms/AIML40 Henk Boelman (@hboelman) The Story Existing pre-built models for text analytics Custom models trained on specific data
4. aka.ms/AIML40 Henk Boelman (@hboelman)
5. aka.ms/AIML40 Henk Boelman (@hboelman)
6. aka.ms/AIML40 Henk Boelman (@hboelman) /Docs alert aka.ms/TextCogSvc Get started with Azure Cognitive Services Text Analytics API
7. aka.ms/AIML40 Henk Boelman (@hboelman) From Cognitive Services to Azure Machine Learning
8. aka.ms/AIML40 Henk Boelman (@hboelman)
9. aka.ms/AIML40 Henk Boelman (@hboelman)
10. aka.ms/AIML40 Henk Boelman (@hboelman) What is Azure Machine Learning? Set of Azure Cloud Services Azure CLI Extension ✓Prepare Data ✓Build Models ✓Train Models ✓Manage Models ✓Track Experiments ✓Deploy Models That enables you to: Python SDK Web interface
11. aka.ms/AIML40 Henk Boelman (@hboelman) Advanced Azure ML Services
12. aka.ms/AIML40 Henk Boelman (@hboelman) NLP-Architect Aspect Based Sentiment Analysis
13. aka.ms/AIML40 Henk Boelman (@hboelman) Managing ML Challenges
14. aka.ms/AIML40 Henk Boelman (@hboelman) Challenges of distributed training  Dependencies and Containers Provision clusters of VMs Schedule jobs Distribute data Gather results Handle failures Scale resources Secure Access
15. aka.ms/AIML40 Henk Boelman (@hboelman) Azure Machine Learning Typical end to end ML process … Prepare Experiment Deploy Orchestrate
16. aka.ms/AIML40 Henk Boelman (@hboelman) Prepare your environment Experiment with your model & data Deploy your model
17. aka.ms/AIML40 Henk Boelman (@hboelman) Step 1: Prepare your environment
18. aka.ms/AIML40 Henk Boelman (@hboelman) Create a workspace ws = Workspace.create( name='<NAME>', subscription_id='<SUBSCRIPTION ID>', resource_group='<RESOURCE GROUP>', location='westeurope') ws.write_config() ws = Workspace.from_config() Create a workspace
19. aka.ms/AIML40 Henk Boelman (@hboelman)
20. aka.ms/AIML40 Henk Boelman (@hboelman) Pipelines – Training workflows Pipeline Endpoints – Training workflow endpoints Azure Machine Learning
21. aka.ms/AIML40 Henk Boelman (@hboelman) Create Compute cfg = AmlCompute.provisioning_configuration( vm_size='STANDARD_NC6', min_nodes=0, max_nodes=6) cc = ComputeTarget.create(ws, '<NAME>', cfg) Create a workspace Create compute
22. aka.ms/AIML40 Henk Boelman (@hboelman) Attach storage ds = ws.get_default_datastore() ds = Datastore.register_azure_blob_container( workspace = ws, datastore_name = ‘<datastore name>', container_name = ‘<blob container name>', account_name = ‘<storage account name>’, account_key = ‘<storage account name>', Create a workspace Create compute Setup storage
23. aka.ms/AIML40 Henk Boelman (@hboelman) Step 1 Prepare your environment Create a workspace Create compute Setup storage
24. aka.ms/AIML40 Henk Boelman (@hboelman) Step 2: Experiment with your model & data
25. aka.ms/AIML40 Henk Boelman (@hboelman) Create an experiment exp = Experiment(workspace=ws, name=“<ExperimentName>”) Create an Experiment
26. aka.ms/AIML40 Henk Boelman (@hboelman) Create a training file Create an Experiment Create a training file
27. aka.ms/AIML40 Henk Boelman (@hboelman) Create an estimator params = {'--data-folder': ws.get_default_datastore().as_mount()} estimator = Estimator( source_directory = script_folder, script_params = params, compute_target = computeCluster, entry_script = 'train.py’, conda_packages = ['scikit-learn','keras','opencv’]) Create an Experiment Create a training file Create an estimator
28. aka.ms/AIML40 Henk Boelman (@hboelman) Submit the experiment to the cluster run = exp.submit(estimator) RunDetails(run).show() Create an Experiment Create a training file Submit to the AI cluster Create an estimator
29. aka.ms/AIML40 Henk Boelman (@hboelman) Create an Experiment Create a training file Submit to the AI cluster Create an estimator Demo: Creating and run an experiment
30. aka.ms/AIML40 Henk Boelman (@hboelman) Notebook VM Compute Target Experiment Docker Image Data store 1. Snapshot folder and send to experiment 2. create docker image 3. Deploy docker and snapshot to compute 4. Mount datastore to compute 6. Stream stdout, logs, metrics 5. Launch the script 7. Copy over outputs
31. aka.ms/AIML40 Henk Boelman (@hboelman) Register the model model = run.register_model( model_name='MyAwesomeModel', model_path='outputs') Create an Experiment Create a training file Submit to the AI cluster Create an estimator Register the model
32. aka.ms/AIML40 Henk Boelman (@hboelman) Create an Experiment Create a training file Submit to the AI cluster Create an estimator Register the model Demo: Register and test the model
33. aka.ms/AIML40 Henk Boelman (@hboelman) Step 3: Deploy your model
34. aka.ms/AIML40 Henk Boelman (@hboelman) AMLS to deploy The Model Score.py Environment file Docker Image
35. aka.ms/AIML40 Henk Boelman (@hboelman) Score.py
36. aka.ms/AIML40 Henk Boelman (@hboelman) Environment File
37. aka.ms/AIML40 Henk Boelman (@hboelman) Inference config python score.py myenv.yml
38. aka.ms/AIML40 Henk Boelman (@hboelman) Deployment using AML
39. aka.ms/AIML40 Henk Boelman (@hboelman) Deploy to ACI ignitett-aci model inference_config aci_deployment_config
40. aka.ms/AIML40 Henk Boelman (@hboelman) Deploy to AKS ignitett-aks model inference_config deployment_config aks_target
41. aka.ms/AIML40 Henk Boelman (@hboelman) Demo: Deploy to ACI & AKS
42. aka.ms/AIML40 Henk Boelman (@hboelman) Distributed Hyperparameter Tuning
43. aka.ms/AIML40 Henk Boelman (@hboelman) What are Hyperparameters? { “learning_rate”: uniform(0, 1), “num_layers”: choice(2, 4, 8) … } Config1= {“learning_rate”: 0.2, “num_layers”: 2, …} Config2= {“learning_rate”: 0.5, “num_layers”: 4, …} Config3= {“learning_rate”: 0.9, “num_layers”: 8, …} …
44. aka.ms/AIML40 Henk Boelman (@hboelman) Typical ‘manual’ approach to hyperparameter tuning Dataset Training Algorithm 1 Hyperparameter Values – config 1 Model 1 Hyperparameter Values – config 2 Model 2 Hyperparameter Values – config 3 Model 3 Model Training Infrastructure Training Algorithm 2 Hyperparameter Values – config 4 Model 4 Complex Tedious Repetitive Time consuming Expensive
45. aka.ms/AIML40 Henk Boelman (@hboelman) Automated Hyperparameter Tuning Manage Active Jobs Evaluate training runs for specified primary metric Early terminate poor performing training runs. Early termination policies include: Early terminate poorly performing runs
46. aka.ms/AIML40 Henk Boelman (@hboelman) Demo: hyperparameter with Azure Machine Learning
47. aka.ms/AIML40 Henk Boelman (@hboelman) Resources aka.ms/AIML40repo aka.ms/AIML40 aka.ms/mymsignitethetour aka.ms/9TipsProdML