Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Azure AutoML

Azure AutoML

Achieving state-of-the-art performance in a given data set is hard. It usually implies carefully selecting the right data pre-processing tasks, picking the algorithm, model, and architecture and pairing it with the right set of parameters. This end-to-end process is usually called Machine Learning Pipeline. Isn't tough and time-consuming?

Automated ML helps create a high-quality model using intelligent automation and optimization. You launch the automated machine learning process by establishing goals and constraints. Then allow the algorithm selection and hyperparameter tuning to happen for you. The AutoML technique iterates over many combinations of algorithms and hyperparameters until it finds the best model based on your criterion.

In this session, we'll see how to set up the environment, configure an Azure Machine Learning service workspace and explore the results

Sivamuthu Kumar

April 27, 2019
Tweet

More Decks by Sivamuthu Kumar

Other Decks in Technology

Transcript

  1. Azure AutoML LEARNING THE
    LEARNING

    View full-size slide

  2. Agenda
    ML – Overview
    Azure Machine Learning
    What is Automated ML?
    Build a ML Model using Automated ML
    Demo

    View full-size slide

  3. Hello !!
    I’m Siva
    Working as Architect in CEI
    #Cloud #Mobile #IoT Solutions
    ksivamuthu
    ksivamuthu

    View full-size slide

  4. Machine Learning … Everywhere …
    ML Overview

    View full-size slide

  5. Artificial Intelligence
    Science of getting machines to do the things what they do in movies. (Mimic human behavior)
    Machine Learning
    Subset of AI - The science of getting computers to act without being explicitly programmed.
    Deep Learning
    Subset of ML - Learning based on deep neural network.

    View full-size slide

  6. Machine Learning
    Using known data, develop a
    model to predict unknown data

    View full-size slide

  7. Machine learns the same way human do.

    View full-size slide

  8. Let’s do an exercise on
    human learning to
    understand machine
    learning.
    @ksivamuthu

    View full-size slide

  9. Virzeth
    https://medium.com/@yakubova92/machine-learning-vs-human-learning-f3f204c8b27d
    @ksivamuthu

    View full-size slide

  10. Virzeth or Not ?
    @ksivamuthu

    View full-size slide

  11. Types of Machine Learning
    Supervised Learning Reinforcement
    Learning
    Train an algorithm to perform
    classification and regression with
    labelled data set.
    Unsupervised
    Learning
    Train an algorithm to find clusters and
    associations in an unlabeled data set.
    Train an agent to take certain actions
    in an environment without a data set.

    View full-size slide

  12. Energy Auto Finance Entertainment
    Media Manufacturing /
    Agriculture
    Retail

    View full-size slide

  13. ML Workflow
    Gathering Data Preparing Data Choosing a model
    Training Evaluation Hyper parameter
    tuning
    Prediction

    View full-size slide

  14. Data
    preprocessing
    ML Model Design
    Tune model
    parameters
    Evaluate Deploy Update
    What preprocessing techniques
    should I use?
    What modeling techniques?
    Which architecture?
    Which set of hyperparameters?
    How do I know my model’s
    performance?
    How can I improve my dataset?
    Which metrics are most important?
    What infrastructure to use to serve
    my model at scale?
    How to meet latency goals?
    ML Workflow

    View full-size slide

  15. Azure Machine
    Learning
    Service

    View full-size slide

  16. Sophisticated pretrained models
    To simplify solution development
    Azure
    Databricks
    Machine Learning
    VMs
    Popular frameworks
    To build advanced deep learning solutions
    TensorFlow Keras
    Pytorch Onnx
    Azure
    Machine Learning
    Language
    Speech

    Azure
    Search
    Vision
    On-premises Cloud Edge
    Productive services
    To empower data science and development teams
    Powerful infrastructure
    To accelerate deep learning
    Flexible deployment
    To deploy and manage models on intelligent cloud and edge
    Machine Learning on Azure
    Cognitive Services

    View full-size slide

  17. Creating ML Workspace

    View full-size slide

  18. Azure ML Workspace Resources

    View full-size slide

  19. ML
    EXPERIMENTS DATASTORE COMPUTE
    MODELS IMAGES DEPLOYMENTS

    View full-size slide

  20. Automated ML

    View full-size slide

  21. • Based on Microsoft
    Research
    • Brain trained with several
    million experiments
    • Collaborative filtering and
    Bayesian optimization
    • Privacy preserving: No need
    to “see” the data
    Automated ML – How it works

    View full-size slide

  22. Build a ML Model
    using AutoML

    View full-size slide

  23. Type of machine
    learning problem
    you are solving …

    View full-size slide

  24. Categories of
    supervised
    learning
    supported
    CLASSIFICATION REGRESSION
    FORECASTING

    View full-size slide

  25. Data
    preparation &
    Compute Target
    • Source and format of training data
    • Data can be read into Numpy arrays or a
    Pandas data frame
    • Configure split options for selecting
    training and validation data
    • You can specify separate training and
    validation datasets.
    • Configure compute target

    View full-size slide

  26. Configure
    your
    experiment
    settings
    automl_classifier = AutoMLConfig(
    task='classification’,
    primary_metric='AUC_weighted’,
    max_time_sec=12000,
    iterations=50,
    X=X,
    y=y,
    n_cross_validations=2
    )

    View full-size slide

  27. Primary Metric
    Classification Regression Time Series Forecasting
    accuracy spearman_correlation spearman_correlation
    AUC_weighted normalized_root_mean_squared
    _error
    normalized_root_mean_squared
    _error
    average_precision_score_weighte
    d
    r2_score r2_score
    norm_macro_recall normalized_mean_absolute_err
    or
    normalized_mean_absolute_err
    or
    precision_score_weighted

    View full-size slide

  28. Run
    Experiment run = experiment.submit(automl_config)

    View full-size slide

  29. Exit Criteria
    No Criteria
    Number of iterations / Iteration Timeout
    Minutes
    Exit after a length of time / Experiment
    Timeout Minute.
    Exit after a score has been reached –
    Experiment Exit Score

    View full-size slide

  30. MODEL CREATION TIME –
    FROM DAYS TO HOURS
    ROBUST BENCHMARKING
    PROCESS FOR ML PROJECTS
    ENABLE DOMAIN EXPERTS
    TO LEVERAGE ML

    View full-size slide

  31. Automated ML in
    Power BI

    View full-size slide

  32. Reference
    • Azure - https://azure.microsoft.com/en-
    us/services/machine-learning-service/
    • Microsoft Machine Learning Blog -
    https://azure.microsoft.com/en-us/blog/tag/azure-
    machine-learning/
    • Azure ML documentation -
    https://docs.microsoft.com/en-us/azure/machine-
    learning/
    • Slides - https://speakerdeck.com/ksivamuthu/azure-
    automl-learning-the-learning

    View full-size slide

  33. Thank You !! Follow me on social medias
    ksivamuthu
    ksivamuthu

    View full-size slide