Revolutions

Forecasting Best Practices, from Microsoft

Microsoft has released a GitHub repository to share best practices for time series forecasting. From the repo:

Time series forecasting is one of the most important topics in data science. Almost every business needs to predict the future in order to make better decisions and allocate resources more effectively.

This repository provides examples and best practice guidelines for building forecasting solutions. The goal of this repository is to build a comprehensive set of tools and examples that leverage recent advances in forecasting algorithms to build solutions and operationalize them. Rather than creating implementations from scratch, we draw from existing state-of-the-art libraries and build additional utilities around processing and featurizing the data, optimizing and evaluating models, and scaling up to the cloud.

The repository includes detailed examples of various time series modeling techniques, as Jupyter Notebooks for Python, and R Markdown documents for R. It also includes Python notebooks to fit time series models in the Azure Machine Learning service, and then operationalize the forecasts as a web service.

The R examples demonstrate several techniques for forecasting time series, specifically data on refrigerated orange juice sales from 83 stores (sourced from the the bayesm package). The forecasting techniques vary (mean forecasting with interpolation, ARIMA, exponential smoothing, and additive models), but all make extensive use of the tidyverts suite of packages, which provides "tidy time series forecasting for R". The forecasting methods themselves are explained in detail in the book (readable online) Forecasting: Principles and Practice by Rob J Hyndman and George Athanasopoulos (Monash University).

You can try out the examples yourself by cloning the repository and knitting the RMarkdown files in R. If you have git installed, a quick and easy way to do this in with RStudio. Choose File > New Project > Version Control > Git, and enter https://github.com/microsoft/forecasting in the Repository URL field. (You might prefer to fork the repository first.)

Open each .Rmd file in turn, accept the prompt to install packages, and click the Knit button to generate the document. The computations can take a while (particularly the Prophet Models example), but if you have a multi-core machine the notebooks do use the parallel package to speed things up. If you don't want to wait, the repository does include HTML versions of the rendered documents, made available at the links below via GitHub Pages:

• Data Preparation
• Basic Models
• ARIMA-Regression models
• Prophet Models

This repository will be updated over time, and contributions are welcome as pull requests to the repository linked below.

GitHub (Microsoft): Forecasting Best Practices

Azure AI and Machine Learning talk series

At last week's Microsoft Ignite conference in Orlando, our team delivered a series of 6 talks about AI and machine learning applications with Azure. The videos from each talk are linked below, and you can watch every talk from the conference online (no registration necessary). Each of our talks also comes with a companion Github repository, where you can find all of the code and scripts behind the demonstrations, so you can deploy and run them yourself.

If you'd like to see these talks live, they will also be presented in 31 cities around the world over the next six months, starting with Paris this week. Check the website for Microsoft Ignite the Tour for event dates and further information.

AIML10 Making sense of your unstructured data with AI

Tailwind Traders has a lot of legacy data that they’d like their developers to leverage in their apps – from various sources, both structured and unstructured, and including images, forms, and several others. In this session, learn how the team used Azure Cognitive Search to make sense of this data in a short amount of time and with amazing success. We discuss tons of AI concepts, like the ingest-enrich-explore pattern, search skillsets, cognitive skills, natural language processing, computer vision, and beyond.

Github repository here. Video here.

AIML20 Using pre-built AI to solve business challenges

As a data-driven company, Tailwind Traders understands the importance of using artificial intelligence to improve business processes and delight customers. Before investing in an AI team, their existing developers were able to demonstrate some quick wins using pre-built AI technologies.

In this session, we show how you can use Azure Cognitive Services to extract insights from retail data and go into the neural networks behind computer vision. Learn how it works and how to augment the pre-built AI with your own images for custom image recognition applications.

Github repository here. Video here.

AIML21 Developers guide to AI: A data story

In this theater session, we show the data science process and how to apply it. From exploration of datasets to deployment of services - all applied to an interesting data story. We also take you on a very brief tour of the Azure AI platform.

Github repository here. Video here.

AIML30: Start building machine learning models faster than you think

Tailwind Traders uses custom machine learning models to fix their inventory issues – without changing their software development life cycle! How? Azure Machine Learning Visual Interface.

In this session, learn the data science process that Tailwind Traders’ uses and get an introduction to Azure Machine Learning Visual Interface. See how to find, import, and prepare data, select a machine learning algorithm, train and test the model, and deploy a complete model to an API. Get the tips, best practices, and resources you and your development team need to continue your machine learning journey, build your first model, and more.

Github repository here. Video here.

AIML40 Taking models to the next level with Azure Machine Learning best practices

Tailwind Traders’ data science team uses natural language processing (NLP), and recently discovered how to fine tune and build a baseline models with Automated ML.

In this session, learn what Automated ML is and why it’s so powerful, then dive into how to improve upon baseline models using examples from the NLP best practices repository. We highlight Azure Machine Learning key features and how you can apply them to your organization, including: low priority compute instances, distributed training with auto scale, hyperparameter optimization, collaboration, logging, and deployment.

Github repository here. Video here.

AIML50 Machine learning operations: Applying DevOps to data science

Many companies have adopted DevOps practices to improve their software delivery, but these same techniques are rarely applied to machine learning projects. Collaboration between developers and data scientists can be limited and deploying models to production in a consistent, trustworthy way is often a pipe dream.

In this session, learn how Tailwind Traders applied DevOps practices to their machine learning projects using Azure DevOps and Azure Machine Learning Service. We show automated training, scoring, and storage of versioned models, wrap the models in Docker containers, and deploy them to Azure Container Instances or Azure Kubernetes Service. We even collect continuous feedback on model behavior so we know when to retrain.

Github repository here. Video here.

New R Support in Azure Machine Learning

Azure Machine Learning has added support for the R language, it was announced at the Ignite conference in Orlando this week.

A new R package azuremlsdk (available to install from Github now, and from CRAN soon), provides the interface to the Azure Machine Learning service. With R functions, you can provision new computing clusters in Azure, and use those to train models with R and deploy them as prediction endpoints for use from any app. You can also launch R-based notebooks in the new Azure Machine Learning studio web interface, or even launch a complete RStudio server instance on your cloud computing resources. Azure Machine Learning service supports the latest version of R (3.6.1) and all R packages (from CRAN, Github, or elsewhere). The video below from The AI Show demonstrates how it all works:

Azure Machine Learning is also great for teams that have both Python and R expertise. You can even call Python models from R (and vice-versa): in this Ignite 2019 talk (presented by me and Daniel Schneider) we deploy R and Python function as a container services, and call them both from a Shiny app. You can also find the slides and associated code from that talk in this Github repository.

To get started with R in Azure Machine Learning, a good place to start is the tutorial "Train and deploy your first model in R with Azure Machine Learning". If you need an Azure subscription, use this link to sign up for Azure and get $200 in free Azure credits.

Azure Machine Learning: ml.azure.com

Microsoft ML Server 9.4 now available

Microsoft Machine Learning Server, the enhanced deployment platform for R and Python applications, has been updated to version 9.4. This update includes the open source R 3.5.2 and Python 3.7.1 engines, and supports integration with Spark 2.4. Microsoft ML Server also includes specialized R packages and Python modules focused on application deployment, scalable machine learning, and integration with SQL Server.

Microsoft Machine Learning Server is used by organizations that need to use R and/or Python code in production applications. For some examples of deployments, take a look at these open-source solution templates for credit risk estimation, energy demand forecasting, fraud detection and many other applications.

Microsoft ML Server 9.4 is available now. For more details on this update, take a look at the announcement at the link below.

SQL Server Blog: Microsoft Machine Learning Server 9.4 is now available 

AI, Machine Learning and Data Science Roundup: February 2019

A monthly roundup of news about Artificial Intelligence, Machine Learning and Data Science. This is an eclectic collection of interesting blog posts, software announcements and data applications from Microsoft and elsewhere that I've noted over the past month or so.

Open Source AI, ML & Data Science News

ONNX, the open interchange format for AI models, updates to version 1.4.0 with support for models larger than 2Gb.

PT-BERT, a PyTorch implementation of Google's BERT language representation model, adds new pre-trained language models: GPT and Transformer-XL.

Stanford University has released StanfordNLP, a natural language analysis package for Python with pre-trained models for 53 languages.

The Python Foundation releases Python 3.7 on the Windows 10 App Store.

Industry News

Amazon offers recommendations to policymakers on the use of facial recognition technology and calls for regulation of its use.

AWS open-sources the Neo-AI project, a machine learning compiler and runtime that tunes Tensorflow, PyTorch, ONNX, MXNet and XGBoost models for performance on edge devices. 

Google introduces FEAST, an open-source "feature store" for managing and discovering features in machine learning models.

Google releases Natural Questions, a corpus of 300,000 questions and human-annotated answers, for question-answering research.

Uber open-sources Ludwig, a platform for training and testing deep learning models without programming, simply by specifying input and output variables in data.

Microsoft News

Microsoft has joined the SciKit-learn Consortium as a Platinum member.

Azure Data Explorer, a high-performance service for real-time analysis of streaming data, is now generally available.

HDInsight Tools for VSCode, providing a lightweight code editor for HDInsight PySpark and Hive batch jobs, is now available.

Azure Cognitive Services can now be used with a unified API key, and in more regions and with new certifications including HIPAA BAA.

An in-depth InfoWorld review of Azure ML services.

Learning resources

Papers with Code, a repository of machine learning papers with associated implementation code and evaluation metrics.

Practical Deep Learning for Coders, the popular free course from fast.ai, has been updated with all-new content. Microsoft Azure and GCP provide pre-configured environments with the required tools, data and materials.

Recommenders Hub: an open source GitHub repository containing Microsoft’s best practices and state-of-the-art algorithms for building recommendation systems.

Presentation videos and workshop materials from the recent RStudio conference are available.

Tutorial: running trained deep learning models on Apple devices, by converting CoreML exported from PyTorch into ONNX.

Tutorials from the MIT Deep Learning courses: Deep learning basics, Driving scene segmentation, DeepTraffic deep reinforcement learning.

Reference Architecture: Batch scoring of Spark models on Azure Databricks.

Applications

Facebook develops a predictive model for global electricity availability, publishes the output and some code.

The British Broadcasting Corporation releases an R package and a guide for data journalists on creating data visualizations in the BBC News style.

TWIMLAI's podcast series on Microsoft's AI for Earth , AI for Accessibility and AI for Humanitarian Action programs.

A profile of Heather Dowdy, the head of Microsoft's AI for Accessibility program.

Find previous editions of the monthly AI roundup here.

AI, Machine Learning and Data Science Roundup: January 2019

A monthly roundup of news about Artificial Intelligence, Machine Learning and Data Science. This is an eclectic collection of interesting blog posts, software announcements and data applications from Microsoft and elsewhere that I've noted over the past month or so.

Open Source AI, ML & Data Science News

Preview of Tensorflow 2.0 (the public preview is expected "early this year").

pcLasso, an R package implementing a new method for supervised learning described by co-author Rob Tibshirani as "principal components regression meets the lasso".

R 3.5.2 has been released.

Industry News

A retrospective of Google Research activities in AI in 2018.

Google Cloud Platform now supports R jobs (via SparkR) in Cloud DataProc.

GCP App Engine support for Python 3.7 now generally available.

Amazon SageMaker now comes preconfigured with support for SciKit-learn.

Microsoft News

Microsoft Professional Program for Data Analysis, a new online course and certification. Other MPP tracks include Data Science and Artificial Intelligence.

[video] A single API key can now be used to access Cognitive Services APIs for language, vision and search.

AzureR, a suite of packages for interfacing with storage, virtual machines, containers and other Azure services from the R language, is now available on CRAN.

Learning resources

E-book by TWIML AI host Sam Charrington: Kubernetes for Machine Learning, Deep Learning and AI (requires free sign-up).

E-book by Patrick Hall and Navdeep Gill: Introduction to Machine Learning Interpretability (requires free registration).

An in-depth introduction to convolutional neural networks, from Ars Technica: How computers got shockingly good at recognizing images.

An on-line course from Databricks, Deep Learning Fundamental Series, with a focus on Keras and TensorFlow.

Getting Started with TensorFlow Probability, from R, a blog post from RStudio.

What can Neural Networks Learn?, an approachable look at the inner workings of neural network classifiers from Brandon Rohrer.

A blog post describing the computational graph concepts behind TensorFlow, including an illustrative implementation of core TensorFlow operations in numpy.

Applications

Using computer vision to monitor shelf stocking policies, a three-part series on complex image classification: Part 1, Part 2, Part 3.

[video] Building a pet breed identification application via transfer learning with Azure ML Service and Python. (Github repo here.)

Find previous editions of the monthly AI roundup here.

Notebooks from the Practical AI Workshop

Last month, I delivered the one-day workshop Practical AI for the Working Software Engineer at the Artificial Intelligence Live conference in Orlando. As the title suggests, the workshop was aimed at developers, bu I didn't assume any particular programming language background. In addition to the lecture slides, the workshop was delivered as a series of Jupyter notebooks. I ran them using Azure Notebooks (which meant the participants had nothing to install and very little to set up), but you can run them in any Jupyter environment you like, as long as it has access to R and Python. You can download the notebooks and slides from this Github repository (and feedback is welcome there, too). 

The workshop was divided into five sections, each with its associated Notebook:

1. The AI behind Seeing AI. Use the web interfaces to Cognitive Services to learn about the AI services behind the "Seeing AI" app
2. Computer Vision API with R. Use an R script to interact with the Computer Vision API and generate captions for random Wikimedia images.
3. Custom Vision with R. An R function to classify an image as a "Hot Dog" or "Not Hot Dog", using the Custom Vision service.
4. MNIST with scikit-learn. Use sckikit-learn to build a digit recognizer for the MNIST data using a regression model.
5. MNIST with Tensorflow. Use Tensorflow (from Python) to build a digit recognizer for the MNIST data using a convolutional neural network.

The workshop was a practical version of a talk I also gave at AI Live, "Getting Started with Deep Learning", and I've embedded those slides below. 

Azure Notebooks: Practical AI for the Working Software Engineer

Gender Diversity in the R and Python Communities

Many (if not most) tech communities have far more representation from men than from women (and even fewer from nonbinary folk). This is a shame, because everybody uses software, and these projects would self-evidently benefit from the talent and expertise from across the entire community. Some projects are doing better than others, though, and data scientist Reshama Shaikh recently published an in-depth comparison of the representation of women in the R any Python communities.

Shaikh's analysis draws from several data sources, which provide evidence that women are better represented in the R community in the Python community. These include:

• The R-Ladies community has 29,500 members compared to PyLadies' 36,500, despite the Python community being 6x larger overall.
• A 2016 study of GitHub contributors estimates R contributors are 9.3% women, and 2.0% for Python contributors.
• In a 2017 R Consortium survey of R users, 14% of respondents identified as women.
• The 2018 New York R conference had 45% women speakers; the 2016 useR! conference had 28% female attendees. 

Several reasons are offered for the relative success of the R community in this regard, but in my opinion the most important of these is the vibrancy of the R-Ladies network, which now comprises 134 chapters worldwide. Shaikh lists some steps that the Python community are taking to address the issue, and provides several additional suggestions as well. You can read the complete analysis and recommendations in the blog post linked below. 

Reshama Shaikh: Why Women Are Flourishing In R Community But Lagging In Python

AI, Machine Learning and Data Science Announcements from Microsoft Ignite

Microsoft Ignite, Microsoft's annual developer conference, wrapped up last week and many of the big announcements focused on artificial intelligence and machine learning. The keynote presentation from Microsoft's Cloud AI lead Eric Boyd showcases the major developments, or you can check out his accompanying blog post for a written summary. 

In this post, I'll dive down into the details of the specific announcements and provide links to the associated recorded presentations from Ignite and other technical documents.

Tools for AI developers

Azure Machine Learning Service (Azure ML) adds several new capabilities. These include a new Python package that developers can use to prepare data for analysis, automate machine learning and deep learning model training, track model performance under various conditions, and automate the model selection process (see below). It supports a variety of execution environments, from local desktops to GPU clusters, and deploy trained models with containers or even high-performance FPGA chips.  (The desktop "Workbench" client, previously in preview, is no longer required and has been deprecated.) Watch the Ignite presentation AI with Azure Machine Learning services: Simplifying the data science process for an overview and demo.

Automated Machine Learning, a new Azure ML service in preview, automates the selection of machine learning algorithms (and associated parameters and hyperparameters) to optimize predictive performance. This blog post provides the technical details, and another blog post describes some practical applications. A Microsoft Research paper describes the probabilistic matrix factorization technique behind the optimization. This Ignite presentation provides further details and includes a demo.  

Visual Studio Code Tools for AI has been updated to provide a convenient interface to Azure ML for users of the popular open-source editor.

SQL Server 2019 is now in public preview, with many improvements shown in this Ignite keynote. Updates include Spark support, and added Linux support for in-database R and Python (watch this Ignite presentation for details).

Azure Databricks is now supported in more regions, offers GPU support for deep learning, and Databricks Delta is now available (in preview) for transactional data capabilities.

Customized AI

Microsoft Bot Framework SDK v4 is now available. The Ignite presentation "Creating Enterprise-Scale Intelligent Agents and Bots" provides an overview and several examples.

Cortana Skills Kit for Enterprise, a development platform based on Azure Bot Service and Language Understanding is now in private preview.  

Pre-trained AI

Speech Service in Azure Cognitive Services is now generally available, and includes a new neural text-to-speech capability for humanlike synthesized speech. This Ignite presentation provides an overview.

This Ignite presentation provides an overview of Cognitive Services with some demos with applications of the pre-trained AI capabilities for computer vision, speech, search and more.

Programs

AI for Humanitarian Action announced. With this new AI for Good program, MIcrosoft will dedicate $40m over five years to deploy AI technologies to assist with disaster recovery; the needs of children, refugees and displaced people; and human rights. A short video summarizes the initiative.

And that's just the highlights — for example, I haven't included the many AI capabilities added to products like Windows and Office. (However I can't resist mentioning the new Insert Data from Picture feature in Excel, which I'm sure I'll be making a lot of use of.) For more from Ignite, take a look at Buck Woody's index of presentations at Ignite, with highlights of videos focused on AI, machine learning, and data science.

If not Notebooks, then what? Look to Literate Programming

Author and research engineer Joel Grus kicked off an important conversation about Jupyter Notebooks in his recent presentation at JupyterCon: 

There's no video yet available of Joel's talk, but you can guess the theme of that opening slide, and walking through the slides conveys the message well, I think. Yuhui Xie, author and creator of the rmarkdown package, provides a detailed summary and response to Joel's talk, where he lists Joel's main critiques of Notebooks: 

1. Hidden state and out-of-order execution
2. Notebooks are difficult for beginners
3. Notebooks encourage bad habits
4. Notebooks discourage modularity and testing
5. Jupyter’s autocomplete, linting, and way of looking up the help are awkward
6. Notebooks encourage bad processes
7. Notebooks hinder reproducible + extensible science
8. Notebooks make it hard to copy and paste into Slack/Github issues
9. Errors will always halt execution
10. Notebooks make it easy to teach poorly
11. Notebooks make it hard to teach well 

Yihui suggests that many of these shortcomings of Notebooks could be addressed through literate programming systems, where the document you edit is plain-text (and so easy to edit, manage, and track), and computations are strictly processed from the beginning of the document to the end. I use the RMarkdown system myself, and find it a delightful way of combining code, output and graphics in a single document, which can in turn be rendered in a variety of formats including HTML, PDF, Word and even PowerPoint.

Yihui expands on these themes in greater detail in his excellent book (with JJ Allaire and Garrett Grolemund), R Markdown: The Definitive Guide, published by CRC Press. Incidentally, the book itself is a fine example of literate programming; you can find the R Markdown source here, and you can read the book in its entirety here. As Joel mentions in his talk, an automatically-generated document of that length and complexity simply wouldn't be possible with Notebooks.

All that being said, RMarkdown is (for now) a strictly R-based system. Are there equivalent literate programming systems for Python? That's a genuine question — I don't know the Python ecosystem well enough to answer — but if you have suggestions please leave them in the comments.

Yihui Xie: The First Notebook War