#AutoML – Automated Machine Learning,AKA #Skynet

Hi!

IMHO one of the most important announcements presented last week in Ignite was the Azure preview for AutoML: Automated Machine Learning.

I’m not going to get into details about AutoML, the best option is to read the official post from the Azure Machine Learning team (see references). I’ll do my best effort to try to summarize that the objective of this new tool if to allows you to automatically identify the best pipeline to work in a machine learning environment / scenario.

A pipeline comprises the basic steps of a process of ML

  • Working with data, this means sorting, filtering, check for nulls, labeling, etc.
  • Select a learning algorithm, SVM, Fast Tree, etc.
  • Define features and Labels, adjust parameters, etc

The [try / error / learn] model in each of these steps help us to improve our model, and to get better results (better accuracy).

AutoML It proposes an automatic service, where the best combination is identified to create a pipeline with the best possible accuracy. As always an image rocks the explanation

01 AutoML process.png

Official description

Automated ML is available to try in the preview of Azure Machine Learning. We currently support classification and regression ML model recommendation on numeric and text data, with support for automatic feature generation (including missing values imputations, encoding, normalizations and heuristics-based features), feature transformations and selection. Data scientists can use automated ML through the Azure Machine Learning Python SDK and Jupyter notebook experience. Training can be performed on a local machine or by leveraging the scale and performance of Azure by running it on Azure Machine Learning managed compute. Customers have the flexibility to pick a pipeline from automated ML and customize it before deployment. Model explainability, ensemble models, full support for Azure Databricks and improvements to automated feature engineering will be coming soon.

From here I strongly recommend reading the official documentation that is where it is explained in detail AutoML. Also, if you are familiar with Jupyter Notebooks, in few seconds you can clone and access a new library with a tutorial to try AutoML from zero. You need to clone a repo from https://github.com/Azure/MachineLearningNotebooks

02 AutoML Jupyter Notebooks

The tutorial is pretty straightforward, and with little Azure resources you can see how you optimize a A Classification model with AutoML

03 AzureML Local tutorial

Although for now only models of classification and regression are supported, AutoML is a tool a Keep in mind when you start working in ML.

See you at the event, Happy Coding!

Greetings @ Toronto

El Bruno

References

Advertisements

#AutoML – Automated Machine Learning, modelos de #MachineLearning que aprenden a optimizarse! (en las movies se llama #skynet)

Buenas!

Una de las noticias mas importantes que se presentaron la semana pasada en Ignite fue la Preview de Azure AutoML: Automated Machine Learning.

Lo mejor para entrar en detalles sobre AutoML es leer el post oficial del equipo de Azure Machine Learning (ver referencias). Yo lo intentare resumir en un nuevo framework que permite identificar de forma automática el mejor pipeline para trabajar con datos.

Un pipeline comprende los pasos básicos de un proceso de ML

  • Trabajar con datos, esto significa ordenarlos, eliminar los nulls, etiquetarlos, etc.
  • Seleccionar un algoritmo de aprendizaje, SVM, Fast Tree, etc.
  • Definir Features y Labels, ajustar parámetros, etc

El modelo de prueba / error / aprendizaje en cada uno de estos pasos define la precisión que tendrá nuestro modelo final.

AutoML propone un servicio automático, donde se identifica la mejor combinación para crear una Pipeline con la mejor precisión posible. Como siempre una imagen ayuda a la explicación

01 AutoML process.png

Y la descripción oficial

Automated ML is available to try in the preview of Azure Machine Learning. We currently support classification and regression ML model recommendation on numeric and text data, with support for automatic feature generation (including missing values imputations, encoding, normalizations and heuristics-based features), feature transformations and selection. Data scientists can use automated ML through the Azure Machine Learning Python SDK and Jupyter notebook experience. Training can be performed on a local machine or by leveraging the scale and performance of Azure by running it on Azure Machine Learning managed compute. Customers have the flexibility to pick a pipeline from automated ML and customize it before deployment. Model explainability, ensemble models, full support for Azure Databricks and improvements to automated feature engineering will be coming soon.

Pues bien, a partir de aquí recomiendo leer la documentación oficial que es donde se explica en detalle AutoML.

Si estas familiarizado con Jupyter notebooks, en pocos segundos puedes tener acceso a un tutorial mas que completo solo clonado una library desde https://github.com/Azure/MachineLearningNotebooks

02 AutoML Jupyter Notebooks

El tutorial es bastante sencillo, y con pocos recursos de Azure puedes ver como se optimiza un un modelo de clasificación con AutoML

03 AzureML Local tutorial

Si bien por ahora solo se soportan modelos de clasificación y regresión, AutoML es una herramienta a tener en cuenta cuando comienzas a trabajar en ML.

Nos vemos en el evento, happy coding!

Saludos @ Toronto

El Bruno

References

#Ignite – A couple of personal highlights on Microsoft Ignite 2018

Hi!

So, the 1st day of MS Ignite is over and I’m putting all together the most interesting topics for me.

Digital transformation is how you compete and win today. It’s the path to better serve customers, build better products, and empower employees. And it’s data that is the foundational resource that allows you to transform and uncover the insights you need to drive business forward.

Adobe, Microsoft, and SAP are partnering on the Open Data Initiative to enable data to be exchanged—and enriched—across systems, making it a renewable resource that flows into intelligent applications.

With a single, comprehensive view of data, you’ll discover in real-time more about your customers, identify ways to maximize operations, and find new ways to provide the amazing experiences that your customers deserve.

And finally, Modern Workplace. I’ve been spending a lot of time on this topic. Helping clients to improve their culture around workplace, collaboration, sharing and more it’s an exciting time. Inside Avanade we have a full offering around this, we called Workplace Experience.

And I really like this article [10 new ways for everyone to achieve more in the modern workplace], specially where they talked about the new feature in Microsoft Teams: Blurred Background. (I tweeted about this a couple of days ago)

And, the following video will never became viral if this feature were available 😀

10-new-ways-for-everyone-to-achieve-more-in-the-modern-workplace-1.gif

 

Greetings @ Toronto

El Bruno

#Ignite – Algunos puntos interesantes de Microsoft Ignite 2018

Buenas!

Pues ha pasado el 1er día de Microsoft Ignite 2018 y la verdad es que un post no da para repasar todo lo que se ha presentado. Aquí van un par de links interesantes.

Digital transformation is how you compete and win today. It’s the path to better serve customers, build better products, and empower employees. And it’s data that is the foundational resource that allows you to transform and uncover the insights you need to drive business forward.

Adobe, Microsoft, and SAP are partnering on the Open Data Initiative to enable data to be exchanged—and enriched—across systems, making it a renewable resource that flows into intelligent applications.

With a single, comprehensive view of data, you’ll discover in real-time more about your customers, identify ways to maximize operations, and find new ways to provide the amazing experiences that your customers deserve.

  • Y por último, me parece que este año también me tocara dedicar mucho tiempo a Modern Workplace. En Avanade estamos trabajando mucho en lo que llamamos Workplace Experience.

Y sobre este tema, me ha llamado la atencion en el siguiente articulo [10 new ways for everyone to achieve more in the modern workplace] la version muy especial del tweet de Blurred Background en Microsoft Teams que publique hace unos días.

Con Teams este video nunca hubiese sido viral

10-new-ways-for-everyone-to-achieve-more-in-the-modern-workplace-1.gif

 

Saludos @ Toronto

El Bruno

#PowerShell – Download all MS Ignite videos and slides in a single click (thanks @mderooij!)

Hi !

Last week I shared a powershell script created by Jon Galloway to download videosn from Channel 9. So, Microsoft Ignite is finished, and someone pick the idea and created a similar one which also includes video and slides download features.

We need to thanks to Michel de Rooij (@mderooij), who is the author of the powershell script to download materials from Microsoft Ignite 2017.

I1

Like in the previous one, we have a couple of parameters to define some filters, like video quality. It took me almost a full day to download all the materials and at the end I finished downloading +420GB which includes 785 PowerPoint Slides and 681 sessions videos.

I2

Happy Coding!

Greetings @ Toronto

El Bruno

References

#PowerShell – Descarga todos los videos y presentaciones de MS Ignite en un click (gracias @mderooij!)

Hola!

La semana pasada compartí un script creado por Jon Galloway para descargar videos desde Channel 9. Pues bien, no se ha hecho esperar las actualizaciones del mismo y ya podemos encontrar otro PowerShell Script, mucho más actual.

Michel de Rooij (@mderooij), ha creado uno para descargar todo el contenido publicado de Microsoft Ignite 2017.

I1

Como en el anterior, tenemos varias opciones para definir el tipo de contenido a descargar, la calidad de los videos, presentaciones, etc,

Estos son bastante GBs de contenido. En mi caso habiendo descargado solo 50 de 1633 elementos ya llevo casi 20 GBs de contenidos

I2

Happy Coding!

Saludos @ Toronto

El Bruno

References