Press "Enter" to skip to content

Category: Machine Learning

Databricks Autologging

Published 2021-08-31 by Kevin Feasel

Corey Zumar and Kasey Uhlenhuth announce a new product:

Machine learning teams require the ability to reproduce and explain their results–whether for regulatory, debugging or other purposes. This means every production model must have a record of its lineage and performance characteristics. While some ML practitioners diligently version their source code, hyperparameters and performance metrics, others find it cumbersome or distracting from their rapid prototyping. As a result, data teams encounter three primary challenges when recording this information: (1) standardizing machine learning artifacts tracked across ML teams, (2) ensuring reproducibility and auditability across a diverse set of ML problems and (3) maintaining readable code across many logging calls.

Read on to see how Databricks Autologging can satisfy these issues.

Comments closed

From API Call to ML Services Prediction

Published 2021-08-20 by Kevin Feasel

Tomaz Kastrun continues a series:

From the previous two blog posts:

– Creating REST API for reading data from Microsoft SQL Server in web browser

– Writing Data to Microsoft SQL Server from web browser using REST API and node.js

We have looked into the installation process of Node.js, setup of Microsoft SQL Server and made couple of examples on reading the data from database through REST API and how to insert data back to database.

In this post, we will be looking the R predictions using API calls against a sample dataset.

Click through to see it in action.

Comments closed

Shrinking Convolutional Neural Networks for TinyML

Published 2021-08-06 by Kevin Feasel

Pete Warden writes up a tip:

A colleague recently asked for more details on an approach I recommended, but which she hadn’t seen any documentation for. I realized that it was something I’d learned from talking to model builders at Google, and I wasn’t sure there was anything written up, so in the spirit of leaving a trail of breadcrumbs for anyone coming after, I thought I should put it into a quick blog post.

The summary is that if you have MaxPool or AveragePool after a convolutional layer in a network, and you’re targeting a resource-constrained system like a microcontroller, you should try removing them entirely and replacing them with a stride in the convolution instead. This has two main benefits, but to explain it’s easiest to diagram out the network before and after.

Click through for the full explanation.

Comments closed

Deploying Custom Docker Images in Azure ML

Published 2021-08-05 by Kevin Feasel

Tsuyoshi Matsuzaki shows us how to deploy an Azure ML model via custom Docker image:

In my early post, I have showed you how to bring your own custom docker image in training with Azure Machine Learning.
On the contrary, here I’ll show you how to bring custom docker image in model deployment.

In Azure Machine Learning, the base docker image in deployment includes the inferencing assets, such as, Flask server, etc. So you should use AML compliant image for base image, even when you use your own custom docker image.
The list of these maintained AML images is available in https://github.com/Azure/AzureML-Containers .

Read on for an example.

Comments closed

An Introduction to ML.NET

Published 2021-07-30 by Kevin Feasel

Ivan Matec gives us a walkthrough of the ML.NET library and its Model Builder component:

Before we dive into our example, let’s talk a bit about ML.NET history and its current state.

ML.NET draws its origins from the 2002’s Microsoft Research project named TMSN, which stands for “test mining search and navigation.” Later it was renamed to TLC, “the learning code.” ML.NET war derived from the TLC library. Initially, it was used on internal Microsoft products.

The first publicly available version ML.NET 1.0 was released in 2019. It included the Model Builder add-in and AutoML (Automated Machine Learning) capabilities.

The current version is 1.6.0. More details about all releases can be found on the official ML.NET release page.

ML.NET is not a bad library if you need to do some fairly simple work

Comments closed

Explaining an ML Model with SHAP

Published 2021-07-29 by Kevin Feasel

Dan Lantos, et al, walk us through one technique for model explainability:

Interpretability has to do with how accurately a machine learning model can associate a cause (input) to an effect (output). 

Explainability on the other hand is the extent to which the internal mechanics of a machine or deep learning system can be explained in human terms. Or to put it simply, explainability is the ability to explain what is happening. 

Let’s consider a simple example illustrated below where the goal of the machine learning model is to classify an animal into its respective groups. We use an image of a butterfly as input into the machine learning model. The model would classify the butterfly as either an insect, mammal, fish, reptile or bird. Typically, most complex machine learning models would provide a classification without explaining how the features contributed to the result. However, using tools that help with explainability, we can overcome this limitation. We can then understand what particular features of the butterfly contributed to it being classified as an insect. Since the butterfly has six legs, it is thus classified as an insect.

Being able to provide a rationale behind a model’s prediction would give the users (and the developers) confidence about the validity of the model’s decision.

Read on to see how you can use a library called SHAP in Python to help with this explainability.

Comments closed

Explaining Chatbots

Published 2021-07-28 by Kevin Feasel

Will Harris takes us through the basics of chatbots:

You can develop your business processes into conversational workflows to help people perform tasks. This can be wide ranging; from looking up records to do with their accounts, through to engaging in new services. There are many processes that can be turned into an effective conversational workflow. This typically helps people perform activities more inclusively and conveniently or helps reduce grunt work for employees.

Brisa2, an automotive company, developed a bot to help find company data and perform tasks like password resets, helping to free up the IT team for other tasks.

Click through for an overview of the concept. I’ve been down on this generation of chatbot because, as a user, they usually show up in places where a well-designed UI would be faster and more effective—as well as less prone to failures in understanding.

Comments closed

Reinforcement Learning and Python 3

Published 2021-07-21 by Kevin Feasel

I have a new post up:

I finally got around to trying out a reinforcement learning exercise this weekend in an attempt to learn about the technique. One of the most interesting blog posts I read is Andrej Karpathy’s post on using reinforcement learning to play Pong on the Atari 2600. In it, Andrej uses the Gym package in Python to play the game.

This won’t be a post diving into the details of how reinforcement learning works; Andrej does that far better than I possibly could, so read the post. Instead, the purpose of this post is to provide a minor update to Andrej’s code to switch it from Python 2 to Python 3. In doing this, I went with the most convenient answer over a potentially better solution (e.g., switching xrange() to range() rather then re-working the code), but it does work. I also bumped up the learning rate a little bit to pick up the pace a bit.

Click through for the (slightly) updated code.

Comments closed

Running Azure ML On-Premises via Azure Arc

Published 2021-07-09 by Kevin Feasel

Tsuyoshi Matsuzaki takes us through running Azure Machine Learning via Azure Arc:

First of all, you must run Azure Arc enabled Kubernetes on-premise or on 3rd party cloud. For running Arc-enabled Machine Learning later, use machines with more than 4 CPUs, since Arc-enabled ML requires enough resources.

In this post, I assume that we run KIND (Kubernetes in Docker) cluster on on-premise Ubuntu server. (For test purpose, I have used Ubuntu 18.04 on a single virtual machine in Azure, Standard D3 v2, which has 4 CPUs and 14 GB memory.)

Click through to see how it’s done.

Comments closed

Optimizing BERT Models on Google Colab

Published 2021-07-08 by Kevin Feasel

Kevin Jacobs fine-tunes some NLP processes:

BERT is a language model and can thus be used for predicting the next word in a sentence. Furthermore, BERT can be used for automatic summarization, text classification and many more downstream tasks. Google Colab provides you with a cloud-based environment on which you can train your machine learning models on a GPU. The downside is that your data is uploaded to the Google cloud. Google Colab gives you the opportunity to finetune BERT.

Click through to see how.

Comments closed