Press "Enter" to skip to content

Category: Machine Learning

Quantifying Model Uncertainty with Tensorflow Probability

Published 2022-04-29 by Kevin Feasel

Vini Jaiswal reviews the Tensorflow Probability library:

In this blog, we look at the topic of uncertainty quantification for machine learning and deep learning. By no means is this a new subject, but the introduction of tools such as Tensorflow Probability and Pyro have made it easy to perform probabilistic modeling to streamline uncertainty calculations. Consider the scenario in which we predict the value of an asset like a house, based on a number of features, to drive purchasing decisions. Wouldn’t it be beneficial to know how certain we are of these predicted prices? Tensorflow Probability allows you to use the familiar Tensorflow syntax and methodology but adds the ability to work with distributions. In this introductory post, we leave the priors and the Bayesian treatment behind and opt for a simpler probabilistic treatment to illustrate the basic principles. We use the likelihood principle to illustrate how an uncertainty measure can be obtained along with predicted values by applying them to a deep learning regression problem.

Read on for an interesting explanation and tutorial.

Comments closed

Custom Model Evaluation Metrics with MLflow

Published 2022-04-22 by Kevin Feasel

Mark Zhang shows off a new bit of functionality in MLflow:

According to an internal customer survey, 75% of respondents say they frequently or always use specialized, business-focused metrics in addition to basic ones like accuracy and loss. Data scientists often utilize these custom metrics as they are more descriptive of business objectives (e.g. conversion rate), and contain additional heuristics not captured by the model prediction itself.

In this blog, we introduce an easy and convenient way of evaluating MLflow models on user-defined custom metrics. With this functionality, a data scientist can easily incorporate this logic at the model evaluation stage and quickly determine the best-performing model without further downstream analysis

Click through to see how to use built-in metrics but also how to create your own.

Comments closed

Iteratively Tuning Graph Neural Networks

Published 2022-04-20 by Kevin Feasel

Luis Bermudez takes us through the process of tuning one flavor of neural network:

We made our own implementations of OGB leaderboard entries for two popular GNN frameworks: GraphSAGE and a Relational Graph Convolutional Network (RGCN). We then designed and executed an iterative experimentation approach for hyperparameter tuning where we seek a quality model that takes minimal time to train. We define quality by running an unconstrained performance tuning loop, and use the results to set thresholds in a constrained tuning loop that optimizes for training efficiency.

Read on to see how they did it.

Comments closed

Topic Modeling with Python

Published 2022-04-08 by Kevin Feasel

Sanil Mhatre takes us through topic modeling:

Topic modeling is a powerful Natural Language Processing technique for finding relationships among data in text documents. It falls under the category of unsupervised learning and works by representing a text document as a collection of topics (set of keywords) that best represent the prevalent contents of that document. This article will focus on a probabilistic modeling approach called Latent Dirichlet Allocation (LDA), by walking readers through topic modeling using the team health demo dataset. Demonstrations will use Python and a Jupyter notebook running on Anaconda. Please follow instructions from the “Initial setup” section of the previous article to install Anaconda and set up a Jupyter notebook.

The second article of this series, Text Mining and Sentiment Analysis: Power BI Visualizations, introduced readers to the Word Cloud, a common technique to represent the frequency of keywords in a body of text. Word Cloud is an image composed of keywords found within a body of text, where the size of each word indicates its frequency in that body of text. This technique is limited in its ability to discover underlying topics and themes in the text, because it only relies on the frequency of keywords to determine their popularity. Topic modeling overcomes these limitations and uncovers deeper insights from text data using statistical modeling for discovering the topics (collection of words) that occur in text documents.

Read on for an informative article with plenty of code.

Comments closed

An Overview of Automatic Text Summarization

Published 2022-04-01 by Kevin Feasel

Kevin Jacobs looks at the state of the art with respect to automatic text summarization:

Automatic text summarization comes in two flavours: extractive summarization and abstractive summarization. Extractive summarization models take exact phrases from the reference documents and use them as a summary. One of the very first research papers on (extractive) text summarization is the work of Luhn [1]. TextRank [2] (based on the concepts used by the PageRank algorithm) is another widely used extractive summarization model.

In the era of deep learning, abstractive summarization became a reality. With abstractive summarization, a model generates a text instead of using literal phrases of the reference documents. One of the more recent works on abstractive summarization is PEGASUS [3] (a demo is available at HuggingFace).

Click through for a couple contemporary examples as well as a few pain points you can experience when using the current set of libraries and algorithms.

Comments closed

A Conceptual Discussion of Active Learning

Published 2022-03-17 by Kevin Feasel

Kevin Jacobs teaches us to learn:

Active Learning is a method in which data is annotated in s smart way. With data annotation, you would normally get to see a randomly selected item which you need to label. This however can lead to a lot of repetition of similar items which you have to label. This is a waste of time. A better way would be to use Active Learning. For Active Learning, a batch of random items is selected first. Then, a lightweight classifier is used for evaluating the previously annotated data.

Basically, run your prediction mechanism, find the things about which the mechanism is least certain, and figure those out. Doing this reduces ambiguity and quickly leads to a better model.

Comments closed

Building a Recommender in Spark

Published 2022-03-03 by Kevin Feasel

Avinash Sooriyarachchi makes a recommendation:

There has been an exponential increase in the volume and variety of data at our disposal to build recommenders and notable advances in compute and algorithms to utilize in the process. Particularly, the means to store, process and learn from image data has dramatically increased in the past several years. This allows retailers to go beyond simple collaborative filtering algorithms and utilize more complex methods, such as image classification and deep convolutional neural networks, that can take into account the visual similarity of items as an input for making recommendations. This is especially important given online shopping is a largely visual experience and many consumer goods are judged on aesthetics.

In this article, we’ll change the script and show the end-to-end process for training and deploying an image-based similarity model that can serve as the foundation for a recommender system. Furthermore, we’ll show how the underlying distributed compute available in Databricks can help scale the training process and how foundational components of the Lakehouse, Delta Lake and MLflow, can make this process simple and reproducible.

Click through for the process.

Comments closed

Scoring Azure ML Models in Azure Synapse Analytics

Published 2022-03-03 by Kevin Feasel

Alex Aleksandrov shows off the PREDICT operator:

We can use Synapse for many activities. We can use it not only for ingesting, querying, storing and visualising data, but for developing machine learning models as well. Of course, one can say that doing data science is another functionality of this platform and this is definitely true. However, in this article, I would like to show you that instead of using Python, one can use T-SQL for doing predictions.

Click through to see how.

Comments closed

AutoML with pycaret

Published 2022-03-01 by Kevin Feasel

Brendan Tierney looks at the pycaret library:

In this post we will have a look at using the AutoML feature in the Pycaret Python library. AutoML is a popular topic and allows Data Scientists and Machine Learning people to develop potentially optimized models based on their data. All requiring the minimum of input from the Data Scientist. As with all AutoML solutions, care is needed on the eventual use of these models. With various ML and AI Legal requirements around the World, it might not be possible to use the output from AutoML in production. But instead, gives the Data Scientists guidance on creating an optimized model, which can then be deployed in production. This facilitates requirements around model explainability, transparency, human oversight, fairness, risk mitigation and human in the loop.

Read on for a tutorial as well as additional resources.

Comments closed

Form Recognizer Updates

Published 2022-02-25 by Kevin Feasel

Vinod Kurpad shares some news:

Form Recognizer continues to improve product capabilities with improved models, support for additional document types and containerized solutions that run in the cloud or on premises either connected or fully disconnected for scenarios where containers need to run in an isolated environment. Recent updates to pricing include commitment tiers for customers who have a predictable volume of documents. Starting February 15th, the pricing for Invoices and General Document API will drop to $10 per 1000 pages, an 80% reduction, making it possible for customers to use invoices and the general document APIs for high volume scenarios to significantly lower cost while providing additional value.

That’s a pretty big improvement.

Comments closed