<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Wes Yiu on Medium]]></title>
        <description><![CDATA[Stories by Wes Yiu on Medium]]></description>
        <link>https://medium.com/@wesyiu?source=rss-1dc22cddc32e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*uNjMsNFSGGCO1LhI.</url>
            <title>Stories by Wes Yiu on Medium</title>
            <link>https://medium.com/@wesyiu?source=rss-1dc22cddc32e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 04:15:14 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@wesyiu/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Machine Learning Isn’t Easy — The Need For Automated Feature Engineering]]></title>
            <link>https://medium.com/@wesyiu/machine-learning-isnt-easy-the-need-for-automated-feature-engineering-75bd5026e410?source=rss-1dc22cddc32e------2</link>
            <guid isPermaLink="false">https://medium.com/p/75bd5026e410</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Wes Yiu]]></dc:creator>
            <pubDate>Wed, 07 Mar 2018 02:42:01 GMT</pubDate>
            <atom:updated>2018-03-07T02:42:01.489Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UWkL4QRmHnvLiIoXPoU2qA.jpeg" /></figure><p>Machine learning is everywhere in the startup ecosystem with applications across healthcare, media, and Fintech to name a few. In the past decade, some of the most prominent applications include self-driving cars and web search. It’s all around us, integrated into many of our daily routines, yet most people are probably not even of aware of it or even know what it really is. A very simple description of machine learning is that involves developing an algorithm or model that is able to analyze data and recognize patterns and then predict similar patterns in a new set of data. Over time as more data is accessed by the model, the model should continue to evolve and improve its predictive power. Machine learning is such a popular buzzword thrown around that perhaps it has given the general public a false sense that it’s very easy to implement. I’m certainly not a data scientist nor a self-professed expert in this space, but I’ve come to appreciate how difficult implementing machine learning is in practice. In fact, many companies struggle with this.</p><p>Despite the widespread use of machine learning in today’s word, it’s not enough to say you’re implementing it. Your machine learning models must actually produce high-quality and meaningful results. That is the ultimate challenge. A source of frustration for data scientists in this regard comes from what is called feature engineering. A feature is an attribute or variable of a set of data that is used for analysis and prediction. They are the bedrock of powering machine learning algorithms. One example of a feature in online retail could be “the total amount spent by a customer in the past month.” To give a crude example, this feature could be used in a model that is designed to predict the total amount spent in a single year.</p><p>Feature engineering is the process of analyzing data sets to extract these predictive features. Not all features are useful for a specific algorithm, so naturally, the quality of features used will have a large influence on whether you have developed a high-performing model. Better features often result in greater flexibility, simpler models, and overall better predictive results. Unfortunately, the problem with feature engineering is that it’s a painstakingly manual, error-prone, and time-consuming process. Data can come in all types of formats, but for machine learning to work, the data needs to be numeric and the features need to be exposed. Often vast amounts of data are presented in an unfriendly format and important features are not readily obvious. Feature engineering is an iterative process that requires data scientists to:</p><p>1. Brainstorm features</p><p>2. Develop the features</p><p>3. Evaluate the models using the developed features</p><p>4. Rinse &amp; Repeat</p><p>If this process is done correctly, it greatly increases the predictive power of machine learning algorithms. This step is very critical, yet it often requires significant domain expertise which presents a significant headwind for machine learning.</p><p>So how do we get by this limitation? We need to bring automation into the fold to help augment the human process of this step. One company that is working to do this called Feature Labs. The concept and company spun out of the MIT Computer Science and Artificial Intelligence Lab. The team calls their innovative process Deep Feature Synthesis (DFS). Applying simple arithmetic operations such as max, min, sum etc… across various types of datasets (retail, financial, enterprise etc…), their platform has been able to enhance the feature engineering process by quickly coming up with many features that could be valuable for a specific machine learning model. The beauty of the platform is that it requires little human intervention thanks to the help of automation.</p><p>The company is still in its early days, but they released an open-source version of their DFS platform on Github and it has already become the most popular library for feature engineering. This has enabled the platform to constantly evolve as more and more people use it to create new and interesting features. According to their website, they already have some big name customers such as NASA and SAP, so they’re clearly trending up. Through the use of Feature Labs’ platform companies now can have a better grasp of the capabilities of their machine learning applications, which historically has been a deep struggle. It looks like machine learning is about to become even more impressive.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=75bd5026e410" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>