<?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 Sasha Petren on Medium]]></title>
        <description><![CDATA[Stories by Sasha Petren on Medium]]></description>
        <link>https://medium.com/@sashapetren?source=rss-b06b981fd751------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*KBSNbu81o0QHaZGQ.</url>
            <title>Stories by Sasha Petren on Medium</title>
            <link>https://medium.com/@sashapetren?source=rss-b06b981fd751------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 23:33:53 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sashapetren/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[The Tensorflow Weakness is a Gluon Strength]]></title>
            <link>https://medium.com/@sashapetren/the-tensorflow-weakness-is-a-gluon-strength-46e818038b26?source=rss-b06b981fd751------2</link>
            <guid isPermaLink="false">https://medium.com/p/46e818038b26</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Sasha Petren]]></dc:creator>
            <pubDate>Mon, 05 Nov 2018 17:04:40 GMT</pubDate>
            <atom:updated>2018-11-05T17:04:40.262Z</atom:updated>
            <content:encoded><![CDATA[<p>On October 12, 2017, <a href="https://www.businesswire.com/news/home/20171012005742/en/AWS-Microsoft-Announce-Gluon-Making-Deep-Learning">Gluon</a> became available to the public. Microsoft’s partnership with Amazon finally has a chance to outshine Google with its great TensorFlow.</p><p>Gluon is a deep learning multiplatform tool which currently utilizes Apache MXNet and soon will work on Microsoft Cognitive Toolkit (CNTK) as well as other platforms to come shortly after.</p><p>This is not the first time Microsoft and Amazon worked together in AI space. On August 2017, they announced that Microsoft Cortana and Amazon Alexa would be able to communicate across a range of tasks, and users will be able to open Cortana through Alexa and vice versa.</p><p>Shortly after partnering on the virtual assistant endeavor, Microsoft and Amazon decided to get together and create a deep learning platform that can capitalize on Google’s TensorFlow weaknesses such as the complexity of coding, slow training performance, and manual construction of the neural network definition.</p><p>TensorFlow was created mostly for developers in mind where Gluon is for researchers and anyone who is passionate about using deep learning platform with minimal training.</p><p>Here are some of the advantages of Gluon:</p><ul><li>Concise — easy-to-understand programming interface</li><li>Dynamic graphs — with Gluon developers can create neural network models on the fly, with any structure and use any of Python’s native control flow</li><li>High performance — the training method has minimal impact on the speed of the underlying engine</li><li>Flexible, imperative structure — Gluon does not require the neural network model to be rigidly defined. Developers can perform model training one step at a time because the Gluon interface brings together the training algorithm and the neural network model. It is much easier to debug, update and reuse neural networks</li><li>Prebuild template — Gluon can use the deep learning templates developed by Microsoft and Amazon</li></ul><p>Even though the Gluon is an impressive tool with the bright future, TensorFlow still dominates the market which makes it the leader in the AI framework and the fifth-most-popular project (by stars) on GitHub. However, Gluon has many competitive advantages as well as the backing from tech giants (Microsoft, Amazon), so it has a potential to take over this space in the coming years.</p><p>In my next Blog, I will deep-dive on Gluon’s framework. Ask me for more!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=46e818038b26" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning Vs. Statistical Learning]]></title>
            <link>https://medium.com/@sashapetren/machine-learning-vs-statistical-learning-365f25a9a649?source=rss-b06b981fd751------2</link>
            <guid isPermaLink="false">https://medium.com/p/365f25a9a649</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[statistical-learning]]></category>
            <dc:creator><![CDATA[Sasha Petren]]></dc:creator>
            <pubDate>Mon, 05 Nov 2018 17:02:39 GMT</pubDate>
            <atom:updated>2018-11-05T17:02:39.332Z</atom:updated>
            <content:encoded><![CDATA[<p>Most of the time as a data scientist I get asked the question, what is the difference between Machine Learning and Statistical Learning? Even though you would think that the answer is obvious, there are a lot of novice data scientists that are still confused about those two approaches.</p><p>As a beginner data scientist, it is hard for you to see the differences between the two, and it is probably due to how we learn Data Science. To become a data scientist, you are required to develop knowledge in multiple subjects such as Statistics, Programming, SQL, Linear Algebra and have the domain expertise. Hopefully, you will start your journey with Statistics, and most of the data scientists believe that this is the foundation in Data Science and I cannot disagree with them.</p><p>After, when you get comfortable with Statistics, then, eventually, expand your horizons within Data Science, by sailing away from all too familiar, small datasets such as Titanic, Iris, Cars, Diamonds, etc. to more uncharted territories, to a new world of Big Data. Nevertheless, with your confidence in Statistical Learning, you will probably take on a big data challenge and hope to generate insight from your data by applying Statistical Learning techniques. I don’t want to disappoint you, but not much value will be formed from this method. This is because you incorrectly approached the situation, you applied a statistical learning solution to a machine learning problem. I cannot stress enough the importance of understanding the differences between those two.</p><p>To prevent novice data scientists from the future disappointments, I have composed a list of differences between Statistical Learning and Machine Learning to aid you on your journey to success.</p><p><strong>Here are some of the differences:</strong></p><ol><li>Both methods are data dependent. However, Statistical Learning relies on rule-based programming; it is formalized in the form of relationship between variables, where Machine Learning learns from data without explicitly programmed instructions.</li><li>Statistical Learning is based on a smaller dataset with a few attributes, compared to Machine Learning where it can learn from billions of observations and attributes.</li><li>Statistical Learning operates on assumptions, such as normality, no multicollinearity, homoscedasticity, etc. when Machine Learning is not as assumptions dependent and in most of the cases ignores them.</li><li>Statistical Learning is mostly about inferences, most of the idea is generated from the sample, population, and hypothesis, in comparison to Machine Learning which emphasizes predictions, supervised learning, unsupervised learning, and semi-supervised learning.</li><li>Statistical Learning is math intensive which is based on the coefficient estimator and requires a good understanding of your data. On the other hand, Machine Learning identifies patterns from your dataset through the iterations which require a way less of human effort.</li></ol><p>Even though most will argue that Machine Learning is superior, and to some extent, I will agree. On the contrary, with the application of Statistical Learning, you familiarize yourself better with your data which help you to build that needed confidence in your modeling.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=365f25a9a649" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Benefits Using Cloud-Based Platforms Data Science]]></title>
            <link>https://medium.com/@sashapetren/benefits-using-cloud-based-platforms-data-science-71a8ab8cec5c?source=rss-b06b981fd751------2</link>
            <guid isPermaLink="false">https://medium.com/p/71a8ab8cec5c</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[azure]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Sasha Petren]]></dc:creator>
            <pubDate>Mon, 05 Nov 2018 17:00:52 GMT</pubDate>
            <atom:updated>2018-11-05T17:08:33.610Z</atom:updated>
            <content:encoded><![CDATA[<h4>Benefits of Using Cloud-Based Platforms for Data Science</h4><p>As everything in this world matures and goes through the steps of evolution, Data Science is not much different. Even though it became more affordable for the companies to jump on the Data Science bandwagon and generate quick wins, Volume, Velocity, and Variety are still the bottleneck for most of the Artificial Intelligence (AI) projects. This makes it challenging for a data scientist to take full advantage of your organization’s data. It’s been found that demand for end-to-end solutions over the past years have grown; 40% in 2016, 51% in 2017 and expected to rise to 70% in 2018, according to Forbes.</p><p>With that growing trend in mind, data scientists have few options available to tackle demand for the end-to-end solution.</p><p>The most straightforward approach for a data scientist is to keep everything as is, stay within the realm of Statistical Learning, continue working with small datasets. Another option for a data scientist is to go lavishly with your company’s dollars and start purchasing big data toys, servers, maintenance, resources, etc., hoping all stays the same. The optimum solution is to utilize available cloud-based platforms such as Amazon Web Services, Google Cloud Platform, IBM Watson or Microsoft Azure for Data Science projects.</p><h3>Here are the benefits of utilizing Data Science on top of a cloud-based platform, that if you decide to go with the optimum solution:</h3><p><strong>No Maintenance is Required<br></strong>By utilizing cloud-based platform you can concentrate only on developing your models, and leave the upkeep work to the cloud platform providers.</p><p><strong>Access to the Latest Tools</strong><br>You don’t need to go through your traditional IT Support to have them install software on your computer. Most of the Data Science software already comes preloaded with cutting-edge tools in your environment, and if it is not, you can instantly download the needed software.</p><p><strong>Ability to Scale</strong><br>You don’t always run your machines at full capacity, which makes it difficult to describe to management why we need, what we need, and for the amount of time we need. Cloud-based platforms give you an ability to scale at need, and for the amount of time you need to do your analysis.</p><p><strong>Pay for What You Use</strong><br>You pay for the resources you used and only for what you use, with some vendors’ storage is free or incredibly cheap.</p><p>There are far many more benefits to Data Science in the cloud than outlined above. Choosing the right platform can be difficult and time-consuming, and if you encounter yourself in that situation, let us know we will help you make the right decision.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=71a8ab8cec5c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Modeling as Proof of Concept (POC)]]></title>
            <link>https://medium.com/@sashapetren/modeling-as-proof-of-concept-poc-4cb9a25accc?source=rss-b06b981fd751------2</link>
            <guid isPermaLink="false">https://medium.com/p/4cb9a25accc</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[modeling]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Sasha Petren]]></dc:creator>
            <pubDate>Mon, 05 Nov 2018 16:58:43 GMT</pubDate>
            <atom:updated>2018-11-05T16:58:43.813Z</atom:updated>
            <content:encoded><![CDATA[<p>Why do I give my precedence to build the model as Proof of Concept (POC) instead of following established methodologies such as CRISP-DM, SEMMA, AIE, MAD Skills, etc.?</p><p>Even though most of the Data scientists will say that they are two different things and used for different purposes; one is a methodology or a step by step approach to deliver the workable model, and the other is to test your idea to make sure your model is possible. Yes, they are different. However, the goal is to have a good working model in both approaches.</p><p>Here are the reasons why I prefer to create a working model as Proof of Concept (POC).</p><ul><li>Flexibility — It gives you the needed flexibility, you can incorporate feedback loop, validation, third-party knowledge and SME into any step of the process.</li><li>Benchmark — It gives you a baseline model that you can use as a reference to compare to your other models. Also, you can demo your baseline model at any time through your engagement.</li><li>Support — Not all Data scientists are Data engineers, working on POC as a collaborative task will guarantee that majority of heavy liftings such as ETL, Parallel Computation, and Data Preprocessing will be done on the server side and let you concentrate on the modeling.</li><li>Holistic View — By creating POC you get to see how your model is fit within the scheme of things, you’ll get to see the fruits of your labor and not just an isolated model. Also, you will know exactly what tools, resources, and how much time is required to create a working prototype, because you will have a complete view of your creation.</li><li>Challenges — You will identify all your challenges early in your development and as they happen will be able to attend them.</li></ul><p>These are my reasons to create the model using Proof of Concept approach; everyone is different. However, I urge you to try it. I’m a strong believer that you will be much satisfied with the model you have created.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4cb9a25accc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is the Future for Data Science Platforms?]]></title>
            <link>https://medium.com/@sashapetren/what-is-the-future-for-data-science-platforms-917d02f4580c?source=rss-b06b981fd751------2</link>
            <guid isPermaLink="false">https://medium.com/p/917d02f4580c</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[customer-data-platform]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Sasha Petren]]></dc:creator>
            <pubDate>Mon, 05 Nov 2018 16:54:44 GMT</pubDate>
            <atom:updated>2018-11-05T16:54:44.908Z</atom:updated>
            <content:encoded><![CDATA[<h4>What will the future hold for data science and machine learning platforms? Most of you already saw <a href="https://pages.alteryx.com/analyst-report-2018-gartner-mq-data-science-machine-learning.html?utm_source=google&amp;lsm=google&amp;utm_medium=cpc&amp;utm_campaign=Gartner-Magic-Quadrant-2018&amp;utm_content=2018-gartner-mq&amp;utm_term=magic%20quadrant&amp;utm_adid=254646986638&amp;gclid=Cj0KCQjwqM3VBRCwARIsAKcekb2GsmwmHPMmVduCCMj7SSnq2HFSUrSEFkbNCXsLcj9sWgiT9vyswBAaAgI8EALw_wcB">Gartner’s perspective</a> in their latest Magic Quadrant on the evolution of data science that was released at the beginning of 2018 and are familiar with the outcome. However, did you ever ask yourself why SAS, KNIME, RapidMiner, and H2O.ai became the leaders?</h4><p>I will be honest with you, I was a little surprised myself, and probably you were too, with the results for the leaders in Data Science and Machine Learning field since I expected to see Microsoft, Google, or IBM. I decided to do a quick investigation into why SAS, KNIME, RapidMiner, and H2O.ai became the leaders.</p><p>Shortly into my research, I started to notice a trend that differentiates Microsoft, Google, and IBM from Gartner’s leaders. This makes a lot of sense in a world where a good data scientist is a scarce resource, algorithms are complicated, and good coding skills are required. Companies like SAS, KNIME, RapidMiner, and H2O.ai found a more attractive approach to aid companies with their analytics by ‘solutionizing’ these challenges.</p><p>Here are the three main differentiators that I believe are the contributors to position SAS, KNIME, RapidMiner, and H2O.ai onto the leader’s quadrant.</p><ul><li>Error proof — SAS, KNIME, RapidMiner, and H2O.ai can be easily installed on any machine. With data channeling through the predesigned nodes, it is hard to make a mistake, which makes it error proof.</li><li>Modeling — Most of the models are embedded into the GUI type of approach, which is simple: you select, move, connect the needed components to generate remarkably accurate results, requiring no knowledge of programming unless you need customization.</li><li>Implementation — you are not required to be a data scientist since these tools can be easily implemented across your organization with minimal training. Keeping things simple has always been the key to success within any organization.</li></ul><p>Gartner has solid reasons for placing SAS, KNIME, RapidMiner, and H2O.ai as leaders into their Magic Quadrant for 2018. If the trend continues, we might see other companies jumping on the analytics roller coaster and emerging as leaders in 2019.</p><p>I would love to hear your opinion on the Gartner Magic Quadrant, and what is the future holds for data science tools and platforms.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=917d02f4580c" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>