<?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[vijay-patha - Medium]]></title>
        <description><![CDATA[Celebrating Failure, Mental Models, Machine Learning , Product - Medium]]></description>
        <link>https://medium.com/vijay-patha?source=rss----81e48ddf1579---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>vijay-patha - Medium</title>
            <link>https://medium.com/vijay-patha?source=rss----81e48ddf1579---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 01:39:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/vijay-patha" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Product Manage Hacks: Visual Guide — Part 2]]></title>
            <link>https://medium.com/vijay-patha/product-manage-hacks-visual-guide-part-2-d5b7361adc0c?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/d5b7361adc0c</guid>
            <category><![CDATA[communication]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Thu, 30 May 2024 14:53:49 GMT</pubDate>
            <atom:updated>2024-05-30T14:53:49.468Z</atom:updated>
            <content:encoded><![CDATA[<h3>Product Manage Hacks: Visual Guide — Part 2</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gZRltJZxzGL8p5y9-6BkkQ.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d5b7361adc0c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/product-manage-hacks-visual-guide-part-2-d5b7361adc0c">Product Manage Hacks: Visual Guide — Part 2</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Product Manager Hacks: Visual Guide— Part 1]]></title>
            <link>https://medium.com/vijay-patha/product-manager-hacks-visual-guide-part-1-2cbcfcb76832?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/2cbcfcb76832</guid>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Tue, 28 May 2024 21:55:48 GMT</pubDate>
            <atom:updated>2024-05-28T21:55:47.997Z</atom:updated>
            <content:encoded><![CDATA[<p>Tips for focused discussions and clear outcomes</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*17GgcRvWH1dPw5hKCDeUSA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2cbcfcb76832" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/product-manager-hacks-visual-guide-part-1-2cbcfcb76832">Product Manager Hacks: Visual Guide— Part 1</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager Guide to: ML Platform — Part 4]]></title>
            <link>https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-4-80b9129848aa?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/80b9129848aa</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Sat, 02 Dec 2023 19:17:59 GMT</pubDate>
            <atom:updated>2023-12-02T19:39:12.099Z</atom:updated>
            <content:encoded><![CDATA[<h3>A Product Manager Guide to: ML Platform — Part 4</h3><p>Avengers, assemble in the ML world!</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Ftlh2lgWp42k%3Fstart%3D13%26feature%3Doembed%26start%3D13&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dtlh2lgWp42k&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Ftlh2lgWp42k%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/1e6197bf0105afd8988fc40e5e2e9cc4/href">https://medium.com/media/1e6197bf0105afd8988fc40e5e2e9cc4/href</a></iframe><p>A lone model is a bit like a superhero taking on Thanos solo — cool but not exactly practical. Even Iron Man needed backup. Enter the Avengers! Just as Earth’s mightiest heroes join forces to tackle big baddies, a machine learning platform is the Avengers of the ML universe. It brings together the power of multiple models, tools, and processes to conquer data and model challenges.</p><p>Your dream team working together like a well-oiled machine. First, they gather and prep the data, setting the stage. Then, they dive into modeling, predicting the future like fortune-tellers. Next up, deployment and evaluation — they make it real, like turning sketches into a blockbuster movie.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tRFEXgl92b6k134pqrqTBA.png" /></figure><p>Why start with that diagram? Why not show the real complexity of an ML platform? Removing the complicated stuff reveals the beauty of simplicity, where things make sense and are clear.</p><p>As Steve Jobs said, ‘To really understand things, you have to strip away the layers and get to the basics.’</p><p>Various experts in the field have shared diverse versions of ML Platforms, yet a significant overlap in components exists. Moreover, the importance of these components varies based on a company’s needs and stage of development, and budget available.</p><p>All big cloud providers are making infrastructure for ML platform accessible. Additionally, they have army of people working on their infrastructure to adopt to various use-cases.</p><p>Here are a three examples that illustrate barebones of ML platform with some variations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tB90beCNOonxCiU_eqvyIg.png" /><figcaption>By Iansiti and Lakhani write in Competing in the Age of AI</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*J2WKuPeZZ21VkZqt.png" /><figcaption>By Chip Huyen in Designing Machine Learning Systems</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_585OoEyy0DW0zpV.png" /><figcaption><a href="https://blog.griddynamics.com/enterprise-grade-ml-platform-in-aws-a-starter-kit/">https://blog.griddynamics.com/enterprise-grade-ml-platform-in-aws-a-starter-kit/</a></figcaption></figure><p>In the next articles, I’ll break down each big part, kind of like the example below. This flowchart shows how the bits and pieces of data collection look when we take a closer look.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6HvArXeoZ1_-j2wjyrdd5A.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=80b9129848aa" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-4-80b9129848aa">A Product Manager Guide to: ML Platform — Part 4</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager Guide to: ML Platform — Part 3]]></title>
            <link>https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-3-e7bb7651cea6?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/e7bb7651cea6</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Mon, 27 Nov 2023 16:25:44 GMT</pubDate>
            <atom:updated>2023-11-27T16:25:44.511Z</atom:updated>
            <content:encoded><![CDATA[<h3>A Product Manager Guide to: ML Platform — Part 3</h3><p>What does a sandwich have in common with MLOps and ML Platform?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pRR53oMQqENhD4KX" /><figcaption>Photo by <a href="https://unsplash.com/@thematteroffood?utm_source=medium&amp;utm_medium=referral">The Matter of Food</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>MLOps, an extension or adaptation of Developments and Operations (DevOps) principles to the machine learning domain, operationalizes or brings a model to production - deploying, monitoring, and maintaining it.</p><p>ML Platform, on the other hand, is a software environment that provides the tools and infrastructure necessary to develop, deploy, and manage ML models.</p><p>MLOps and a Machine Learning Platform are interrelated but serve different aspects of the machine learning lifecycle, with MLOps focusing on the overall operationalization and collaboration, and the platform providing the technical foundation. ML platform is the foundation that ML Ops is built on. ML Ops cannot exist without ML platform, but ML platform can exist without ML Ops.</p><p>MLOps is analogous to preparing a sandwich, covering ingredient gathering, assembly, and serving. A machine learning platform is like a well-organized kitchen, providing the tools and infrastructure necessary for development and delivery of machine learning models (sandwiches).</p><p>In summary, to effectively implement machine learning, organizations should combine the strengths of MLOps and ML Platforms. MLOps streamlines the machine learning lifecycle, ensuring smooth data collection, model training, deployment, and maintenance. Meanwhile, ML Platforms provide the tools and infrastructure for efficient data storage, model training, and deployment. This holistic approach leads to robust machine learning solutions that align with business goals.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e7bb7651cea6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-3-e7bb7651cea6">A Product Manager Guide to: ML Platform — Part 3</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager Guide to: ML Platform — Part 2]]></title>
            <link>https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-2-b7596949244a?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/b7596949244a</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Fri, 10 Nov 2023 02:01:46 GMT</pubDate>
            <atom:updated>2023-11-10T02:06:32.539Z</atom:updated>
            <content:encoded><![CDATA[<h3>A Product Manager Guide to: ML Platform — Part 2</h3><p>The Culinary Comedy of Data Science and Production Systems, with the ML Platform as the Star Sous-Chef</p><p>Imagine a bustling kitchen where</p><ul><li>Data Science, the head chef fluent in Python, and</li><li>Production Systems, the dynamic kitchen staff well-versed in various tech languages,</li></ul><p>come together to whip up a tech-tastic feast.</p><p>But there’s a twist in this culinary comedy — meet the ML Platform, the star sous-chef who ensures this tech kitchen runs like a well-oiled machine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oyKEmCym39-UGv7j" /><figcaption>Photo by <a href="https://unsplash.com/@simonkaremann?utm_source=medium&amp;utm_medium=referral">Simon Karemann</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><strong>Selecting the Right Ingredients: Python and The Tech Stack with ML Platform</strong></p><p>Data Science, our head chef, presents its unique recipes. Production Systems, the diverse kitchen crew, speaks different tech languages, creating a comical tower of Babel situation. Enter the ML Platform, the multilingual sous-chef. With a dash of humor, this platform bridges the culinary gap, making sure everyone understands the secret ingredients for a tech extravaganza.</p><p><strong>Preparing the Dish: Model Serialization and Deployment with ML Platform</strong></p><p>Data Science prepares (serializes) its Python-based models, like a chef whipping up a culinary masterpiece. The ML Platform dons the hat of the “Model Courier.” It’s like the tech delivery hero, ensuring that these models are packed and delivered for a tech road trip. ML Platform can use containers to package and deploy serialized models. Containers are a lightweight virtualization technology that allows applications to be packaged with all of their dependencies into a single unit. This makes it easy to deploy and run applications on any platform.</p><p><strong>Serving the Dish: Performance and Efficiency with ML Platform</strong></p><p>In our bustling restaurant, orders are flying in like confetti at a party. Python, the head chef, is known for its skills but tends to take its sweet time. The Production Systems team is like the servers, dashing to ensure prompt delivery. It’s a race in the kitchen, and the customers can’t go hungry. The ML Platform orchestrates this dance, ensuring a seamless and hilarious relay race to serve users efficiently.</p><p><strong>Meeting the Demand: Scalability with the ML Platform</strong></p><p>The restaurant gets busier, just like a kitchen overwhelmed by a food critic’s rave review. The kitchen staff must expand smoothly. Production Systems, with their varied tech languages, also need to accommodate a growing audience. The ML Platform takes on the role of the kitchen’s tech expansion expert, ensuring everything runs smoothly as the restaurant’s tech audience grows.</p><p><strong>Quality Control: Monitoring and Feedback with ML Platform</strong></p><p>In the restaurant, chefs constantly taste their dishes, akin to diligent quality control. The ML Platform serves as the kitchen’s ever-watchful food critic, keeping a close eye on everything. If something’s not quite right, it’s like the critic sending back a dish for a little more seasoning. The ML Platform is the tech gourmet who ensures everything is up to par.</p><p><strong>Recipe Updates: Model Versioning with ML Platform</strong></p><p>Just as chefs update their menus, Data Science wants to bring new flavors to its tech menu. Model versioning is like launching a limited-time offer. The ML Platform is the chef’s best friend, ensuring these tech updates are introduced without creating chaos in the kitchen. It’s all about adding excitement to the tech palate, and the ML Platform is the tech menu maestro.</p><p><strong>Handling Mistakes Gracefully: Error Handling with ML Platform</strong></p><p>In the restaurant business, sometimes a dish doesn’t turn out as expected. It’s all about saving face with humor. Likewise, in tech integration, you need some robust error handling. The ML Platform takes on the role of the restaurant’s smooth-talking manager, ensuring that hiccups are sorted without causing a tech kitchen disaster. It’s the tech troubleshooter who keeps the show running smoothly.</p><p><strong>Securing the Recipe: Data Privacy and Security with ML Platform</strong></p><p>Just like a restaurant takes customer privacy and food safety seriously, the tech integration process must prioritize security. The ML Platform is the tech bouncer, guarding the recipe vault and ensuring the diners’ data is safe and sound. It’s the tech security sentinel that ensures everything is locked up tight.</p><p><strong>Conclusion</strong></p><p>So, in this culinary comedy of Data Science and Production Systems, the ML Platform takes the center stage. It’s the unsung hero that makes sure the tech feast is served with precision, leaving users satisfied and maybe even a bit amused.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7596949244a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-2-b7596949244a">A Product Manager Guide to: ML Platform — Part 2</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager Guide to: ML Platform — Part 1]]></title>
            <link>https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-1-9e8cd228a1bd?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/9e8cd228a1bd</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Fri, 20 Oct 2023 20:23:26 GMT</pubDate>
            <atom:updated>2023-10-20T23:58:42.406Z</atom:updated>
            <content:encoded><![CDATA[<h3>A Product Manager Guide to: ML Platform — Part 1</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GfwAGxeLJlEWFGBFNAazMQ.png" /></figure><p>This article is a fun coffee shop conversation between Alex and Sarah about ML platforms and why some organizations use them. A nosy eavesdropper provides additional commentary. In the process, you’ll be introduced to different components of an ML platform.</p><p>So, grab a snack, sit back, and let’s get started!</p><blockquote>Alex: Hey, Sarah, have you heard about Machine Learning (ML)? It’s all the rage nowadays!</blockquote><blockquote>Sarah: Yeah, I’ve heard about ML. It’s about creating those fancy algorithms and models, right?</blockquote><p><em>A random eavesdropper:: Alright, ML is becoming a legacy term. AI is the new rage. But lets be honest 90% of AI is ML. But the discussion is still revelant. Keep talking!</em></p><blockquote>Alex: Well, that’s a part of it. But think about ML as if it’s not just a lone rockstar in the corner but a full-blown rock band!</blockquote><blockquote>Sarah: (Laughs) A rock band, really?</blockquote><p><em>Random eavesdropper:: That make sense. Unless you are building a model for kaggle competition to evaluate model performance, ML in business settings do need support from a large number of systems and stakeholders. Each stakeholder have their own agenda too!</em></p><blockquote>Alex: Absolutely! So, you see, ML isn’t just about the cool algorithms. It’s about the entire system — the instruments, the stage, the lighting, the roadies — everything that makes a rock concert a success.</blockquote><blockquote>Sarah: Okay, I’m intrigued. Tell me more.</blockquote><blockquote>Alex: Well, imagine you have this super talented guitarist, which is your ML model. But without the right stage, the audience can’t see or appreciate their talent. And that’s where the ML platform comes in. It’s like the stage manager, making sure everything runs smoothly.</blockquote><p><em>Random eavesdropper:: o! I like that comment. It’s not just about building a model; it’s about the entire ecosystem. An ML platform handles data pipelines, right environment, model deployment, monitoring, and even governance.</em></p><blockquote>Sarah: But why do we need all this? Can’t we just enjoy the guitarist (the model) without all the stage drama?</blockquote><blockquote>Alex: Well, sure, you can. But remember that without a good stage, the guitarist might not be at their best, and the audience might not even show up. We need ML to be deployed, integrated, and monitored to make it useful. For example, our life got even easier when Nextflix and YouTube or Google’s began giving ML based recommendations. But what if the ML model takes 60 seconds to provide recommendations? 60 seconds is too late! Its the platform job to make those recommendations available in nano seconds to customers.</blockquote><blockquote>Sarah: I get it now. Its not just ML, its the ML system. Its about the whole system working together to create something amazing.</blockquote><blockquote>Alex: Bingo! It’s not just about the rockstar; it’s about the whole concert. And that’s why we need ML platforms — to make sure our ML “concert” is a hit every time!</blockquote><p><em>Random eavesdropper:: I like this coffee shop. Great coffee and great conversations. I learned something new — ML platform is a group of systems that work together to make ML happen. These systems could include data stores to create and store data assets, model data stores to store data from model, model and feature orchestrators, and monitoring and alerting mechanisms.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9e8cd228a1bd" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platform-part-1-9e8cd228a1bd">A Product Manager Guide to: ML Platform — Part 1</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager Guide to: ML Platforms (Why, What, and How)]]></title>
            <link>https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platforms-why-what-and-how-3fe724db93f6?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/3fe724db93f6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[machine-learning-platform]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Sat, 14 Oct 2023 11:59:48 GMT</pubDate>
            <atom:updated>2023-10-14T14:00:44.044Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*u9gfOERpnSiJ09rb" /><figcaption>Photo by <a href="https://unsplash.com/@chrislawton?utm_source=medium&amp;utm_medium=referral">Chris Lawton</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>ML/AI models are notoriously difficult to productionize.</p><ul><li>“ML models are like snowflakes: no two are the same.” — Andrew Ng, former Chief Scientist at Baidu</li><li>“Productionizing ML models is like trying to herd cats.” — Martin Casado, former SVP of Infrastructure at VMware</li><li>“ML/AI productionization is the dark matter of the universe: it makes up most of the universe, but we can’t see it.” — Anonymous ML engineer</li></ul><p>Machine learning (ML) platforms are becoming increasingly popular as businesses and organizations look to scale their ML efforts. An ML platform provides a centralized environment for developing, deploying, and managing ML models. It can help to streamline the ML workflow, reduce costs, and improve the quality and reliability of ML models.</p><p>I’ve been building a sophisticated ML platform for the past year and half. I’m excited to share my learnings in building the platform, launching models, and strategizing the future.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e3YpbbQsmjAIicggdytH9Q@2x.jpeg" /><figcaption>Photo by Clark Tibbs on Unsplash</figcaption></figure><p>This series of articles will explore the why, what, and how of ML platforms. We will discuss the benefits of using an ML platform, the different types of ML platforms available, and how to choose the right ML platform for your needs. We will also provide practical tips and advice on how to use ML platforms effectively.</p><p>Here are some topics that you may be interested in learning about in this series:</p><ul><li>What comes before ML platform</li><li>The benefits of using an ML platform</li><li>The different types of ML platforms available</li><li>How to choose the right ML platform for your needs</li><li>How to use ML platforms to develop, deploy, and manage ML models effectively</li><li>Best practices for using ML platforms in production</li><li>Case studies of how businesses and organizations are using ML platforms to achieve their goals</li></ul><p>Please let me know if you have any specific topics that you would like me to cover in this series. I am always looking for new ideas and feedback.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fe724db93f6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-manager-guide-to-ml-platforms-why-what-and-how-3fe724db93f6">A Product Manager Guide to: ML Platforms (Why, What, and How)</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager’s guide to SQL Window Functions]]></title>
            <link>https://medium.com/vijay-patha/a-product-managers-guide-to-sql-window-functions-13d33953cde1?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/13d33953cde1</guid>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Sat, 27 Mar 2021 12:54:10 GMT</pubDate>
            <atom:updated>2021-03-29T13:50:37.892Z</atom:updated>
            <content:encoded><![CDATA[<h3>A Product Manager’s Guide to Machine Learning: SQL Window Functions</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5PtXHCDh1qbMO3lf" /><figcaption>Photo by <a href="https://unsplash.com/@kalenemsley?utm_source=medium&amp;utm_medium=referral">Kalen Emsley</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The purpose of this article is to provide PMs with &lt;2 minutes intro to SQL window functions, with syntax. Use this article as a reference guide to source data to evaluate a use case or report the status or impact of your product.</p><h4>What are window functions?</h4><p><em>Window describes a set of rows. Window function is a function that performs calculations on set of rows and returns a single aggregated value for each row. unlike regular aggregate functions, use of a window function does not cause rows to become grouped into a single output row — the rows retain their separate identities.</em></p><h4>Anatomy of Window Function</h4><pre>window_function (column_name) <br>OVER (PARTITION BY column_name ORDER BY column_name)</pre><pre>Example:<br>SELECT colA, colB, colC, <br><strong>avg(colB) OVER(PARTITION BY colA ORDER BY colC)</strong></pre><pre>-- <strong><em>OVER: </em></strong><em>This is what syntactically distinguishes window functions from a regular function or aggregate function. </em>You could read the above aggregation as:<br>-- aggregate window function over the window (set of rows) of colA<br>-- take the avg of colB <em>over</em> the partition set in colA, and order by colC.</pre><pre>FROM<br>table1;</pre><h4>Type of Window Functions</h4><ul><li><strong><em>Aggregate Window Functions</em></strong><em>: SUM(), MAX(), MIN(), AVG(). COUNT()</em></li><li><strong><em>Ranking Window Functions</em></strong><em>: RANK(), DENSE_RANK(), ROW_NUMBER(),</em></li><li><strong><em>Paging Window Functions</em></strong><em>: NTILE() (splits the data into chucks)</em></li><li><strong><em>Fetch/Value Window Functions</em></strong><em>: Relatives: LAG(), LEAD(), Absolutes: FIRST_VALUE(), LAST_VALUE()</em></li></ul><h4>Window Functions Sub-clause</h4><p>Sub-clause are used to change the behavior of window functions.</p><ul><li><strong><em>PARTITION BY: </em></strong><em>Defines the set of rows or window for window functions. If PARTITION BY is not specified, grouping will be done on entire table and values will be aggregated accordingly.</em></li><li><strong><em>ORDER BY : </em></strong><em>Sorts the rows within each partition.</em></li></ul><h4>Window Frame</h4><p>Window frames indicate specific number of rows around the current row that a window function should capture. The syntax for a window frame is ROWS BETWEEN X AND Y specifying the start of the frame and the end of the frame.</p><pre>SELECT month,<br>       sales,<br>       <em>--- Calculate the 3-month moving average sales</em><br>       <strong>Avg</strong>(sales)<br>         over (<br>           ORDER BY month ASC <strong>ROWS BETWEEN </strong>2 preceding<strong> AND </strong>CURRENT ROW) AS<br>       MA_Sales<br>FROM   sales_table</pre><pre>Others Frames:<br><strong>ROWS BETWEEN</strong> n PRECEDING <strong>AND </strong>n FOLLOWING: Capture n row before the current rows and n rows after the current row.</pre><pre><strong>ROWS BETWEEN</strong> current row <strong>AND n </strong>following: Capture current row and the next row</pre><pre><strong>ROWS BETWEEN</strong> UNBOUNDED PRECEDING <strong>AND</strong> CURRENT ROW: Capture Current Row and all the rows before it.</pre><pre><strong>ROWS BETWEEN</strong> CURRENT ROW <strong>AND </strong>UNBOUNDED Following: Capture Current Row and all the rows after it.</pre><h4>Pivoting</h4><pre><strong>CREATE EXTENSION IF NOT EXISTS tablefunc;</strong></pre><pre><strong>SELECT * FROM CROSSTAB($$</strong></pre><pre>WITH temp_table name AS <br>(Select *<br>from a table<br>Where<br>Group by)</pre><pre>Select *<br>from temp_table;</pre><pre><strong>$$) AS ct ( Column1 VARCHAR, &quot;Column2&quot; Integer)</strong><br>Order by Column1 ASC;</pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=13d33953cde1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-managers-guide-to-sql-window-functions-13d33953cde1">A Product Manager’s guide to SQL Window Functions</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager’s Guide to Machine Learning: Tenets]]></title>
            <link>https://medium.com/vijay-patha/a-product-managers-guide-to-machine-learning-tenets-6fbfc386f178?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/6fbfc386f178</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Sat, 22 Aug 2020 16:46:59 GMT</pubDate>
            <atom:updated>2020-08-22T16:46:59.085Z</atom:updated>
            <content:encoded><![CDATA[<p>What and Why of tenets for Machine Learning(ML) Products</p><p>This article reviews my experience with a very ambitious and ambiguous ML project and summarizes my learning with an easy to adopt one-page plan.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DxlzxWNTUpSahJzi" /><figcaption>Photo by <a href="https://unsplash.com/@joshstyle?utm_source=medium&amp;utm_medium=referral">JOSHUA COLEMAN</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>Situation</h4><p>A predominant consensus of key stakeholders on a recently envisioned ambitious ML project was that it is grossly ambiguous. My first week on the the job, I heard stakeholders describe this project as: “it is a great idea, but what do we do with it”. “It is a cauldron of amorphous ideas” . “Everyone has their own thoughts about it because there is lot of visibility on it.”</p><h4>Complication</h4><p>These are the symptoms that come from calling an idea a project. Ideas, by nature, are amorphous. A Product Manager’s job is to lead an idea to reality by bringing clarity to the vision and a destination; all the while by gaining the alignment of the leadership and cross functional teams. When everyone has their own thoughts about the idea, it is a clear indication that there is no consensus on the deliverable or the deliverable is undefined. As a result, you will be bogged down with an inefficient decision making process.</p><h4>Question</h4><p>So, how do you bring shape to an amorphous idea in a product setting?</p><p>In the past, like other product managers, I would have focused on creating a roadmap that describes the timeline of feature launches. A roadmap brings some certainty and that puts people in a “zen mode”. In the last month, I learned that creating a roadmap should be the least of your problems in this situation. You still need to put the roadmap in front of people ASAP but for you, a leader of the program, that should be last thing you should work on when there is no consensus on the vision. When you don’t know the destination, you will not know how good your plan is to get to the destination.</p><p>Here is an idea to “bring all the ducks in the row” on your project:</p><h4>Solution</h4><p>Write a Page-0 for your project. That’s right, I said write a document with prose, paragraphs, punctuation and all. This doesn’t mean creating a slide or writing bullet points.</p><p><strong>What is Page-0?</strong> — Think of it is a focused one-pager that clearly communicates to key stakeholders what your project is, what problem it solves, and how. The objective of this page is to bring alignment.</p><p>Here is the outline for page 0.</p><ol><li>Purpose — What is the purpose of the document and what specific feedback are you seeking.</li><li>Define your project in 4–5 sentences. If you can’t, then you don’t understand the project, and need to work to refine the definition.</li><li>Tenets — Beliefs that guide the project and team.</li><li>Themes — These are problems that you promise to solve.</li><li>Roadmap — Theme based roadmap in Now, Next, Later format. Provide dates for those items in “Now”, these dates will bring some needed certainty to stakeholders. “Next” could be next quarter or so.</li></ol><p>Getting alignment should be ruthlessly prioritized. Get feedback.</p><h4>More about Tenets</h4><p>A tenet is a principle or belief that guides the project and team. Because your actions are guided by your beliefs, Tenets should appear at the beginning of narratives, in roadmaps, and in other documents. This will help your audience know what is guiding your actions. By telling the reader how you approach problems, you tell them a lot about your space.</p><p>Developing a good set of tenets requires relentless scrutiny and questioning. It is incredibly hard work. It is ok to have tension between two or more tenets. Debate with your team and get alignment. At Amazon, we aim to have less than 7 tenets.</p><p>Here are the three tenets that I wrote for my project._________ is redacted information.</p><p>1. Because ______ happens at both a macro and micro level, ____ doesn’t take one-size fits all solution. ____ adapts ____ definition based on the use cases by leveraging individual and/or blended components.</p><p>2. ______ biases towards machine intelligence but does not back away from using heuristics where needed.</p><p>3. ____ prioritizes and relies on ground-truth to define and refine ______, components, and use cases.</p><h4>Conclusion</h4><p>This format can be applied to most projects, not only ML. Ironically, though a science based project, ML projects are very ambiguous. Stake in the success of the project with tenets.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6fbfc386f178" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-managers-guide-to-machine-learning-tenets-6fbfc386f178">A Product Manager’s Guide to Machine Learning: Tenets</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Product Manager’s Guide to Machine Learning: ECDFs]]></title>
            <link>https://medium.com/vijay-patha/a-product-managers-guide-to-machine-learning-ecdfs-b66ce4a99349?source=rss----81e48ddf1579---4</link>
            <guid isPermaLink="false">https://medium.com/p/b66ce4a99349</guid>
            <category><![CDATA[ecdf]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Vijay Patha]]></dc:creator>
            <pubDate>Wed, 08 Jul 2020 21:24:57 GMT</pubDate>
            <atom:updated>2020-07-09T19:39:43.933Z</atom:updated>
            <content:encoded><![CDATA[<h4>Empirical cumulative distribution functions (ECDFs)</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lm6VuO9bdBh-YrJZ" /><figcaption>Photo by <a href="https://unsplash.com/@ignitedit?utm_source=medium&amp;utm_medium=referral">Mark Basarab</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Here is an another tool to add to Exploratory Data Analysis (EDA) toolkit. The ECDF maps every data point in the dataset to a quantile. Here is the code that you can use to create Histogram and ECDF:</p><pre>import pandas as pd<br>import numpy as np<br>import matplotlib.pyplot as plt<br>df = pd.read_csv(&#39;ecdf.csv&#39;, names=[&quot;sales&quot;])<br>df.head()</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/134/1*u9SnaXXfZmYpP1CI44UhyA.png" /></figure><h3>Percentiles</h3><pre>data = df.to_numpy()</pre><pre>percentiles = np.array([2.5,25,50,75,100])</pre><pre># Compute percentiles: ptiles_sales<br>ptiles_sales=np.percentile(data,percentiles)<br>print(ptiles_sales)</pre><pre>[1.395500e+02 2.691800e+04 4.545300e+04 8.829950e+04 1.590546e+06]</pre><h3>Histogram</h3><pre># Import numpy<br>import numpy as np</pre><pre># Compute number of data points: n_data<br>n_data = len(data)</pre><pre># Number of bins is the square root of number of data points: n_bins<br>n_bins = np.sqrt(n_data)</pre><pre># Convert number of bins to integer: n_bins<br>n_bins = int(n_bins)</pre><pre># Plot the histogram<br>plt.hist(data, bins=n_bins)</pre><pre># Label axes<br>_ = plt.xlabel(&#39;Sales&#39;)<br>_ = plt.ylabel(&#39;count&#39;)</pre><pre># Show histogram<br>plt.show()</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/653/1*6gFgQUC6SVyM_PoiQmT15A.png" /></figure><h3>ECDF</h3><p>ECDFs stand for the “empirical cumulative distribution function”, and they map every data point in the dataset to a quantile, which is a number between 0 and 1 that indicates the cumulative fraction of data points smaller than that data point itself.</p><p>It takes as input a 1D array of data and then returns the x and y</p><pre>def ecdf(a,b):<br>    &quot;&quot;&quot;State what function does here&quot;&quot;&quot;<br>    # Computation performed here<br>    return x, y</pre><p>Your ecdf() function returns two arrays so you will need to unpack them. So unpacking is x, y = ecdf(data)</p><ul><li>The x-values are the sorted data. Use the np.sort() function to perform the sorting.</li><li>The y data of the ECDF go from 1/n to 1 in equally spaced increments. You can construct this using np.arange(). Since np.arange() is not inclusive, will need to go from 1 to n+1</li></ul><h4>ECDF Function</h4><pre>def ecdf(data):<br> ””” Compute ECDF for a one-dimensional array of measurements.”””<br> # Number of data points: n<br> n = len(data)</pre><pre># x-data for the ECDF: x<br> x = np.sort(data)</pre><pre># y-data for the ECDF: y<br> y = np.arange(1, n+1) / n</pre><pre>return x, y</pre><h4>Plot Sales using ECDF</h4><pre># Compute ECDF for sales data: x_sales, y_sales<br>x_sales, y_sales = ecdf(data)<br># Generate plot<br>plt.plot(x_sales, y_sales, marker = &#39;.&#39;, linestyle = &#39;none&#39;)<br># Label the axes<br>plt.xlabel(&#39;sales&#39;)<br>plt.ylabel(&#39;ECDF&#39;)<br># Overlay percentiles as red diamonds.<br>plt.plot(ptiles_sales, percentiles/100, marker=&#39;D&#39;, color=&#39;red&#39;,<br>         linestyle=&#39;none&#39;)<br># Display the plot<br>plt.show()</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/631/1*znLr7KQWlrAmDL-gcUeElQ.png" /><figcaption>Red dots are the percentiles.</figcaption></figure><h3>Conclusion</h3><p>ECDFs provide forest landscape view of percentiles, values, outliers of a feature.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b66ce4a99349" width="1" height="1" alt=""><hr><p><a href="https://medium.com/vijay-patha/a-product-managers-guide-to-machine-learning-ecdfs-b66ce4a99349">A Product Manager’s Guide to Machine Learning: ECDFs</a> was originally published in <a href="https://medium.com/vijay-patha">vijay-patha</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>