<?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 Channabasava H on Medium]]></title>
        <description><![CDATA[Stories by Channabasava H on Medium]]></description>
        <link>https://medium.com/@channachannu.97?source=rss-76a97bcb3541------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*-s-vRxkBV25K407o</url>
            <title>Stories by Channabasava H on Medium</title>
            <link>https://medium.com/@channachannu.97?source=rss-76a97bcb3541------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 19:54:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@channachannu.97/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[Designing AI Around Evidence, Not Confidence]]></title>
            <link>https://medium.com/learning-data/designing-ai-around-evidence-not-confidence-1049aab687f0?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/1049aab687f0</guid>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Mon, 17 Aug 2026 10:07:36 GMT</pubDate>
            <atom:updated>2026-09-02T17:32:36.156Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tEVSebNvAjMg9_i9l9JbaQ.png" /></figure><p>Most AI systems are built around a simple expectation: ask a question and get an answer.</p><p>For financial analysis, I started wondering whether that is always the right behaviour.</p><blockquote>Sometimes, the better answer is simply — <strong>the evidence isn’t there.</strong></blockquote><p>I have been building a small IPO analysis system that answers questions using only a company’s regulatory filing. The idea is deliberately narrow. No market opinions, no assumptions, and no filling in missing information from what the underlying model may already know.</p><p>The system works from a structured knowledge base created from the filing. Each fact carries its source page, section and disclosure status. The language model sits at the end of the process and narrates only the information that survives the earlier checks.</p><h3>The model doesn’t get the final say</h3><p>One lesson from earlier experiments was that asking a model to <em>please stay within the provided context</em> isn’t a particularly strong control.</p><p>So I moved that responsibility outside the model.</p><p>The current system uses a three-stage pipeline.</p><p>First, certain questions such as investment recommendations, future projections and fair-value opinions are blocked deterministically.</p><p>For legitimate questions, the model can identify which part of the knowledge base may contain the answer. But that choice is then validated in code. If the requested information does not actually exist, it never reaches the final answering step.</p><p>Python also handles the numbers. The language model does not perform arithmetic or invent missing values.</p><blockquote>Its job is much smaller: <strong>Explain the evidence it has been given.</strong></blockquote><h3>A surprisingly useful failure</h3><p>During testing, I deliberately asked questions that a general-purpose model could easily answer from its training data.</p><p>A common financial ratio was one example.</p><p>But if the relevant definition wasn’t available within the information I had allowed the system to use, it didn’t quietly reach into general knowledge and complete the answer.</p><blockquote>It stopped — That was exactly the behavior I wanted.</blockquote><p>The distinction becomes even more important with company-specific information. Understanding what a financial ratio means is one thing. Claiming what that ratio is for a particular company is something entirely different.</p><p>The second requires evidence.</p><h3>Why this matters</h3><p>In many applications, a plausible answer may be good enough.</p><p>Financial documents are less forgiving.</p><p>A number that looks reasonable but cannot be traced back to the filing can easily be mistaken for a disclosed fact. An estimate can start looking like evidence simply because it was written confidently.</p><p>That makes restraint part of the system design.</p><p>The goal is not to make the model less capable. It is to decide where that capability is allowed to operate.</p><p>Most of my career has involved getting systems to do more. This project has been an interesting reversal: part of the value comes from making the system deliberately do less.</p><p>For financial AI, reliability is not only about producing the correct answer.</p><blockquote>It is also about controlling what the system is allowed to claim.</blockquote><p>And sometimes, <strong>“the filing doesn’t provide enough evidence”</strong> is exactly the answer the system should give.</p><p>Thank you for reading…</p><p><em>The contents of external submissions are not necessarily reflective of the opinions or work of </em><a href="http://mavenanalytics.io/"><em>Maven Analytics</em></a><em> or any of its team members.</em></p><p><em>We believe in fostering lifelong learning and our intent is to provide a platform for the data community to share their work and seek feedback from the Maven Analytics data fam.</em></p><p><a href="https://medium.com/learning-data/how-to-get-your-work-published-by-learning-data-with-maven-analytics-7df21e466a3e?source=friends_link&amp;sk=020dfac485597d602e218968d9ffb395"><em>Submit your own writing here</em></a><em> if you’d like to become a contributor.</em></p><p><em>Happy learning!</em></p><p><em>-Team Maven</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1049aab687f0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/learning-data/designing-ai-around-evidence-not-confidence-1049aab687f0">Designing AI Around Evidence, Not Confidence</a> was originally published in <a href="https://medium.com/learning-data">Learning Data</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Simple Mental Model for Modern AI Infrastructure]]></title>
            <link>https://medium.com/learning-data/a-simple-mental-model-for-modern-ai-infrastructure-733bd90f5a9c?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/733bd90f5a9c</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[agentic-rag]]></category>
            <category><![CDATA[ai-infrastructure]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Sat, 01 Aug 2026 14:00:28 GMT</pubDate>
            <atom:updated>2026-09-02T17:29:39.203Z</atom:updated>
            <content:encoded><![CDATA[<h4>The Six-Layer Map I Wish I Had as a Beginner</h4><p>If you follow AI discussions or news, you’ve probably encountered terms like &quot;LLM,&quot; &quot;Claude,&quot; &quot;GPT,&quot; &quot;Ollama,&quot; &quot;LangChain,&quot; &quot;RAG,&quot; and &quot;GraphRAG&quot;—without a clear sense of where they fit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WFrRkGv5ZOHUycrL5H2GWw.png" /></figure><blockquote>It sounds like a lot. It isn’t, actually. The problem isn’t that these things are individually hard — it’s that nobody ever draws you the map of how they fit together. So let’s just draw the map.</blockquote><h3>Why this feels confusing (it’s not you)</h3><p>Every one of those words describes a completely different <em>kind</em> of thing. Some are models. Some run models. Some serve models over an API. Some coordinate the whole app. Some are just strategies for finding information before answering a question.</p><p>Mixing all of these into one sentence—&quot;we used LangChain with RAG on top of vLLM serving a Llama model”—is like describing a restaurant by saying “the menu, the chef, the delivery app, and the gas line” all in one breath. Technically all real, all connected, and all completely different jobs.</p><h3>The full stack, top to bottom</h3><p>Here’s the entire thing. Every AI app, no matter how fancy, is basically this stack:</p><ol><li><strong>The App</strong> — where you type your question. A chatbot, a website, a CLI, whatever.</li><li><strong>Orchestration</strong> — decides what happens next. Should we look something up? Call a tool? Just answer? This is LangChain, LangGraph, or LlamaIndex.</li><li><strong>Retrieval</strong> — goes and finds actual evidence before answering. This is RAG or GraphRAG.</li><li><strong>Model serving</strong> — loads the AI model and runs it efficiently. Ollama, vLLM, SGLang, llama.cpp.</li><li><strong>The model (LLM)</strong> — the actual trained network. Llama, Qwen, Mistral, GPT, Claude.</li><li><strong>Hardware</strong> — CPU/GPU. The thing quietly paying for all of this.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CyZfJnNsGEcU4m4lcDv_UQ.png" /></figure><blockquote>That’s the whole map. Every confusing tool name you’ve ever seen fits into exactly one of these six boxes. Once you know the box, the tool stops being intimidating — it’s just “oh, that’s an orchestration thing” instead of a mystery word.</blockquote><p>One honest caveat: real tools don’t always respect these boundaries neatly. llama.cpp does inference <em>and</em> has its own little server built in. vLLM handles serving <em>and</em> dips into orchestration-adjacent territory. Think responsibilities, not strict rulebooks, and you’ll be fine.</p><p>Here’s that same map, one card per layer:</p><h3>Layer 1: The model isn’t the whole app</h3><p>An LLM — Llama, Qwen, Mistral, GPT, Claude — is just a trained neural network that predicts the next word, over and over, extremely fast. By itself, it’s a file. It can’t take your question, can’t run itself, can’t do anything until something else loads it up and starts feeding it prompts.</p><blockquote>This trips people up constantly: “we have access to GPT” does not mean “we have a working app.” You still need everything below it.</blockquote><h3>Layer 2: Who actually runs the model</h3><p>This is the model-serving layer — the software between “a trained model exists” and “you can actually talk to it.”</p><ul><li><strong>llama.cpp</strong> — runs models locally, even on a regular laptop CPU, especially compressed (“quantized”) versions. Great for offline or budget setups.</li><li><strong>vLLM</strong> — built to handle lots of requests at once, production-grade, usually GPU-powered.</li><li><strong>SGLang</strong> — optimized for speed and more complex, multi-step generation, also production-grade.</li><li><strong>Ollama</strong> — the easy button. Downloads models, manages them, gives you a simple local API to call. It is not the model itself, and it’s not raw low-level inference — it’s the convenience layer sitting on top of it.</li></ul><h3>Layer 3: Who decides what happens next</h3><p>LangChain and LangGraph don’t do any of the actual “thinking” — no neural network math happens inside them. They just decide the sequence: check this, call that, then respond.</p><ul><li><strong>LangChain</strong> gives you pre-built pieces to plug together — tools, integrations, model calls.</li><li><strong>LangGraph</strong> is for when the steps aren’t a straight line — there’s branching, memory, maybe looping back before it’s actually done.</li></ul><h3>Layer 4: Getting the model real information first</h3><p>This is where <strong>RAG</strong> and <strong>GraphRAG</strong> live, and it’s probably the single most misunderstood part of the whole stack—because neither one is a model, a database, or a serving engine. They’re both just <em>strategies for finding good evidence</em> before the model writes an answer.</p><p><strong>RAG (Retrieval-Augmented Generation):</strong> search for relevant text — usually using something called vector search, which finds passages that mean something similar to your question — and staple that text onto the prompt before sending it to the model. Simple, effective, and the reason a chatbot can answer questions about stuff way outside what it was originally trained on.</p><p><strong>GraphRAG:</strong> Instead of grabbing loose, similar-sounding chunks of text, it pulls a <em>connected map</em> of facts—entities and the relationships between them—usually from something called a knowledge graph (Neo4j is a common tool for this). It’s the difference between handing someone a stack of relevant articles versus handing them a diagram that already shows how everything connects.</p><p>Here’s the one sentence worth remembering from this entire post:</p><p><strong>RAG and GraphRAG never change how the model “thinks.” They only change what it’s given to think with.</strong> Same model, same hardware, same underlying math — just better material in front of it.</p><h3>Layer 5: When AI starts making its own decisions—&quot;agentic AI”</h3><p>Everything so far has been mostly fixed steps: get a question, maybe retrieve something, send it to the model, get an answer. An <strong>agent</strong> is different—it&#39;s a setup where the model gets to decide, on its own, what to do next: search for something, run a calculation, call another tool, or ask a follow-up question—and keep looping through these choices until an actual goal is met, not just until one step is technically finished.</p><p>Agents live in the orchestration layer. They use retrieval, model serving, and the LLM as tools toward a goal, instead of following a fixed script top to bottom. LangGraph shows up a lot here specifically because it’s built for this kind of branching, multi-step, sometimes-looping process.</p><h3>Quick myths, busted</h3><ul><li><strong>“Ollama is an LLM.”</strong> Nope — Ollama runs and manages models. The LLM is what it’s running.</li><li><strong>“RAG talks directly to the model’s inner workings.”</strong> Nope — RAG just preps a finished prompt, which then goes through the serving layer like any other request.</li><li><strong>“GraphRAG replaces the LLM.”</strong> Nope — it just hands the LLM better material to work with.</li><li><strong>“GraphRAG is always the better choice.”</strong> Also nope. For a quick, simple fact lookup, plain RAG is faster and cheaper. Save the relationship-graph approach for genuinely connected, multi-step questions.</li><li><strong>“Every tool fits neatly into one box.”</strong> Mostly, but not perfectly — remember the caveat above. Use this map as a mental model, not a legal contract.</li></ul><h3>The whole thing, compressed</h3><blockquote><strong>LLM</strong> = the model. Can’t run itself.</blockquote><blockquote><strong>Serving layer</strong> (Ollama/vLLM/SGLang/llama.cpp) = runs the model.</blockquote><blockquote><strong>Orchestration</strong> (LangChain/LangGraph) = decides the steps.</blockquote><blockquote><strong>RAG/GraphRAG</strong> = fetch good evidence before answering.</blockquote><blockquote><strong>Agents</strong> = orchestration that makes its own calls, in a loop, toward a goal.</blockquote><blockquote><strong>Hardware</strong> = pays the electricity bill.</blockquote><p>Once this map is in your head, new AI tool names stop being scary. You’ll just ask yourself, &quot;Which of these six boxes does this thing actually belong to?” — and nine times out of ten, you’ll already know the answer.</p><h4><strong>Final Thoughts</strong></h4><p>And that’s the stack. The next time someone says, “We’re using RAG with a LangGraph agent on vLLM,” you won’t have to nod thoughtfully and hope nobody asks a follow-up. You’ll know they mean: retrieve the right information, coordinate the workflow, and run the model efficiently.</p><blockquote>Go ahead — use “orchestration layer” in a meeting like it has always been part of your vocabulary.</blockquote><blockquote><em>Thank you for Reading</em> 🙌 <em>and Keep an eye out for similar blogs</em>!!</blockquote><p><strong>Editorial note:</strong> This guide presents a responsibility-based view of AI infrastructure based on my current understanding. It is intended as a learning framework, not a representation of universal industry practice. Because product capabilities evolve quickly, always validate deployment decisions against the latest official documentation.</p><p><em>The contents of external submissions are not necessarily reflective of the opinions or work of </em><a href="http://mavenanalytics.io/"><em>Maven Analytics</em></a><em> or any of its team members.</em></p><p><em>We believe in fostering lifelong learning and our intent is to provide a platform for the data community to share their work and seek feedback from the Maven Analytics data fam.</em></p><p><a href="https://medium.com/learning-data/how-to-get-your-work-published-by-learning-data-with-maven-analytics-7df21e466a3e?source=friends_link&amp;sk=020dfac485597d602e218968d9ffb395"><em>Submit your own writing here</em></a><em> if you’d like to become a contributor.</em></p><p><em>Happy learning!</em></p><p><em>-Team Maven</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=733bd90f5a9c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/learning-data/a-simple-mental-model-for-modern-ai-infrastructure-733bd90f5a9c">A Simple Mental Model for Modern AI Infrastructure</a> was originally published in <a href="https://medium.com/learning-data">Learning Data</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[✨ Modelling Stock Market Regimes with Markov Chains: A Practical, Data-Driven Study]]></title>
            <link>https://wire.insiderfinance.io/modelling-stock-market-regimes-with-markov-chains-a-practical-data-driven-study-712d98300ccf?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/712d98300ccf</guid>
            <category><![CDATA[markov-chains]]></category>
            <category><![CDATA[quantitative-trading]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[stock-market]]></category>
            <category><![CDATA[algorithmic-trading]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Sun, 16 Nov 2025 02:29:53 GMT</pubDate>
            <atom:updated>2025-11-17T14:09:16.602Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>How I Used Markov Chains to Predict Stock Market Regimes (And Why </strong>62%<strong> Accuracy Is a Big Deal)</strong></h4><p><em>A practical guide for researchers, analysts, and curious finance enthusiasts on Quantitative Trading.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*V5zc2b77egH5xDRAwgcSQQ.png" /><figcaption>Modelling Stock Market Regimes with Markov Chains</figcaption></figure><h3>👋<strong> Introduction</strong></h3><p>If you’ve ever looked at the stock market and <strong><em>Thought</em></strong>:</p><blockquote>“Why does my portfolio act like it’s in a long-distance relationship with my hopes?” — welcome to the club.</blockquote><p>This article is a practical walk-through of a project where I used <strong>Markov chains</strong> to model daily <strong>market states</strong> for a particular stock and build a probabilistic forecast system.</p><p><strong>The objective?</strong></p><blockquote>Understand the stock’s behavioural tendencies and evaluate whether tomorrow’s regime can be inferred probabilistically from today’s.</blockquote><p>If that sounds technical — don’t worry, I’ll break it down like I’m explaining it to my cousin who still thinks SIP is an energy drink.</p><p>By the end of this article, you will understand:</p><ol><li>How to classify market days into regimes</li><li>How to Build a Markov model on real stock data</li><li>How to Predict Next Possible Stock Movement</li><li>How to Backtest Predictions</li><li>And how to generate early warning risk signals</li></ol><h3>🧩<strong> 1. The Foundation — Identifying Market States</strong></h3><p>Before building any predictive model, we selected “<strong>SunPharma</strong>” <strong>(SUNPHARMA.NS)</strong> Stock data from the last 24 months and needed to classify each trading day into a <strong>state</strong>.</p><p>Think of states as the “moods” of the stock — because stocks, like humans, have them.</p><p>We used metrics such as:</p><ol><li>Daily % Change</li><li>ATR and ATR% (volatility)</li><li>Volume Change</li><li>Price deviations from MA20 (momentum)</li></ol><p>This gave us five meaningful regimes:</p><pre>✔ Uptrend - Positive momentum, above MA20, controlled volatility.<br>✔ Downtrend - Negative momentum, weakness below MA20, falling volume.<br>✔ Neutral - Sideways consolidation. (The stock-market equivalent of &quot;I don&#39;t know what I want, but I&#39;m fine&quot;.)<br>✔ Reversal - Turning points where the market switches direction. Fun to analyse, dangerous to predict.<br>✔ Volatile - Large ranges, high ATR, big candles. Your stop loss&#39;s worst nightmare.</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/993/1*-sOG2IcPOFptfST7nLkfxA.png" /><figcaption>Stock market regimes</figcaption></figure><h3>🎯<strong> Calculating Volatility — Average True Return (ATR)</strong></h3><p>We computed <strong>True Range (TR)</strong>:</p><pre>TR = max(High-Low, |High-PrevClose|, |Low-PrevClose|)</pre><p>And then smoothed it using an <strong>exponential moving average</strong> (EWM) to. avoid spiky, unstable volatility measures:</p><pre>ATR = TR.ewm(span=14).mean()</pre><p>This makes volatility responsive — perfect for fast-changing market conditions and a <strong>Responsive volatility estimator</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/450/1*v_SnygkF-vaz-2ZsSUbeEQ.png" /><figcaption>Average True Return and Volatility</figcaption></figure><h3>🧠<strong> Building the Markov Chain</strong></h3><p>Markov chains help answer one simple question:</p><blockquote>“Given today’s state, what is the probability of tomorrow’s state?”</blockquote><p>To do this, we:</p><ol><li>Took ~24 months of historical states</li><li>Counted all transitions (e.g., Neutral → Uptrend)</li><li>Normalised rows to build the <strong>transition probability matrix</strong></li></ol><p>Here’s the matrix we used:</p><pre>| From/To   | Downtrend | Neutral   | Reversal | Uptrend | Volatile  |<br>| --------- | --------- | --------- | -------- | ------- | --------- |<br>| Downtrend | 22.8%     | **45.7%** | 8.5%     | 20%     | 2.8%      |<br>| Neutral   | 9.2%      | **52.0%** | 8.6%     | 27%     | 2.8%      |<br>| Reversal  | 15.1%     | **36.3%** | 9.0%     | 27.2%   | 12.1%     |<br>| Uptrend   | 6.0%      | **47.4%** | 10%      | 35%     | 1%        |<br>| Volatile  | 0%        | 4.6%      | 1.3%     | 1.3%    | **92.7%** |</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/920/1*BFTSi0Hv-I_F7NFNd0D_cQ.png" /><figcaption>Probability of Next State from given Current State</figcaption></figure><h3>🔥<strong> The First Big Insight — Neutral Dominates</strong></h3><p>This stock loves being Neutral. If this stock were a person, its relationship status would be “It’s complicated… with the sideways market.”</p><ul><li>Neutral is the <em>gravitational centre</em></li><li>Most states revert to Neutral</li><li>Downtrends rarely persist</li><li>Volatile days cluster heavily</li></ul><p>This is the <strong>exact behaviour</strong> for a fundamentally stable, liquid stock.</p><h3>🧪 Backtesting the Markov Predictions</h3><p>We validated the model by comparing:</p><blockquote><strong><em>Predicted state (based on today)</em></strong><em><br> vs<br> </em><strong><em>Actual state (next day)</em></strong></blockquote><blockquote>🥇<strong> 62% Predictive Accuracy</strong></blockquote><p>Which, in regime modelling terms, is <strong>very good</strong>. Anything above 55–60% is considered meaningful. And contrary to what Instagram trading gurus say, a 62% probabilistic edge <strong>can beat the market</strong> when used correctly.</p><h3>🧾 Conclusion</h3><p>Markov models don’t predict prices.</p><blockquote>They predict <strong>behavior</strong> — the <em>regimes</em> that shape price movement.</blockquote><p>This project is just one way to turn raw data into intuition. I hope it inspires you to experiment, explore, and maybe build your own market model.</p><blockquote>Thank you for Reading<em> 🙌 </em>and Keep an eye out for Part 2 <em>😉!!</em></blockquote><h4>A Message from InsiderFinance</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/301/0*QCVCC4NQvshiTPdI.png" /></figure><p>Thanks for being a part of our community! Before you go:</p><ul><li>👏 Clap for the story and follow the author 👉</li><li>📰 View more content in the <a href="https://wire.insiderfinance.io/">InsiderFinance Wire</a></li><li>📚 Take our <a href="https://learn.insiderfinance.io/p/mastering-the-flow">FREE Masterclass</a></li><li><strong>📈 Discover </strong><a href="https://insiderfinance.io/?utm_source=wire&amp;utm_medium=message"><strong>Powerful Trading Tools</strong></a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=712d98300ccf" width="1" height="1" alt=""><hr><p><a href="https://wire.insiderfinance.io/modelling-stock-market-regimes-with-markov-chains-a-practical-data-driven-study-712d98300ccf">✨ Modelling Stock Market Regimes with Markov Chains: A Practical, Data-Driven Study</a> was originally published in <a href="https://wire.insiderfinance.io/">InsiderFinance Wire</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ From Clicks to Journeys: Using Probability to Understand Campaign Performance]]></title>
            <link>https://medium.com/learning-data/from-clicks-to-journeys-using-probability-to-understand-campaign-performance-a298c248e0a9?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/a298c248e0a9</guid>
            <category><![CDATA[online-business]]></category>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[digital-marketing]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Thu, 25 Sep 2025 23:51:51 GMT</pubDate>
            <atom:updated>2025-09-25T23:51:51.698Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>Marketing campaigns live and die by numbers.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/350/1*sWrOlg5Dpb9rdGAbV6Yjvg.png" /></figure><p>We send thousands of emails… But beneath the surface of simple “open rates” and “click-through rates” lie powerful probability models that can transform you from a <strong>reporter of data</strong> into a <strong>strategic architect of success</strong>.</p><p>Let’s explore three lenses for looking at campaign outcomes: <strong>Binomial distribution, Poisson distribution, and Markov chains</strong>. Each tells a different part of the story.</p><h3>📊 Binomial Distribution: The Basics of “How Many”</h3><p>The Binomial distribution is your basic yes-or-no coin flip. Instead of heads or tails, your customer either <strong>shops (success)</strong> or <strong>doesn’t shop (failure)</strong> 🛒.</p><ul><li><strong>Business framing</strong>: If you send 50 emails and your historical shop probability is 80%, the Binomial model tells you the likelihood of getting exactly 30, 40, or 50 purchases.</li><li><strong>Why it matters</strong>: It quantifies the <em>risk of underperformance</em>. If you expected 40 buyers, but only got 30, Binomial shows that this outcome had only a 0.1% probability. In other words, it’s a red flag worth investigating.</li></ul><blockquote><strong>Takeaway</strong>: Use Binomial when you want to validate whether your campaign results match expectations. It answers the question:<br> 👉 <em>“How likely was it to get this many shoppers, given what I know?”</em></blockquote><p>But what happens when your success rate is so tiny that the Binomial math gets clunky? We turn to Poisson</p><h3>🔢 Poisson Distribution: When Events Are Rare</h3><p>Sometimes the math of Binomial gets messy — especially when you have a large audience but very small conversion rates. That’s where Poisson shines.</p><ul><li><strong>Business framing</strong>: Imagine sending 100 emails with a 0.5% shop rate. You expect about 0.5 purchases. Poisson helps you model the probability of seeing exactly 0, 1, or 2 shoppers.</li><li><strong>Why it matters</strong>: It’s perfect for forecasting “rare but important” events. In campaigns, that might mean low-probability actions such as high-value product purchases, survey responses, or referral sign-ups.</li></ul><blockquote><strong>Takeaway</strong>: Use Poisson for quick forecasting when success rates are tiny but audience sizes are large. It answers:<br> 👉 <em>“Given a very low response rate, what’s the chance of seeing 0, 1, or 2 conversions?”</em></blockquote><h3>🔄 Markov Chains: Beyond Counts, Into Journeys</h3><p>Binomial and Poisson stop at counting. But customers don’t just “buy” or “not buy” — they move through a <strong>funnel</strong>: open, click, browse, purchase. Markov chains give you a map of that journey.</p><ul><li><strong>Business framing</strong>: In one campaign with 100 customers, 70 never open the email, 30 open it, but 22 stop there, 8 click, but only 3 go on to purchase.</li><li><strong>How:</strong> While Binomial just tells you <strong>how many</strong> ultimately bought, Markov tells you the story of <strong>the ones who quit.</strong> It reveals, for example, that the biggest probability drop-off isn’t between the click and the purchase — it’s between <strong>receiving the email and opening it.</strong> This immediately signals that your time is better spent optimising <strong>subject lines</strong> than optimising the checkout page.</li><li><strong>Why it matters</strong>: Markov chains reveal <strong>where customers drop off</strong>. Maybe your subject lines are strong (good open rates), but your CTA is weak (few clicks). Or maybe you drive clicks but fail to close sales.</li></ul><blockquote><strong>Takeaway</strong>: Use Markov chains to optimize the funnel. It answers:<br> 👉 <em>“Where exactly do customers fall out, and how would improving one step shift the outcome?”</em></blockquote><h3>💡 Bringing It Together</h3><ul><li><strong>Binomial</strong> helps you test whether campaign results are within the realm of expectation.</li><li><strong>Poisson</strong> helps you forecast rare events quickly and simply.</li><li><strong>Markov chains</strong> help you see the bigger picture: the customer journey across touchpoints.</li></ul><blockquote>Together, they move you from <em>counting conversions</em> to <em>understanding behavior</em>.</blockquote><blockquote>And in marketing, that shift is the difference between reporting yesterday’s numbers and actively designing tomorrow’s success.</blockquote><p>✨ <strong>Closing thought</strong>:<br> Next time you’re analysing a campaign, don’t just ask</p><blockquote>“How many bought?”. Ask “How did they get there, what are the odds, and where did we lose the rest?”.</blockquote><p>Probability models give you those answers — and in the process, make you a more strategic marketer.</p><p>Thank you for staying till the end! 😃</p><blockquote><em>If you enjoyed reading, please give it a CLAP! 🙌</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a298c248e0a9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/learning-data/from-clicks-to-journeys-using-probability-to-understand-campaign-performance-a298c248e0a9">🎯 From Clicks to Journeys: Using Probability to Understand Campaign Performance</a> was originally published in <a href="https://medium.com/learning-data">Learning Data</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hold My Stocks: The Art of Playing the Long Game (Without Losing Your Mind)]]></title>
            <link>https://wire.insiderfinance.io/hold-my-stocks-the-art-of-playing-the-long-game-without-losing-your-mind-c1f1ee3327c0?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/c1f1ee3327c0</guid>
            <category><![CDATA[portfolio-management]]></category>
            <category><![CDATA[asset-management]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[finance]]></category>
            <category><![CDATA[stocks]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Sun, 10 Aug 2025 18:20:19 GMT</pubDate>
            <atom:updated>2025-11-19T01:16:13.579Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0EHv8IKXm_3GzLk2HOvSGA.png" /></figure><p>Investing can feel like a wild ride — one minute you’re celebrating wins, the next you’re sweating over every dip. Should you sell and lock in those gains, or hold on for bigger wins? It’s the question every investor battles with.</p><blockquote>Let’s break it down with the <strong><em>Garden Approach</em></strong>…</blockquote><h4>Active vs Passive Investing — The Basics</h4><p>Think of your portfolio like a garden. An <strong>active investor</strong> is the gardener who’s always pruning, rearranging plants, and trying new seeds to get the best bloom. This means regularly checking your stocks, selling some, buying others, and rebalancing your mix.</p><p>On the other hand, a <strong>passive investor</strong> plants the garden once and waters it occasionally — no fuss, just steady care. They buy good stocks and hold them for the long haul, adding a little more now and then without disrupting their portfolio.</p><p>Both approaches have fans and critics. Active investing promises you can beat the market if you’re smart and quick. Passive investing relies on patience and letting your money grow over time with less stress.</p><blockquote>So, which one works better? Let’s meet Tom and Harry’s to find out..</blockquote><h3>Meet Tom and Harry — Two Gardeners, Two Strategies</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*gEugqfiyvsmtYnIVVnRT0g.png" /></figure><blockquote>Back in January 2017, Apple shares were priced at $30 each.</blockquote><ul><li>Tom, our steady gardener, bought 10 shares for $300. He held onto them,<strong><em> occasionally buying more</em></strong> shares every year. He followed the <strong><em>passive strategy.</em></strong></li><li>Harry, also a gardener but a bit more hands-on, bought the same 10 shares in 2017. Unlike Tom, he <strong><em>sold 5 shares</em></strong> by the end of that year at $42 each, booking a neat 40% return on that chunk. In 2018, he bought and sold shares again with 15% returns, <strong><em>actively managing his portfolio</em></strong> over the years.</li></ul><blockquote>Now, the big question: Who ended up with the fatter portfolio — Tom or Harry?</blockquote><p>Let’s look at the table below…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v09xeL4RC0lkwpkLCEhV1A.png" /><figcaption>Portfolio value ($) of Tom and Harry since 2017</figcaption></figure><p>Harry profited every year by selling a part of his portfolio, but if we visualise the portfolio value over the years, Tom’s approach gave better returns.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2gp08fo7SS_n3lVkk5d7wA.png" /></figure><p>Here’s what the numbers say about Tom’s and Harry’s investing approaches:</p><blockquote><strong>Tom’s Passive Strategy</strong></blockquote><ul><li><strong>Method: </strong>Tom bought Apple shares steadily — starting with 10 shares in 2017 and adding more each year — never selling, just accumulating.</li><li><strong>Outcome: </strong>His portfolio grew relentlessly, compounding both share count and value. By 2025, he held <strong>101 shares worth about $23,230</strong>.</li><li><strong>Mindset: </strong>Patience, discipline, and faith in long-term growth, with minimal tinkering.</li><li>Portfolio Value — $23K with no booked profits</li></ul><blockquote><strong>Harry’s Active Strategy</strong></blockquote><ul><li><strong><em>Method:</em> </strong>Harry started similarly but sold shares regularly to lock in profits throughout, notably in 2017, 2018, and subsequent years. He bought new shares each year but frequently booked gains.</li><li><strong><em>Outcome:</em> </strong>While he realised cash profits along the way via sales, his end portfolio was smaller — about <strong><em>49 shares worth $11,270</em></strong> — with profits booked separately each year.</li><li><strong><em>Mindset:</em> </strong>Opportunistic, focused on harvesting gains and actively managing portfolio risk.</li><li>Portfolio Value — $11K with $4.5K booked profits</li></ul><h3><strong>Key Insight:</strong></h3><ul><li><strong>Compounding Power Wins:</strong> Tom’s passive approach multiplied his wealth by letting Apple’s long-term growth work its magic — proving that patience often beats chasing quick profits.</li><li><strong>Steady Cash Flow &amp; Lower Risk:</strong> Harry’s active style brought regular returns and helped manage market swings, but this safety net sometimes means missing out on those huge market rallies.</li><li><strong>Find Your Sweet Spot:</strong> Most investors aren’t strictly active or passive. Knowing the strengths and trade-offs of each style helps you craft a personalised strategy that fits your goals and comfort with risk.</li></ul><h3>Callouts for the Readers</h3><ul><li><strong><em>Letting Winners Run Matters:</em></strong> In markets or stocks with long-term upward momentum, staying the course — without trimming wins — can pay off spectacularly.</li><li><strong><em>Profit Booking Isn’t Wrong:</em></strong> In choppy or uncertain times, harvesting gains and reducing exposure can help preserve capital and sleep better at night.</li><li><strong><em>No One-size-fits-all: </em></strong>Your best strategy depends on temperament, confidence in your investment, and time horizon.</li><li><strong><em>Clarity is Key: </em></strong>Know your “why” for every trade or non-trade. Are you here to maximise compounding, or is cash-in-hand more important for your journey?</li></ul><h3>Bottom Line… 🎯</h3><p>Investing isn’t one-size-fits-all. Tom’s patient approach shows the power of <strong>letting winners run</strong> 🌱, while Harry’s active style locks in profits but risks missing big gains 💸.</p><p>Both have pros and cons — your best strategy depends on your goals and risk tolerance.</p><p><strong>What’s your style?</strong> Share your thoughts below! 💬👇</p><p>Thank you for staying till the end! 😃</p><blockquote>If you enjoyed reading, please give it a <strong>CLAP!</strong> 🙌</blockquote><p>Follow me and don’t miss out on any future posts — subscribe to my profile for interesting concepts on investing and finance 📈.</p><h4>A Message from InsiderFinance</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/301/0*10x5_2smmKq8oIlf.png" /></figure><p>Thanks for being a part of our community! Before you go:</p><ul><li>👏 Clap for the story and follow the author 👉</li><li>📰 View more content in the <a href="https://wire.insiderfinance.io/">InsiderFinance Wire</a></li><li>📚 Take our <a href="https://learn.insiderfinance.io/p/mastering-the-flow">FREE Masterclass</a></li><li><strong>📈 Discover </strong><a href="https://insiderfinance.io/?utm_source=wire&amp;utm_medium=message"><strong>Powerful Trading Tools</strong></a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c1f1ee3327c0" width="1" height="1" alt=""><hr><p><a href="https://wire.insiderfinance.io/hold-my-stocks-the-art-of-playing-the-long-game-without-losing-your-mind-c1f1ee3327c0">Hold My Stocks: The Art of Playing the Long Game (Without Losing Your Mind)</a> was originally published in <a href="https://wire.insiderfinance.io/">InsiderFinance Wire</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ Why Choosing the Right Scaling Method Can Make or Break Your Model]]></title>
            <link>https://medium.com/learning-data/why-choosing-the-right-scaling-method-can-make-or-break-your-model-887c403dbcd0?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/887c403dbcd0</guid>
            <category><![CDATA[feature-scaling]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[python-programming]]></category>
            <category><![CDATA[k-means]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Mon, 21 Jul 2025 20:01:11 GMT</pubDate>
            <atom:updated>2025-11-17T00:33:19.834Z</atom:updated>
            <content:encoded><![CDATA[<p>A step-by-step guide with examples, charts, and visuals — so you never overlook scaling again.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/382/1*2NRL4F4OVjioJJBfDDP5hA.png" /></figure><h3>✨ Introduction</h3><p>When working with machine learning models — especially clustering techniques like K-Means — many people focus on the algorithm but forget one critical step: <strong>scaling the data.</strong></p><p>This article shows you why scaling is crucial, using simple <strong>RFM (Recency, Frequency, Monetary)</strong> customer data and visual examples to demonstrate how Standard Scaling, Min-Max Scaling, and Log Scaling affect model results.</p><h4>🔷 Sample RFM Database</h4><p>We start with a small dataset of 5 customers (skewed dataset):</p><pre>| Customer | Recency (days) | Frequency | Monetary ($) |<br>| -------- | -------------- | --------- | ------------- |<br>| C1       | 5              | 50        | 5000          |<br>| C2       | 20             | 10        | 1000          |<br>| C3       | 60             | 5         | 500           |<br>| C4       | 90             | 2         | 200           |<br>| C5       | 200            | 1         | 100           |</pre><h4>🔷 Without Scaling:</h4><ul><li>K-Means minimizes Euclidean distances.</li><li>Here, <strong>Monetary dominates</strong> because its scale (100–5000) is much larger than Frequency (1–50) and Recency (5–200).</li><li>Result: clusters may reflect mostly Monetary, ignoring Recency/Frequency.</li></ul><pre># Create RFM data<br>raw_data = pd.DataFrame({<br>    &#39;Customer&#39;: [&#39;C1&#39;, &#39;C2&#39;, &#39;C3&#39;, &#39;C4&#39;, &#39;C5&#39;],<br>    &#39;Recency&#39;: [1, 5, 20, 100, 500],         # long tail<br>    &#39;Frequency&#39;: [200, 50, 10, 2, 1],       # steep drop<br>    &#39;Monetary&#39;: [100000, 5000, 1000, 200, 50] # extreme skew<br>}).set_index(&#39;Customer&#39;)</pre><pre>from sklearn.cluster import KMeans<br><br># Fit KMeans directly on raw data<br>kmeans = KMeans(n_clusters=2, random_state=42)<br>kmeans.fit(raw_data)</pre><h4>🔷 How Scaling Helps</h4><p>We tested three common scaling techniques:</p><ul><li>🔷 <strong>StandardScaler:</strong> Centers data (mean=0, std=1), balances feature contributions.</li></ul><pre>from sklearn.preprocessing import StandardScaler<br><br># Scale features to mean=0, std=1<br>scaler = StandardScaler()<br>scaled = scaler.fit_transform(raw_data)<br><br>kmeans.fit(scaled)</pre><ul><li>🔷 <strong>MinMaxScaler:</strong> Scales data to [0,1], keeps values positive, preserves distribution shape.</li></ul><pre>from sklearn.preprocessing import MinMaxScaler<br><br># Scale features to [0,1]<br>scaler = MinMaxScaler()<br>scaled = scaler.fit_transform(raw_data)<br><br>kmeans.fit(scaled)</pre><ul><li>🔷 <strong>Log Scaling:</strong> Reduces skew by compressing large values with log(x+1).</li></ul><pre>import numpy as np<br><br># Reduce skew with log(x+1)<br>log_scaled = np.log1p(raw_data)<br><br>kmeans.fit(log_scaled)</pre><p>These methods make each feature contribute meaningfully to the clustering, rather than letting one feature dominate.</p><h4>📊 Results — See the Difference!</h4><p>Once we ran K-Means with each scaling method, the results spoke for themselves.</p><p>Here’s how each method performed, measured by <strong>Silhouette Score</strong> (higher is better ):</p><pre>| Scaling Method    | Silhouette Score &amp; Notes |<br>|--------------------|---------------------------|<br>| 🚫 Unscaled        | ~0.00 — meaningless clusters |<br>| 🔷 StandardScaler  | ~0.40 — good separation |<br>| 🔷 MinMaxScaler    | ~0.42 — similar to StandardScaler |<br>| 🔷 Log Scaling     | ~0.49 — best result, reduced skew |</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a8PeA63PB_svpQVtFtoKbw.png" /></figure><blockquote>You can clearly see how proper scaling transforms random blobs into meaningful, actionable clusters — <strong>especially when the data is skewed.</strong></blockquote><h3>🎯 Key Takeaways</h3><p>✅ Without scaling, models favour the largest-scale feature → bad clusters.<br>✅ StandardScaler &amp; MinMaxScaler are great for balanced or normal-ish distributions.<br>✅ Log Scaling shines when data is skewed — as in our case.<br>✅ Always check distributions &amp; scale your data before training!</p><p>Without scaling, the largest feature dominates. With thoughtful scaling, every feature gets a fair say — and your clusters make sense!</p><h3>📌 Final Thoughts</h3><blockquote>Next time you train a model — especially a distance-based one like K-Means — <strong>don’t skip the scaling step.</strong><br> Proper scaling ensures your model focuses on what matters and delivers actionable insights.</blockquote><p>If you found this helpful, feel free to clap 👏 and share your experience with scaling in the comments!</p><p><em>The contents of external submissions are not necessarily reflective of the opinions or work of </em><a href="http://mavenanalytics.io/"><em>Maven Analytics</em></a><em> or any of its team members.</em></p><p><em>We believe in fostering lifelong learning and our intent is to provide a platform for the data community to share their work and seek feedback from the Maven Analytics data fam.</em></p><p><a href="https://medium.com/learning-data/how-to-get-your-work-published-by-learning-data-with-maven-analytics-7df21e466a3e?source=friends_link&amp;sk=020dfac485597d602e218968d9ffb395"><em>Submit your own writing here</em></a><em> if you’d like to become a contributor.</em></p><p><em>Happy learning!</em></p><p><em>-Team Maven</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=887c403dbcd0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/learning-data/why-choosing-the-right-scaling-method-can-make-or-break-your-model-887c403dbcd0">📊 Why Choosing the Right Scaling Method Can Make or Break Your Model</a> was originally published in <a href="https://medium.com/learning-data">Learning Data</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Nothing Feels New !!!]]></title>
            <link>https://medium.com/@channachannu.97/when-nothing-feels-new-256dec2b24d8?source=rss-76a97bcb3541------2</link>
            <guid isPermaLink="false">https://medium.com/p/256dec2b24d8</guid>
            <category><![CDATA[lifestyle]]></category>
            <category><![CDATA[stand-out]]></category>
            <dc:creator><![CDATA[Channabasava H]]></dc:creator>
            <pubDate>Mon, 20 Jan 2025 14:30:53 GMT</pubDate>
            <atom:updated>2025-01-20T14:33:18.902Z</atom:updated>
            <content:encoded><![CDATA[<h3>When Nothing Feels New !!!</h3><p><em>I</em><strong><em>t’s a strange feeling </em></strong>— being bored of being bored. You find yourself stuck in a space where nothing excites you, where the things others find meaningful or thrilling seem to fall flat. Take tourist attractions, for instance. People rave about them and call them must-see destinations but for me? They don’t feel worth the trouble. (I’m sure some of you can relate to that, right?)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/369/1*gOp45J7Y8LuLE7IyO1pzxA.jpeg" /></figure><p>So, here I am, in this peculiar state of monotony, trying to break free.</p><p>This article is my first step out of that boredom — my attempt to shake off the dullness and wake up to something new. I wrote this without preparation, without overthinking, just a simple thought: <em>What if I start here? What if this is the beginning of a fresh chapter?</em></p><p>For now, this is my way of saying goodbye to monotony and hello to the possibility of a new lifestyle.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=256dec2b24d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>