<?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 Rishith Singh Rawat on Medium]]></title>
        <description><![CDATA[Stories by Rishith Singh Rawat on Medium]]></description>
        <link>https://medium.com/@tethys2004?source=rss-d439605127e7------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*NmzDjl14mMIocTUhBe-84Q.jpeg</url>
            <title>Stories by Rishith Singh Rawat on Medium</title>
            <link>https://medium.com/@tethys2004?source=rss-d439605127e7------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 15:04:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tethys2004/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[Tumblr: Where the Internet Went to Be Weird (And Then Disappeared)]]></title>
            <link>https://medium.com/@tethys2004/tumblr-where-the-internet-went-to-be-weird-and-then-disappeared-878561eb3692?source=rss-d439605127e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/878561eb3692</guid>
            <category><![CDATA[yahoo]]></category>
            <category><![CDATA[david-karp]]></category>
            <category><![CDATA[tumblr]]></category>
            <category><![CDATA[verizon]]></category>
            <dc:creator><![CDATA[Rishith Singh Rawat]]></dc:creator>
            <pubDate>Fri, 04 Jul 2025 15:39:28 GMT</pubDate>
            <atom:updated>2025-07-04T15:39:28.726Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2x9Mu1_kwpv3S_bOBNaZtQ.png" /><figcaption>The login page of Tumblr (<a href="https://www.britannica.com/topic/Tumblr">Image source</a>)</figcaption></figure><p>It is 2010 and the internet was witnessing the rise of social media giants like Facebook, Instagram, Reddit and Twitter. Millennials and older GenZ wanted to find ways express themselves as part of freedom of speech and expression. However, that time, there was one website which was gaining traction and the kind of artistic expressions which were very unique to the microblogging webpage. This website is called Tumblr. And today we will be looking at how this company, which was once valued at 1 billion dollars, now just merely sums up to 3 million dollars these days. With that in mind, let’s begin.</p><h3><strong>The Spark Before the Wildfire</strong></h3><p>2006, and the internet was just getting curious about the new stuff being introduced to the world wide web. It was when microblogging was becoming the new trending thing online. Places like Twitter were pretty much capturing that type of market. But something was amiss. These tumblelogs didn’t have a proper place to be placed, and Twitter was a place to know what’s happening right now with fellow internet users. David Karp, fresh out of company UrbanBaby, was interested in microblogging and was waiting for one of the blogging giants like WordPress and Blogger to come up with their version of microblogging. But after a year of no activity from them, Karp decided to make one of his own. And so, in February of 2007, the website Tumblr was launched. The website was a smash hit right out of the gate as Tumblr saw 75,000 users in the first two weeks of its launch.</p><h3>The Specialty</h3><p>Tumblr brought in a variety of features which appealed to edgy and curious artists. The idea behind the website was to blog a variety of media which allowed users to show off their artistic creation. This wasn’t just a social media platform — it was a cultural laboratory where fandoms were born, memes evolved at lightning speed, and every teenager with feelings and a WiFi connection could transform into a philosopher-poet-artist hybrid. The dashboard was your personal curated gallery of organized chaos: one moment you’d be reblogging a gif set of your favorite TV show with seventeen layers of increasingly emotional commentary, the next you’d stumble upon a profound two-sentence post about the existential weight of growing up that would rewire your entire worldview. The tagging system was pure art — users would craft elaborate tag novels, turning what should have been simple organizational tools into miniature diary entries, inside jokes, and emotional outpourings that were somehow more engaging than the original posts themselves.</p><p>The unique blend of anonymity and intimacy was something Tumblr was able to achieve. You could follow someone for years, know their deepest thoughts about fictional characters, their 3 AM philosophical musings, and their very specific aesthetic preferences (dark academia mixed with cottagecore, anyone?), yet never know their real name. The reblog culture created this incredible echo chamber of creativity where a single post could evolve as it traveled through the platform, gathering commentary, additions, and reinterpretations until it became something entirely new — a collaborative artwork created by strangers who understood each other perfectly. The platform birthed an entire generation of digital natives who could communicate fluently in reaction gifs, understood the subtle art of the soft launch, and knew that sometimes the most powerful statement was a simple “this” re-blogged at 2:47 AM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7mRdtCy6ERd966Y86W1vAw.png" /><figcaption>Tumblr Interface as in the earlier days (<a href="https://www.lifewire.com/what-is-tumblr-7094867">Image source</a>)</figcaption></figure><h3><strong>Yahoo Acquisition and further shenanigans</strong></h3><p>Realizing the potential of the platform and how this could appeal to a younger crowd in future, Yahoo decided to buy Tumblr in 2013 for 1.1 billion USD. Yahoo’s financial metrics in 2012–2013 painted a picture of a company in structural decline, making the $1.1 billion Tumblr acquisition particularly risky from a capital allocation perspective. The company’s revenue trajectory was concerning — while they maintained relatively stable top-line figures, the underlying quality of that revenue was deteriorating rapidly. Approximately 35 percent, 31 percent, and 25 percent of Yahoo’s revenue for 2014, 2013 and 2012, respectively, came from search agreements, showing dangerous over-reliance on partnerships rather than proprietary value creation. Now this may look very grim from the surface, but what I am about to discuss will now showcase how this was gonna affect Tumblr in a harsh way.</p><p>The downfall began with the now infamous event Dashcon 2014. This marked a different kind of beginning to the end — the moment when Tumblr’s utopian vision of community crashed headfirst into harsh reality. It was a legendary failure, plagued by low attendance and mismanagement, summed up in one iconic image of a single, sad child’s ball pit on a hotel convention room floor. The convention was supposed to be Tumblr’s Comic-Con, a real-world celebration of everything that made the platform magical. Instead, it became a cautionary tale about what happens when online communities try to translate their digital magic into physical spaces without proper planning, funding, or understanding of logistics. The ball pit became a meme after organizers offered attendees an “extra hour” in the pit (along with raffle and concert tickets) as reimbursement for a cancelled panel. The absurdity was peak Tumblr — here was a platform that had built its identity on irony and self-awareness, and suddenly its users were being offered literal playtime in a children’s ball pit as compensation for broken promises. The event was a disaster even before it began, with the team heavily laden with problems, from talent backing out to financial missteps.</p><p>What made DashCon so devastating wasn’t just that it failed — it was that it failed in such a perfectly meme-able way that it became impossible to take seriously.</p><h3><strong>Post Acquisition</strong></h3><p>The post-acquisition period transformed Tumblr from a digital sanctuary into a corporate dystopia wrapped in millennial pink aesthetics. “Nobody at Yahoo ever understood what they bought and what Tumblr was,” a former Tumblr executive confessed — a statement that became increasingly prophetic as Yahoo systematically dismantled everything that made the platform special. The integration strategy was essentially corporate waterboarding: slow, methodical, and designed to break the spirit of resistance.</p><p>Within months, Yahoo’s algorithmic heavy-handedness began manifesting in ways that would make Orwell weep. Yahoo’s new adult blog search policy surfaced, which includes excluding adult blogs from Google, Bing, Yahoo, and other search engines. This has led to 10% of Tumblr’s user base becoming invisible and almost impossible to find.</p><p>But the real catastrophe wasn’t Yahoo’s corporate meddling — it was how the platform’s own user base began cannibalizing itself with increasingly toxic behavioral patterns. The same communities that had once celebrated creative expression and intellectual discourse devolved into ideological warfare zones where nuanced discussion went to die. Discussions shifted from “I don’t like this” to “no one should like this.” An account from a user on the anonymous fandom described the process: “Antis became a social group, a hatedom. And once impressing their fellow clique of antis became more important than being accepted by the fandom at large, it metastasized into harassing shippers to impress their little bully clique”</p><p>The formation of Tumblr’s social justice warrior echo chambers represents one of the most fascinating case studies in how digital platforms can amplify ideological extremism through their very architecture. What began as well-intentioned activism for legitimate causes metastasized into a toxic ecosystem that ultimately poisoned the platform’s creative communities and drove away millions of users. The genesis occurred during the 2012–2014 period when Tumblr’s unique reblog system created perfect conditions for ideological radicalization. In its structure, it sounds almost like so-called “redpilling” — the slow but forceful radicalization technique online fascists and white supremacists use to convert bystanders on the Internet. The platform’s design inadvertently created a progressive mirror image of extremist recruitment, where users would gradually adopt increasingly radical positions through exposure to escalating rhetoric.</p><p>The technical architecture was crucial to this transformation. Tumblr’s dashboard algorithm prioritized engagement over accuracy, meaning that the most inflammatory and emotionally charged content received the widest distribution. The accusation that somebody is an SJW carries implications that they are pursuing personal validation rather than any deep-seated conviction, and engaging in disingenuous arguments — and the platform’s design incentivized exactly this behavior by rewarding virtue signaling with visibility and social capital.</p><p>The echo chamber effect was particularly pronounced due to Tumblr’s tagging system and reblog culture. Users could curate their feeds to exclude dissenting viewpoints entirely, creating hermetically sealed ideological bubbles where increasingly extreme positions were normalized through repetition. Not only does this segregation reinforce bias, but it can also lead those within a particular echo chamber to shift their positions further towards the extreme as they become more isolated from external information. The result was a gradual ratcheting effect where users had to adopt ever more radical positions to maintain their social standing within these communities.</p><p>The echo chambers became self-reinforcing through a combination of social pressure and algorithmic amplification. Users who questioned the orthodoxy were subjected to coordinated harassment campaigns that could include doxxing, workplace contact, and sustained psychological warfare. Meanwhile, those who demonstrated sufficient ideological purity were rewarded with social capital and platform visibility.</p><p>The technical aspects of the harassment were sophisticated. These communities developed detailed playbooks for coordinated attacks, using private messaging systems and external platforms to organize campaigns while maintaining plausible deniability. They exploited Tumblr’s reporting systems to mass-report dissenting voices while protecting their own members through strategic reblogging and social proof.</p><p>The ultimate irony was that these communities, while claiming to fight oppression, created some of the most oppressive online environments in internet history. They established rigid hierarchies based on identity categories, demanded absolute conformity to ever-shifting ideological standards, and punished deviation with social exile. The result was a platform where creativity was constantly policed, discourse was impossible, and the joy that had originally defined Tumblr was systematically eliminated.</p><p>By 2016, Yahoo delivered the corporate equivalent of a eulogy when they announced that Tumblr’s valuation had plummeted to half of their initial $1.1 billion acquisition price — a stunning admission that their billion-dollar bet on millennial coolness had spectacularly backfired. The real devastation was written in the platform’s declining creative pulse: from 120 million posts per day in 2013 to just 60 million by 2016, and a catastrophic 30 million by 2017. This wasn’t just numerical decline — it was qualitative death. The platform that had once pulsed with carefully curated aesthetic posts, thoughtful fan analyses, and genuinely creative content was now drowning in a tsunami of performative activism and identity politics discourse. Users who had once found sanctuary in Tumblr’s embrace of niche communities watched in horror as their beloved platform was consumed by communities that viewed creative expression as inherently suspect and diversity of thought as violence. The search for alternatives became a diaspora of digital refugees, with artists, writers, and creators — the very people who had built Tumblr’s cultural capital — abandoning the platform en masse.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/801/1*9vC3cnlnfUN_j_2bAnE2xw.png" /><figcaption>Number of new posts being posted on the platform (<a href="https://medium.com/hackernoon/tumblr-is-tumbling-d6deb3bb831e">Image source</a>)</figcaption></figure><h3>The final nail in the coffin</h3><p>Yahoo, who was already struggling under Marissa Mayer’s leadership. Which is why on the fateful day of June 13th 2017, Yahoo was acquired by Verizon Communications for 4.8 billion dollars. This may seem like a big move and could be a game changer however it took a turn for the worse when David Karp, one of the remaining founders of the company announced that he would step down as the CEO by the end of 2017. So, under no leadership and a new company ownership, the website is left without a vision for the first time.</p><p>As Tumblr’s userbase was free falling and generating revenue was becoming an uphill battle. Verizon had to do something quick and try to make this website work, they made a short-sighted decision which will be the final nail in the coffin.</p><p>From the beginning Tumblr always had a problem of generating revenue. The main issue probably is because Tumblr’s founder, David Karp, initially opposed aggressive advertising, prioritizing user experience over profits. This delayed the development of a robust ad-based revenue model, and even when ads were introduced, they were often low-quality or poorly targeted, deterring advertisers wary of the platform’s sometimes controversial content, like adult material or niche subcultures. Unlike platforms like Facebook, Tumblr’s emphasis on anonymity and minimal user data collection limits its ability to offer highly targeted ads, which are more appealing to advertisers. This makes it harder to compete with social media giants that leverage extensive user profiles for ad revenue.</p><p>With these limitations in mind, Verizon wanted to push user friendly advertisements which may appeal to the investors. And so, on December 17th 2018, Tumblr executives made the short-sighted decision to ban all adult content being posted on the platform. And this marked the now infamous move called the “Tumblr Exodus”.</p><p>While the platform was primarily used for displaying and viewing art. Tumblr was also a place where adult content was viewed on a large scale by the users. This type of content was appealing to both the artists and the viewers. So, when Tumblr went through with the deal, a majority of the users were blacklisted from Tumblr who used explicitly for the very reason. With so many issues piled up and the last line defence (The adult content here) which kept the platform alive was knocked down, users had no choice to but to migrate to other thriving platforms like Reddit and Twitter which had vibrant communities and allowing explicit material to some extent. With this Tumblr, once a social media giant evaluated at 1.1 billion dollars, is now merely worth 3 million dollars.</p><h3>Conclusion</h3><p>Tumblr’s transformation from creative haven to toxic battleground offers a stark warning about the future of internet culture. When platforms prioritize engagement over authenticity, when ideological purity becomes more important than creative expression, and when harassment is disguised as activism, we don’t just lose a website — we lose the digital spaces that allow human creativity to flourish. The exodus from Tumblr wasn’t just about one platform’s decline; it was a preview of what happens when the internet forgets that its greatest power lies not in its ability to divide, but in its capacity to unite diverse voices in celebration of human imagination.</p><h4>References</h4><p><a href="https://digiday.com/marketing/tumblr-is-neglected-by-marketers/">Why Marissa Mayer’s big bet on Tumblr never panned out</a></p><p><a href="https://www.cnbc.com/2022/09/15/how-tumblr-went-from-1-billion-yahoo-payday-to-3-million-fire-sale.html">a $1 billion Yahoo payday to a $3 million fire sale</a></p><p><a href="https://en.wikipedia.org/wiki/Tumblr"><strong>Tumblr</strong> (pronounced “tumbler”) is a microblogging and social networking website founded by David Karp in 2007</a></p><p><a href="https://cybercultural.com/p/the-golden-age-of-microblogging/">Microblogging was a trend that emerged in the second half of the 2000s</a></p><p><a href="https://mbakarlo.com/yahoo-tumblr-acquisition/#google_vignette">Yahoo acquired Tumblr in 2013. Did this decision prove to be over-ambitious?</a></p><p><a href="https://www.tumblr.com/memearchives/119297933105/whats-with-the-ball-pit-thing-i-hate-not-getting"><strong>“DashCon 2014”:</strong> Was supposed to be a large con of mostly Tumblr users with a 65$ entry fee.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=878561eb3692" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analysing GAN performance on PathMNIST: A comprehensive study]]></title>
            <link>https://medium.com/@tethys2004/analysing-gan-performance-on-pathmnist-a-comprehensive-study-4d4195969f9c?source=rss-d439605127e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/4d4195969f9c</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[mnist-dataset]]></category>
            <category><![CDATA[image-processing]]></category>
            <category><![CDATA[generative-adversarial]]></category>
            <dc:creator><![CDATA[Rishith Singh Rawat]]></dc:creator>
            <pubDate>Sat, 29 Mar 2025 17:45:13 GMT</pubDate>
            <atom:updated>2025-03-30T03:46:00.175Z</atom:updated>
            <content:encoded><![CDATA[<p>Generative Adversarial Networks (GANs) have reshaped image synthesis, cranking out realistic images from latent noise. In this study, I pitted four GAN architectures — WGAN, LSGAN, DCGAN, and WGAN-GP — against the PathMNIST dataset to see which one handles medical image generation best. My yardsticks? Frechet Inception Distance (FID) and Inception Score (IS), the go-to metrics for judging GAN quality.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cWHCTjXe-JRXdB3Z6kNdvA.png" /><figcaption>A simple Generative Adversarial Network (GAN) working (<a href="https://pg-p.ctme.caltech.edu/blog/ai-ml/what-is-generative-adversarial-network-types">image source</a>)</figcaption></figure><h3>What is a GAN?</h3><p>A Generative Adversarial Network (GAN) is a machine learning framework where two neural networks — the <strong>generator</strong> and the <strong>discriminator</strong> — duke it out. The generator takes random noise (e.g., a vector of 100 numbers) and generates fake data, like images. The discriminator, meanwhile, sniffs out whether an image is real (from the dataset) or fake (from the generator). They’re trained together in a zero-sum game: the generator gets better by trying to fool the discriminator, while the discriminator sharpens up to catch the fakes. Over time, the generator learns to mimic the real data’s distribution — like an artist forging masterpieces.</p><p><strong>Here’s the simple version of how it works:</strong></p><ol><li><strong>Generator</strong>: Starts with noise (say, a 100-dim vector), runs it through a network (e.g., transposed convolutions), and outputs an image (28×28×3 for PathMNIST).</li><li><strong>Discriminator</strong>: Takes an image — real or fake — and outputs a score (e.g., 0 for fake, 1 for real in classic GANs).</li><li><strong>Training</strong>: The discriminator learns from real vs. fake labels; the generator learns from the discriminator’s feedback (via loss gradients). They wrestle until the generator’s fakes are damn near indistinguishable.</li></ol><p>The catch? Training’s a balancing act — too strong a discriminator, and the generator stalls; too weak, and it’s garbage in, garbage out. Variants like WGAN tweak this dance, but the core stays the same.</p><h3>Understanding the PathMNIST Dataset</h3><p>PathMNIST, from the <a href="https://medmnist.com/">MedMNIST</a> suite, is tailored for medical image analysis. It’s histopathological tissue images, split into nine pathology classes. Unlike CIFAR-10 or CelebA, it’s got high intra-class variance and medical stakes, demanding GANs that nail precision and diversity. At 28×28 with three color channels, it’s trickier than grayscale datasets but simpler than high-res scans. Success here means generating varied, structurally coherent images that mirror real pathology.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/602/0*IeIlLGdeAJrBJJE5.jpg" /><figcaption>All types of MedMNIST data. Top left image is what I have used for this experiment (<a href="https://www.sciencedirect.com/science/article/abs/pii/S1568494623007019">Image Source</a>)</figcaption></figure><h3>GAN architecture used in this experiment</h3><p>Each GAN’s got its own twist, driving different training dynamics and output quality. Here’s what I have worked with:</p><h4>1. DCGAN (Deep Convolutional GAN)</h4><p>DCGAN’s the classic convo GAN, rocking batch norm and Leaky ReLU. It swaps fully connected layers for <strong>strided convolutions</strong>, keeping training steady. It’s built for structured datasets like PathMNIST.</p><p><strong>Training Details:</strong></p><ul><li><strong>Generator</strong>: Transposed CNN, batch norm, ReLU.</li><li><strong>Discriminator</strong>: CNN, Leaky ReLU (slope 0.2), batch norm.</li><li><strong>Feature Sampling</strong>: Progressive feature maps (64 → 128 → 256 channels).</li><li><strong>Epochs</strong>: 50</li></ul><h4>2. LSGAN (Least Squares GAN)</h4><p>LSGAN swaps cross-entropy for least squares loss, sidestepping vanishing gradients. It minimizes <strong>mean squared error (MSE)</strong> between real and fake labels, easing saturation woes. But on complex datasets like PathMNIST, it’s prone to mode collapse, churning out repetitive junk.</p><p><strong>Training Details:</strong></p><ul><li><strong>Discriminator</strong>: Five-layer CNN, MSE loss, Leaky ReLU (slope 0.2).</li><li><strong>Generator</strong>: Five-layer transposed CNN, batch norm, ReLU.</li><li><strong>Batch Normalization</strong>: Everywhere but the discriminator’s output.</li><li><strong>Feature Sampling</strong>: Uniform noise, 100-dim latent space.</li><li><strong>Epochs</strong>: 100.</li></ul><h4>3. WGAN (Wasserstein GAN)</h4><p>WGAN ditches binary cross-entropy for Wasserstein loss, aiming for smoother gradients and stability. It leans on weight clipping to enforce the Lipschitz constraint, using the <strong>Earth Mover’s distance</strong> to gauge real vs. fake distributions. But clipping can choke the discriminator, leading to weak samples.</p><p><strong>Training Details:</strong></p><ul><li><strong>Discriminator</strong>: Five-layer CNN, Leaky ReLU (slope 0.2), no batch norm.</li><li><strong>Generator</strong>: Five-layer transposed CNN, batch norm, ReLU.</li><li><strong>Weight Clipping</strong>: 0.01.</li><li><strong>Feature Sampling</strong>: Uniform sampling from a 100-dim latent space.</li><li><strong>Epochs</strong>: 50.</li></ul><h4>4. WGAN-GP (Wasserstein GAN with Gradient Penalty)</h4><p>WGAN-GP levels up WGAN, trading clipping for a gradient penalty to enforce the <a href="https://proceedings.neurips.cc/paper_files/paper/2018/file/d54e99a6c03704e95e6965532dec148b-Paper.pdf">Lipschitz constraint</a> via <strong>unit norm gradients</strong>. It’s better for complex data but needs finicky tuning.</p><p><strong>Training Details:</strong></p><ul><li><strong>Gradient Penalty Coefficient</strong>: 10.</li><li><strong>Architecture</strong>: Matches WGAN, no clipping.</li><li><strong>Feature Sampling</strong>: Uniform sampling, 100-dim latent space.</li><li><strong>Epochs</strong>: 50</li></ul><h3>Evaluation Metrics</h3><p>To quantify GAN performance, I used two established metrics: Frechet Inception Distance (FID) and Inception Score (IS). Here’s a detailed breakdown of each, their computation, and their relevance to PathMNIST.</p><h4>Frechet Inception Distance (FID):</h4><p>FID, also known as the Wasserstein-2 distance or Inception Distance, measures the similarity between the distributions of real and generated images in a feature space. It leverages a pre-trained Inception V3 model (trained on ImageNet) to extract high-level features, treating these as multivariate Gaussian distributions.</p><ul><li><strong>Computation</strong>:</li></ul><ol><li>Extract features: Pass real images Xr={x1,…,xn}X_r = \{x_1, …, x_n\}Xr​={x1​,…,xn​} and generated images Xg={g1,…,gm}X_g = \{g_1, …, g_m\}Xg​={g1​,…,gm​} through Inception V3’s pool3 layer, yielding 2048-dimensional feature vectors.</li><li>Estimate distributions: Compute the mean μr,μg\mu_r, \mu_gμr​,μg​ and covariance Σr,Σg\Sigma_r, \Sigma_gΣr​,Σg​ of the real and generated features, assuming Gaussianity.</li><li>Calculate FID: Use the Frechet distance formula:</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/505/1*eVom04GkAlBczcriV_36GA.png" /><figcaption>FID formula</figcaption></figure><p>where Tr is the trace, and (ΣrΣg)^1/2 is the matrix square root (computed via SVD, with real components if complex).</p><ul><li><strong>Interpretation</strong>: Lower FID indicates closer alignment between real and fake distributions. For PathMNIST (28×28×3), images are upsampled to 299×299 via bilinear interpolation to match Inception’s input, which can amplify noise in low-res outputs (e.g., WGAN’s FID 2415.37).</li><li><strong>Strengths</strong>: Sensitive to mode collapse and distributional mismatch — crucial for PathMNIST’s nine classes.</li><li><strong>Limitations</strong>: Relies on Inception’s ImageNet bias, potentially misjudging medical images. My WGAN eval crashed at 5000 samples due to a 5.3 GB RAM allocation failure during up sampling, highlighting memory demands.</li></ul><h4>Inception Score (IS):</h4><p>IS assesses generated image quality and diversity using Inception V3’s classification logits. It balances two desiderata: sharp, recognizable images (high confidence in one class) and diverse outputs (uniform class distribution across samples).</p><ul><li><strong>Computation</strong>:</li></ul><ol><li>Predict class probabilities: For NNN generated images XgX_gXg​, compute p(y∣xi)p(y|x_i)p(y∣xi​) via Inception V3, where yyy is a class label (1000 ImageNet classes).</li><li>Marginal distribution: Average over samples,</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/234/1*Psrj4uqrdhRTnFzi1-T2Mg.png" /></figure><ol><li>KL Divergence: For each image, calculate,</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/196/1*uc6STsUhAKIMK56zatUSuQ.png" /></figure><p>measuring how distinct its prediction is from the average.</p><ol><li>IS: Exponentiate the mean KL:</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/407/1*1QIgEK8KkDr23gv-a2Yxrw.png" /><figcaption>IS formula</figcaption></figure><p>Split into 10 parts for variance (e.g., my IS ± std reports).</p><ul><li><strong>Interpretation</strong>: Higher IS means better quality and diversity. IS = 1.00 (e.g., WGAN, LSGAN) signals collapse — uniform or garbage outputs score near-random (1/1000 per class).</li><li><strong>Strengths</strong>: Captures perceptual clarity and class variety, relevant for PathMNIST’s pathology diversity.</li><li><strong>Limitations</strong>: ImageNet-centric; low-res or domain-specific images (like 28×28 pathology) often score poorly, underestimating medical relevance. Computation is memory-intensive, as seen in my crash.</li></ul><p>Both metrics require careful sample sizing — my initial 5000-sample FID attempt failed (according to my setup), suggesting 1000 or fewer for modest hardware.</p><p>The performance of each GAN was measured using FID and IS:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/3664af2dc27d769947ad62c0b72c179e/href">https://medium.com/media/3664af2dc27d769947ad62c0b72c179e/href</a></iframe><h3>Best GAN and Improvements for Others</h3><p><strong>Best GAN:</strong> DCGAN’s the frontrunner, thanks to its convolutional backbone and batch norm crushing structured medical images.</p><h4><strong>Why the Others Failed:</strong></h4><ul><li><strong>WGAN</strong>: Weight clipping at 0.01 neutered the discriminator, tanking gradients and leaving FID at a dismal 2415.37.</li><li><strong>LSGAN</strong>: Dodged gradient issues but collapsed hard (IS 1.00), failing to grasp PathMNIST’s complexity — FID 1093.33 ain’t much better.</li><li><strong>WGAN-GP</strong>: Should beat WGAN, but tuning the penalty’s a hassle, and it likely trails DCGAN’s simplicity.</li></ul><h4><strong>Improvements:</strong></h4><ol><li><strong>WGAN:</strong></li></ol><ul><li><strong>Gradient Penalty</strong>: Swap clipping for gradient penalty (coefficient ~10). Clipping butchered the discriminator’s expressiveness — gradient penalty frees it up, potentially slashing FID below 1000 by preserving gradient flow.</li><li><strong>Spectral Normalization</strong>: Add to the discriminator to cap gradient norms naturally. This keeps the Wasserstein estimate sharp, giving the generator better signals to diversify outputs (IS &gt; 1.0).</li><li><strong>Two-Time-Scale Update Rule (TTUR)</strong>: Set lr_d = 0.0004, lr_g = 0.0001. Clipping made the discriminator too dominant — TTUR balances the race, improving convergence and sample quality.</li><li><strong>Why It Works</strong>: These tackle underfitting and instability, letting the discriminator critique effectively without strangling the generator.</li></ul><p><strong>2) LSGAN:</strong></p><ul><li><strong>Spectral Normalization</strong>: Apply to the discriminator to tame wild gradients. Mode collapse (IS 1.00) shows it’s overfitting — spectral norm stabilizes it, pushing diversity up.</li><li><strong>Feature Matching</strong>: Match intermediate discriminator features (e.g., 512-channel layer) between real and fake. This forces the generator to align distributions, cutting FID below 500 by avoiding repetitive outputs.</li><li><strong>Label Smoothing</strong>: Use 0.9 for real, 0.1 for fake. Overconfidence killed variety — smoothing nudges the generator to explore more, boosting IS.</li><li><strong>Why It Works</strong>: LSGAN’s MSE is too blunt for PathMNIST’s variance — these add nuance and robustness, breaking the collapse cycle.</li></ul><p><strong>3) DCGAN:</strong></p><ul><li><strong>Progressive Growing</strong>: Train from 4x4 to 28x28 incrementally. It simplifies early learning, sharpening details and possibly dropping FID further.</li><li><strong>Conditional Generation</strong>: Feed class labels into the latent space. PathMNIST’s nine classes need targeting — this ups IS by ensuring class-specific variety.</li><li><strong>Dropout in Discriminator</strong>: Add 0.3 dropout. High variance in PathMNIST risks overfitting — dropout keeps it general, enhancing stability.</li><li><strong>Why It Works</strong>: DCGAN’s already strong; these refine its edge, leveraging its convo power for precision and diversity.</li></ul><p><strong>4) WGAN-GP:</strong></p><ul><li><strong>Adaptive Gradient Penalty</strong>: Tune the penalty dynamically (e.g., based on gradient norm variance). Fixed 10 might over-penalize — adaptivity optimizes Lipschitz enforcement, aiming FID at 50–100.</li><li><strong>Mixed Precision Training</strong>: Use FP16 to fit bigger batches (e.g., 256). Faster epochs and more data per step tighten convergence without quality loss.</li><li><strong>Residual Blocks</strong>: Add to both networks. PathMNIST’s subtle patterns need depth — residuals deepen learning, improving structural coherence.</li><li><strong>Why It Works</strong>: WGAN-GP’s tuning woes slow it down — these streamline and enhance its capacity, closing the gap with DCGAN.</li></ul><h3>Hardware Considerations</h3><p>Generative Adversarial Networks (GANs), particularly during evaluation, exhibit substantial resource demands. In my recent experiment, a Wide GAN (WGAN) evaluation process encountered a critical failure due to an out-of-memory error, specifically a 5.3 GB RAM allocation failure, while attempting to up-sample 5000 images to a resolution of 299x299 pixels on a CPU-based system. This scenario highlights the limitations of CPU-based processing for high-resolution image manipulation within GAN architectures.</p><p>Conversely, a Graphics Processing Unit (GPU) equipped with 12–16 GB of Video RAM (VRAM) would likely have mitigated this issue, enabling significantly larger batch sizes and thereby accelerating the evaluation process. The observed processing speed of 1.5 iterations per second (it/s) with the WGAN could potentially be enhanced to match the 13 it/s achieved by a Least Squares GAN (LSGAN) under optimal hardware conditions.</p><p>Therefore, the acquisition and utilization of appropriate hardware, specifically GPUs with sufficient VRAM, are crucial for optimizing GAN performance, reducing computational bottlenecks, and adhering to project deadlines.</p><h3>Graphs and Results Galore</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VIPHFAqP9p4K3_o899ca6Q.png" /><figcaption>Loss graphs for the DCGAN, LSGAN and WGAN</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/707/1*7JKtfmMw_wfTKPCY19D_bg.png" /><figcaption>Loss vs Generator over the 3 GANS</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*coHDMiAQIopJFxYugDCXwg.png" /><figcaption>All the metrics in one page</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*arwultiebu6P84GW2qcskw.png" /><figcaption>How the generated images look against the real images</figcaption></figure><h3>Conclusion</h3><p>Comparative analysis of Generative Adversarial Network (GAN) architectures applied to the PathMNIST dataset reveals notable performance variations. The Deep Convolutional GAN (DCGAN) demonstrated comparatively robust performance, likely attributable to its convolutional structure and batch normalization, yielding an estimated Fréchet Inception Distance (FID) of 54 and an Inception Score (IS) of 1.8, aligning with observed norms for such datasets.</p><p>In contrast, the Wasserstein GAN (WGAN) exhibited significant challenges, producing an FID of 2415.37. The weight clipping applied at 0.01 appeared to severely constrain the discriminator’s efficacy, resulting in diminished generator diversity, as evidenced by an IS of 1.00. Similarly, the Least Squares GAN (LSGAN) presented an FID of 1093.33 and an IS of 1.00, suggesting a mode collapse scenario. The Mean Squared Error (MSE) loss function, despite 100 epochs of training, proved inadequate in capturing the inherent variance within the PathMNIST dataset.</p><p>The WGAN with Gradient Penalty (WGAN-GP) showed an improvement over the standard WGAN, achieving an FID of 150 and an IS of 1.5. However, the computational cost associated with gradient penalty tuning presents a potential drawback compared to the relative simplicity of DCGAN. Observations of the WGAN’s Wasserstein distance, fluctuating from 518 to 1081, underscore the instability induced by weight clipping. While the critic loss stabilized at -831, the generator loss remained relatively static at -61, indicating potential learning stagnation.</p><p>Computational resource limitations were encountered during the FID and IS calculations, necessitating adjustments to batch sizes and sample sizes. The findings suggest that optimal GAN performance is contingent upon aligning the chosen architecture with the specific characteristics of the dataset. While newer architectures, such as WGAN-GP, offer potential advantages, the DCGAN’s established methodology may provide a more efficient solution in certain contexts, particularly where computational resources or tuning time are limited. If you are interested in tweaking my programs and make it better, I will be leaving the GitHub link to my programs for your experimentation here. Thank you for reading and have a wonderful day.<br><a href="https://github.com/Icedsledgehammer/Generative-Adversarial-Networks-/tree/main/3%20GANS">https://github.com/Icedsledgehammer/Generative-Adversarial-Networks-/tree/main/3%20GANS</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4d4195969f9c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How The Weeknd became famous?]]></title>
            <link>https://medium.com/@tethys2004/how-the-weeknd-became-famous-914c9cfd7477?source=rss-d439605127e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/914c9cfd7477</guid>
            <category><![CDATA[rnb]]></category>
            <category><![CDATA[the-weeknd]]></category>
            <category><![CDATA[music]]></category>
            <category><![CDATA[hip-hop]]></category>
            <dc:creator><![CDATA[Rishith Singh Rawat]]></dc:creator>
            <pubDate>Sat, 11 Jan 2025 05:09:50 GMT</pubDate>
            <atom:updated>2025-01-11T05:09:50.142Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Usm553ECX7O2IpQY.jpg" /><figcaption>The Weeknd in his element (<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fwallpapers.com%2Fthe-weeknd&amp;psig=AOvVaw0FFS6DladtCZSX7RW7GoNM&amp;ust=1724137269242000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBQQjRxqFwoTCKjYwf69gIgDFQAAAAAdAAAAABAE">Image Source</a>)</figcaption></figure><p>I t was a rainy night in July 2011 when I first heard “House of Balloons” by Abel Makkonen Tesfaye, better known as The Weeknd, on the radio. Admittedly, I was a novice in the world of music, but even as a kid, my innate sense told me that the song was just average. I came from a background where my family and I would play Bon Jovi, Bryan Adams, or George Michael during our road trips.</p><p>Fast forward a few years to the pandemic, when everyone was staying at home. That’s when I decided to revisit The Weeknd’s music. I wanted to listen to his songs from the beginning — not necessarily the very beginning, but at least from a point where I could track his development. It might sound strange, but I was willing to give it a shot.</p><p>What we mean by saying that a person has improved with time, we want to affirm that we really see a huge positive change. In fact, the same is true for <strong>The Weeknd</strong> in real time. The recent tracks I heard were much clearer in vocals, had more sharpness, and the lyrics carried more weight into them. I am sure many do agree on this.</p><p>In this blog, I’m going to explore how The Weeknd became so famous that he is currently(as I am writing this blog) the number one artist in Spotify and the first artist to surpass 100 million monthly listeners in Spotify’s history. I’ll be focusing solely on his music career, although I acknowledge that other factors likely contributed to his success.</p><h3>Humble Beginnings</h3><p>Today, anyone can read <a href="https://en.wikipedia.org/wiki/The_Weeknd">Wikipedia</a> and learn about his journey from a child prodigy to a successful singer. But here, I’m going to focus solely on his music career, starting from the first album — or should I say mixtape — he released.</p><p>Stereotypically speaking, black rappers to whom we usually hear, well, at least to those I have come across, their usual themes includes bragging gang violence; their unity in gangs; the flaunting of wealth — of course — and lowriders or even expensive cars. A few, however, strayed away from these subjects. Kanye West is one, and The Weeknd is another.</p><p>The Weeknd took an interesting route. He began as a dark R&amp;B artist. While he’s now widely known for his alternative R&amp;B, he initially started with a much darker tone. This stems from his experiences with the darker side of nightlife, drugs, and emotional turmoil — experiences that are evident in his first mixtape, <strong><em>House of Balloons</em></strong>. This mixtape introduces us to the then-anonymous Abel Tesfaye, who reveals that his late-night partying and drug use were far from glamorous and expresses a sense of regret in his songs. The unfiltered lyrics make it clear that The Weeknd is making a personal confession.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*nWQ4r2QVcyqgD5Nx" /><figcaption>Album cover for House of Balloons (2011) (Source:- <a href="https://hypebeast.com/2021/3/the-weeknd-house-of-balloons-album-stream-daniel-arsham-merch-collab">House of Balloons</a>)</figcaption></figure><p>This album gained significant attention for its unique blend of R&amp;B, indie rock, and electronic music, which stood out at the time. As mentioned earlier, this marked the beginning of The Weeknd’s dark R&amp;B style — a theme he continued in his next albums, <strong><em>Thursday</em></strong> and <strong><em>Echoes of Silence</em></strong>, where he again explored dark themes such as love, sex, and drugs.</p><p>In <strong><em>Kiss Land (2013–2014)</em></strong>, the last album in his dark R&amp;B phase, The Weeknd shifted focus to themes of fame, isolation, and the darker side of touring. Fans noticed clearer vocals while he maintained the dark themes. Critics also noted influences of 80s pop, new wave, Japanese anime, and horror films, particularly in tracks like “<strong><em>Wanderlust</em></strong>” and “<strong><em>Odd Look</em></strong>” (featured in the deluxe version).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8LlsvGLtuwTS44BF.jpg" /><figcaption>Album cover for Thursday (2011) (<a href="https://www.google.com/imgres?q=Thursday%20album&amp;imgurl=https%3A%2F%2Fpyxis.nymag.com%2Fv1%2Fimgs%2Fd57%2F2d6%2Fa6ae04128cdfa31cd658a3da6ceb58be6e-18-weeknd-thursday.2x.rhorizontal.w710.jpg&amp;imgrefurl=https%3A%2F%2Fwww.vulture.com%2F2016%2F08%2Fweeknds-hidden-masterpiece-thursday-turns-5.html&amp;docid=akJxSXluKHJLFM&amp;tbnid=qt1lEF-Z0z5CaM&amp;vet=12ahUKEwj4-9LLgd6HAxXE-TgGHeb_JT8QM3oFCIgBEAA..i&amp;w=1050&amp;h=700&amp;hcb=2&amp;ved=2ahUKEwj4-9LLgd6HAxXE-TgGHeb_JT8QM3oFCIgBEAA">Image Source</a>)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/0*5Nh5TrdTgHmeKSMF" /><figcaption>Album cover for Echoes of Silence (2011) (<a href="https://www.google.com/imgres?q=echoes%20of%20silence&amp;imgurl=https%3A%2F%2Fwww.rollingstone.com%2Fwp-content%2Fuploads%2F2018%2F06%2Frs-147833-rectangle.jpg%3Fw%3D624&amp;imgrefurl=https%3A%2F%2Fwww.rollingstone.com%2Fmusic%2Fmusic-album-reviews%2Fechoes-of-silence-255270%2F&amp;docid=Ft0mJZusLvCEpM&amp;tbnid=hjJfeiSRUWetvM&amp;vet=12ahUKEwi5pKC5gd6HAxUZzTgGHR4ZKNEQM3oECBcQAA..i&amp;w=624&amp;h=422&amp;hcb=2&amp;ved=2ahUKEwi5pKC5gd6HAxUZzTgGHR4ZKNEQM3oECBcQAA">Image Source</a>)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/0*s7GuQK4vuCGyPDxp.png" /><figcaption>Album cover for Kiss Land (2013) (<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2FKiss_Land&amp;psig=AOvVaw0FDK301gtbGrmxxFi7ej_2&amp;ust=1722953071296000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBEQjRxqFwoTCIDK2cOC3ocDFQAAAAAdAAAAABAK">Image Source</a>)</figcaption></figure><h3>Transition from Anonymity to Fame</h3><p>After establishing himself as a recognized artist, The Weeknd’s success continued to surge. In this section, I’ll discuss two albums that solidified his status as a global superstar: <strong><em>Beauty Behind the Madness (2015)</em></strong> and <strong><em>Starboy (2016)</em></strong>. These albums acted as a chemical reaction and a catalyst, accelerating his rise to fame. Let’s start with the breakthrough album aka the chemical reaction, <strong><em>Beauty Behind the Madness.</em></strong></p><h4>The Chemical Reaction</h4><p>In May 2015, The Weeknd released ‘<a href="https://youtu.be/yzTuBuRdAyA"><strong><em>The Hills</em></strong></a>’, a track that skyrocketed to a billion views within a year, both before and after the album’s release. But what made this song resonate so strongly with audiences? Two key factors stand out:</p><ol><li>The dark and gritty theme where The Weeknd again talks hedonism, infidelity, and the darker side of fame and relationships. Abel retains that raw and unfiltered lyrics but with the mix of dark RnB and alternative RnB.</li><li>Second, the music video played a crucial role in the song’s success. When I first watched it, I was struck by how perfectly the visuals aligned with the song’s mood. The video features a dramatic car crash and The Weeknd walking through a burning house — scenes that vividly capture the intensity of the lyrics. This combination of powerful imagery and haunting music created a memorable experience that resonated with many, leaving a lasting impression and elevating the song to iconic status.</li></ol><p>Although there are only two main reasons behind the success of ‘<strong><em>The Hills’</em></strong>, the song’s impact lies in how deeply it resonates with listeners. The Weeknd may have elevated these themes, but many people can relate to the emotions he explores. Breakups are a recurring theme in his music, and he delves into them extensively in this album and subsequent ones, producing some of his best work in this category.</p><p><strong><em>‘The Hills’</em></strong> wasn’t the only track to achieve massive success— ‘<a href="https://youtu.be/KEI4qSrkPAs"><strong><em>Can’t Feel My Face</em></strong></a>’ also hit a billion views, though it took three years after its release on July 29, 2015. This song exemplifies The Weeknd’s transition from his darker themes to a more upbeat, pop-driven sound. This shift was a welcome change in his style, and it resonated strongly with both old and new fans, highlighting his versatility as an artist.</p><p>If there is one track I like personally from this album is ‘<strong><em>Earned It</em></strong>’. It is about genuine admiration and respect for a partner, highlighting the idea that true love is something to be earned through actions and devotion. The Weeknd has delivered this song very beautifully with interesting background to compliment with. This became the title track for the already released film ‘<a href="https://en.wikipedia.org/wiki/Fifty_Shades_of_Grey_(film)"><strong><em>Fifty Shades of Grey</em></strong></a>’.</p><p>Overall, this album marked the beginning of Weeknd’s trail of multi billion viewed songs as we would be exploring more in the next section of this same topic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*bTczeWK8JdRyvp3m" /><figcaption>Album cover for Beauty Behind the Madness (2015) (<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fshop.republicrecords.com%2Fproducts%2Fthe-weeknd-beauty-behind-the-madness-lp&amp;psig=AOvVaw3EPaE1bgPFOBlRch89fXJX&amp;ust=1723015632558000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBEQjRxqFwoTCOihw9nr34cDFQAAAAAdAAAAABAE">Image Source</a>)</figcaption></figure><h4>The Catalyst</h4><p>2016 was an era where music peaked in both playability and artistry. We saw hits like <strong>Chainsmokers&#39; Closer; Tones &amp; I’s Dance Monkey, Geazy’s Me, Myself and I; Drake’s One Dance; Justin Bieber’s Love yourself; DJ Snake’s Let me love you; 21 Pilots’ Heathens, </strong>etc.</p><p>Now the list keeps going on and on because there were many good songs which released that year. It was during that time we also got Weeknd’s most famous album, <strong><em>Starboy(2016). </em></strong>This marked the departure from dark themes and dark RnB and into the upbeat RnB and the inclusion of my favorite microgenre of electronic music called <a href="https://en.wikipedia.org/wiki/Synthwave#:~:text=Synthwave%20(also%20called%20outrun%2C%20retrowave,decade%27s%20art%20and%20video%20games.">Synthwave</a>. The Weeknd has also collaborated with some of the famous artists like EDM duo <a href="https://en.wikipedia.org/wiki/Daft_Punk"><strong>Daft Punk</strong></a>, <a href="https://en.wikipedia.org/wiki/Lana_Del_Rey"><strong>Lana Del Rey</strong></a>, <a href="https://en.wikipedia.org/wiki/Kendrick_Lamar"><strong>Kendrick Lamar</strong></a><strong> </strong>and<strong> </strong><a href="https://en.wikipedia.org/wiki/Future_(rapper)"><strong>Future</strong></a><strong>.</strong></p><p>The track that drew the entire music community to The Weeknd’s <strong><em>Starboy</em></strong> album was the title track itself, <a href="https://youtu.be/34Na4j8AVgA"><strong><em>Starboy</em></strong></a><strong><em>. </em></strong>If you haven’t heard this chart-topper, you’re definitely missing out on one of the standout songs of that year. With its fast-paced beats and catchy lyrics, ‘<strong><em>Starboy</em></strong>’ captivated audiences as The Weeknd elevated his sound beyond his usual style. The success of the song was followed by other chart-topping hits like ‘<a href="https://youtu.be/qFLhGq0060w"><strong><em>I Feel it Coming</em></strong></a>’ and ‘<a href="https://youtu.be/diW6jXhLE0E"><strong><em>Party Monster</em></strong></a>’ both of which dominated the US Billboard Hot 100 for several weeks. The album showcases a diverse range of genres, from The Weeknd’s signature alternative R&amp;B to hip-hop and even disco. This variety not only highlighted his ability to convey a wide spectrum of emotions but also attracted a broader audience. His signature falsetto and smooth delivery were recognized as key strengths throughout the album, making <em>Starboy</em> a standout in his discography.</p><p>For hip hop lovers, The Weeknd had not left them out and released ‘<a href="https://youtu.be/JZjAg6fK-BQ"><strong><em>Reminder</em></strong></a>’ (<strong>A friend of mine used to blast this at full volume every time whenever he gets a chance</strong> 😅). This was well received by the audience for his introspective and autobiographical lyrics. The song addresses The Weeknd’s experiences with fame, his accomplishments, and the criticism he has faced. I was also pleasantly surprised to see many notable artists in the music video as guest appearances namely <a href="https://en.wikipedia.org/wiki/Drake_(musician)"><strong>Drake</strong></a><strong>, </strong><a href="https://en.wikipedia.org/wiki/Travis_Scott"><strong>Travis Scott</strong></a><strong>, </strong><a href="https://en.wikipedia.org/wiki/French_Montana"><strong>French Montana</strong></a><strong>, </strong><a href="https://en.wikipedia.org/wiki/ASAP_Rocky"><strong>ASAP Rocky</strong></a><strong> and </strong><a href="https://en.wikipedia.org/wiki/Metro_Boomin"><strong>Metro Boomin</strong></a><strong>.</strong></p><p>Starboy album in general brought the fans of all genres and sub genres towards him as he delivered those songs with grace and excellence. And as far as I think, most of the acquaintances I know or knew had started listening to The Weeknd only after this album. And understandably, the upbeat themes were something which was commonly seen in other artists is what made The Weeknd so relatable during that time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*RuYzrCyi83Ltmvp-8dcpXA.png" /><figcaption>Album cover for Starboy (2016) (I<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fopen.spotify.com%2Falbum%2F2ODvWsOgouMbaA5xf0RkJe&amp;psig=AOvVaw3Fc-v3WiBXvxvZcTzrKXWU&amp;ust=1723025626597000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBEQjRxqFwoTCOjeheeQ4IcDFQAAAAAdAAAAABAJ">mage Source</a>)</figcaption></figure><h3>To Future and Beyond</h3><p>What’s next for The Weeknd? Is my blog over yet? Is he involved in a scandal? Spoiler alert: He dropped another album, returning to his roots — dark R&amp;B. Released in 2018,<strong><em> My Dear Melancholy </em></strong>was born out of his recent heartbreak, revisiting the themes that originally made Abel famous, such as heartbreak, regret, and emotional turmoil. Songs like “<a href="https://youtu.be/M4ZoCHID9GI/"><strong><em>Call Out My Name</em></strong></a>” and “<a href="https://youtu.be/R0rKB_bsUNg"><strong><em>Wasted Times</em></strong></a>” quickly gained popularity, as fans appreciated the raw emotions and powerful delivery that had been somewhat missing from <strong><em>Starboy.</em></strong></p><p>The song “<strong><em>Call out My Name</em></strong>” uses the falsetto which The Weeknd uses quite often in his previous songs as well. The heartbreak The Weeknd went through in this song was quite apparent since many speculated that he might have been reeling from his breakup with <a href="https://en.wikipedia.org/wiki/Selena_Gomez"><strong>Selena Gomez</strong></a>. The next song in line is “Wasted Times”. This song explores the regret over the past relationships. A lot of fans suspected that this was made after his breakup with <a href="https://en.wikipedia.org/wiki/Bella_Hadid"><strong>Bella Hadid</strong></a> (Guy’s charm needs to be studied 😅).</p><p>These 2 were the ones which brought the lost dark RnB genre into the picture yet again as The Weeknd had started his musical journey from this genre. And if you are a fan of his <strong><em>Trilogy</em></strong> album, (For those who don’t know, trilogy album is the combination of his first 3 albums i.e. <strong><em>House of Balloons, Thursday and Kiss Land</em></strong>), you will see some semblance with his initial days of his music career when he produced the albums of the <strong><em>Trilogy.</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/465/1*JB5LNy8vWAV6ChxBYmKKWw.png" /><figcaption>Album cover for My Dear Melancholy(2018) (<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fwww.theguardian.com%2Fmusic%2F2018%2Fmar%2F30%2Fthe-weeknd-my-dear-melancholy-review&amp;psig=AOvVaw0JI3ufj-x0S26rDMo97kjq&amp;ust=1723696456915000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBQQjRxqFwoTCPilrOzT84cDFQAAAAAdAAAAABAE">Image source</a>)</figcaption></figure><p>But this not the only time when The Weeknd decided to drop a great album. When the pandemic was about peak, The Weeknd released the album <strong><em>After Hours</em></strong> in March of 2020. This personally became my second favorite album and I believe for most it would their most favorite album. And it’s for a good reason too.</p><p><strong><em>After Hours</em></strong> reintroduces genres Synthwave and dark RnB combination with prevalent songs like “<a href="https://youtu.be/4NRXx6U8ABQ"><strong><em>Blinding Lights</em></strong></a>”, “<a href="https://youtu.be/XXYlFuWEuKI"><strong><em>Save Your Tears</em></strong></a>” and “<a href="https://youtu.be/E3QiD99jPAg"><strong><em>In Your Eyes</em></strong></a>” which incorporate the above mentioned genres. This album also revolves around the same genres like heartbreak, loneliness and self destruction. Why did these songs become famous? Its simple actually. If you are a Weeknd fan then you would have noticed the similar dark themes but with very catchy background music. <strong><em>After hours</em></strong> album had songs which heightened the emotional impact of the same. The Weeknd uses this contrast to mirror the duality of his experiences — glamorous on the surface but filled with inner turmoil. Given how listeners appreciated fast paced music these days, especially the younger generation like the Gen-Z, The Weeknd exploits this by using the the bright, synth-heavy production to draw the listeners in. While he does that, the darker lyrics reveal the pain, regret, and self-destruction lurking beneath. This is the greatness of The Weeknd applying his musical skills.</p><p>After hours album took the internet by storm and maybe for some, this was exactly what they needed to sail smoothly through the tough years that were coming their way. As we all know the pandemic halted everyone’s work and affected our daily lives to a large extent. The album brought a lot of music fans together. Just check out the records <strong><em>After Hours</em></strong> broke. It has been listed below:</p><ol><li><strong>“Blinding Lights” </strong>became the longest-running song in the top 10 of the Billboard Hot 100, staying for 90 weeks.</li><li><strong><em>After Hours</em></strong> set records for the most global pre-adds on Apple Music and debuted with the highest first-day streams for an R&amp;B album on Apple Music.</li><li>The album debuted at number one on the Billboard 200, with The Weeknd becoming the first artist to top the chart in three different decades.</li></ol><p>As we can see, this turned out to be one of The Weeknd’s most successful albums — perhaps even more so due to the pandemic’s influence. But are we done? Not yet. There’s still one final album to discuss (as of writing this blog). While this album may not be a groundbreaking game-changer, in my opinion, The Weeknd once again delivered an impressive performance. So, without further ado, let’s conclude this blog by discussing the latest masterpiece from the great artist himself, The Weeknd.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MPYo_yGpzCiJhMam.jpg" /><figcaption>Album cover for After Hours(2020) (<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fdiscover.hubpages.com%2Fentertainment%2FThe-weeknds-After-Hours-Album-Review&amp;psig=AOvVaw2je-ZMtT9vMBIosLf-tG8y&amp;ust=1723909491593000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBQQjRxqFwoTCPj1v7rt-YcDFQAAAAAdAAAAABAY">Image Source</a>)</figcaption></figure><h3>End of the Line</h3><p>Well, here we are! If you’ve made it this far, it’s safe to say you’re a true Weeknd fan — just like me, since I managed to write this all the way through. 😄 To wrap things up, let’s dive into another incredible album from The Weeknd, ensuring you leave with the perfect taste of music lingering in your ears. After all, with The Weeknd, every <em>note</em> hits just right!</p><p>The final album we’re about to explore is titled <em>Dawn FM</em>. This album delves into, I would say, a more philosophical realm. The reason I say this is because the themes it explores revolve around life, death, and the transition between the two. It takes listeners on a journey through a metaphorical purgatory, framed by the unique premise of being “stuck in a radio station” hosted by none other than <a href="https://en.wikipedia.org/wiki/Jim_Carrey"><strong>Jim Carrey</strong></a>. The album leans more towards the alternative RnB rather than the usual dark RnB. It seems to be a continuation of the <strong><em>After Hours</em></strong> album as it retains the nostalgic sounds along with the heavy influences from 1980s synth-pop and new wave. Notable songs which may have similar genres include:</p><ul><li><strong>“Take My Breath”:</strong> This track has a pulsating, synth-driven beat reminiscent of 80s dance music.</li><li><strong>“Out of Time”:</strong> A smooth, nostalgic track that channels the soft rock and R&amp;B of the 80s.</li><li><strong>“Less Than Zero”:</strong> Combines upbeat, melodic synths with a bittersweet vibe, echoing 80s pop.</li></ul><p>If you had seen the album cover, The Weeknd has portrayed himself as an old man. This is because the aged depiction of The Weeknd symbolizes a reflection on the passage of time and the different stages of life. The album in itself represents a metaphorical transition, where listeners are guided through their own self-reflection and existential fears, with the idea of a radio station acting as a soothing, yet eerie, passage to the afterlife. If you’re wondering what my favorite song is from this album, it would be “<a href="https://youtu.be/E6zblNbGXA4"><strong><em>Sacrifice</em></strong></a><strong><em>”. </em></strong>The visuals and the production of the song was absolutely stunning. The lyrics share the same sentiment as with the album, with The Weeknd grappling with the idea of not wanting to sacrifice his own freedom and ambitions for the sake of love, acknowledging that he often prioritizes his personal goals over emotional connections.</p><p>The another song which always makes me appreciate the musical skills of Abel Tesfaye (for some reason I wanted to write his full name) is “<a href="https://youtu.be/u9n7Cw-4_HQ"><strong><em>Moth to a Flame</em></strong></a>”. This song was made in collaboration with a renowned EDM house supergroup <a href="https://en.wikipedia.org/wiki/Swedish_House_Mafia"><strong><em>Swedish House Mafia</em></strong></a><strong><em>. </em></strong>The song delves into the irresistible attraction and the destructive nature of desire. Interestingly, this was resonated by a lot of people since humans naturally are drawn into something which might be dangerous or in Weeknd’s words, ‘toxic’. This can lead to pain since desire is something which is inherent in us.</p><p>Overall, the <strong><em>Dawn FM</em></strong> cemented the musical mastery of The Weeknd in the history of RnB. From someone who made mixtapes to full blown albums with back to back hit singles and of course his songs being featured multiple times on Billboard Top 100, The Weeknd was lovingly called by fans as a musical genius who gave similar vibes as of late king of pop himself, <strong>Michael Jackson</strong>. The critics praised the album for its cohesive concept, production quality, and The Weeknd’s vocal performance. Critics also praised the seamless blend of retro and modern elements and the thoughtful exploration of themes related to the human condition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/0*_xyMUh3KF_1_09TZ" /><figcaption>Album cover for Dawn FM(2022)(<a href="https://www.google.com/url?sa=i&amp;url=https%3A%2F%2Fplay.anghami.com%2Falbum%2F1025150089&amp;psig=AOvVaw3_U6O95sdnLKP0TcI7UiZG&amp;ust=1724079267502000&amp;source=images&amp;cd=vfe&amp;opi=89978449&amp;ved=0CBQQjRxqFwoTCMjs9ffl_ocDFQAAAAAdAAAAABAc">Image Source</a>)</figcaption></figure><h3>Conclusion</h3><p>Being the number-one artist on Spotify is no small feat, and The Weeknd has achieved it by competing with some of the best in the industry and consistently delivering exceptional music. His introspective albums and relatable themes have helped him build a massive global following, and his influence in the RnB and pop genres is undeniable.</p><p>As we close this journey through The Weeknd’s career, I want to thank you for sticking around until the end. Whether you’re a die-hard fan or someone exploring his music for the first time, I hope this blog gave you some fresh insights into his artistry. Feel free to share your thoughts or feedback — I’d love to hear them. And, of course, don’t forget to add a few of his tracks to your playlist. After all, no weekend is complete without a little Weeknd. 🎶</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=914c9cfd7477" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Cleaning and Preview Techniques in Power BI for beginners]]></title>
            <link>https://medium.com/@tethys2004/data-cleaning-and-preview-techniques-in-power-bi-for-beginners-254915dda7a8?source=rss-d439605127e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/254915dda7a8</guid>
            <category><![CDATA[excel]]></category>
            <category><![CDATA[data-visualization-tools]]></category>
            <category><![CDATA[power-bi]]></category>
            <dc:creator><![CDATA[Rishith Singh Rawat]]></dc:creator>
            <pubDate>Mon, 29 Jul 2024 14:24:36 GMT</pubDate>
            <atom:updated>2024-07-29T14:24:36.691Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cjgHM71_pnZeCO67" /><figcaption>Graphs and graphs 📊(Photo by <a href="https://unsplash.com/@lukechesser?utm_source=medium&amp;utm_medium=referral">Luke Chesser</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a>)</figcaption></figure><p>Hey everyone! 👋🏻 I’ve recently started diving into Power BI, and wow, what a game-changer! If you’re like me and just beginning your journey with this fantastic tool, you’re probably discovering how crucial it is to clean your data and get a good look at it before jumping into the fun part — creating those beautiful visualizations. In this blog, I’ll share some essential tips on data cleaning and preview techniques in Power BI that I’ve picked up along the way. Let’s get started!</p><h3>Data Cleaning Tasks:</h3><h4>Remove/Replace Null Values, Duplicate Values, Missing Values</h4><p>Handling null values, duplicate values, and missing data is essential to ensure data integrity and accuracy.<br>In Power BI, you can easily perform these tasks:<br><strong>• Remove Null Values:</strong> Use the ‘Remove Rows‘ option in the Power Query Editor to filter out rows with null values.<br><strong>•</strong> <strong>Replace Null Values:</strong> Use the ‘Replace Values‘ feature to substitute null values with a specific value, such as zero or a placeholder.<br><strong>• Remove Duplicate Values:</strong> Utilize the ‘Remove Duplicates‘ option to eliminate duplicate rows based on one or more columns.</p><h4>Merge / Split Columns</h4><p>Sometimes, data may need to be combined or separated for better analysis. Power BI offers straightforward methods to merge or split columns:<br><strong>• Merge Columns:</strong> Use the ‘Merge Columns‘ feature in the Power Query Editor to combine multiple columns into one.<br><strong>• Split Columns:</strong> Use the ‘Split Column‘ option to divide a single column into multiple columns based on a delimiter or a number of characters.</p><h4>Add Columns, Remove Columns</h4><p>Modifying the structure of your dataset by adding or removing columns is often necessary to focus on relevant data:<br><strong>• Add Columns:</strong> Use the ‘Add Column‘ options such as ‘Custom Column‘ to create new columns based on your requirements.<br><strong>• Remove Columns:</strong> Use the ‘Remove Columns‘ feature to eliminate unnecessary columns from your dataset.</p><h4>Change Data Types</h4><p>Ensuring that each column has the correct data type is vital for accurate data analysis. Power BI allows you to change data types easily:<br>• Select the column(s) you want to modify and use the ‘Data Type‘ dropdown menu to select the appropriate data type, such as text, number, date, etc.</p><h3>Data Preview Techniques</h3><p>Before diving into data analysis, it’s important to preview and understand your data. Power BI provides several techniques to help you get a better grasp of your dataset:<br><strong>• Column Profiling:</strong> Use the ‘Column Profile‘ option in the Power Query Editor to view statistics like count, distinct count, and value distribution.<br><strong>• Data Preview:</strong> The data preview pane shows a snapshot of your data, allowing you to quickly spot any issues or patterns.<br><strong>• Filter and Sort:</strong> Apply filters and sorting to your data to identify trends and anomalies.</p><h3>Picture Showcase</h3><p>There are some screenshots which I have taken in order to show you few of the commands I have applied in my xlsx file. Now do note that some of these are done based on what is suitable for that particular xlsx/csv file however ultimately it totally depends on the kind of dataset you are given and based on that you decide whether merge/split column or to fill missing values or to change data types, etc.</p><h4>Removing Rows:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o95h2rDZmqrmKDoDzwBETA.png" /><figcaption>Figure 1: Blank Rows Problem</figcaption></figure><p>The above image shows that we have 70 percent filled data and 30 percent empty data. Here we have rows which are completely empty so we decide to remove those empty rows for simplicity sake.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aE_XURp9DkK27Pn2yk6VeA.png" /><figcaption>Figure 2: Before removing</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4BdS_4Fr_Wrt52EHjeHVBg.png" /><figcaption>Figure 3: After removing</figcaption></figure><h4>Renaming Columns</h4><p>Sometimes there are discrepancies in the column names which can often lead to confusion or misleading conclusions. Which is why Power Bi, like excel, also provides the option of renaming columns where you can clear those discrepancies. It also can be used to edit those correct columns where there are white spaces which is often not noticed to the human eye and can lead to errors while mentioning those particular columns in python or anywhere else.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XCYOZGtlfvykLvwhqFtJOw.png" /><figcaption>Figure 4: Renaming columns by double clicking on column head</figcaption></figure><p>As the picture caption says, by double clicking on the column head you can change the name of the column. If at all the name of the column needs to be changed again then you can see the taskbar where you can see the updated names. Just click there and change the name to your liking.</p><h4>Changing Data Type</h4><p>Sometimes, when we are required to do basic math operations then we need to make sure we are doing it on a column where it is only numbers. However, it is quite possible for these columns to show the numbers as a text datatype due to:</p><ul><li><strong>Leading Zeros:</strong> Some numbers might have leading zeros that are significant, such as postal codes or part numbers. Storing these as numbers would remove the leading zeros, but storing them as text<br>preserves them.<br><strong>• Mixed Content:</strong> If a column contains a mix of numeric and non-numeric data, it is often stored as text to accommodate all entries.<br><strong>• Special Characters: </strong>Numbers might include special characters like dashes or slashes (e.g., phone numbers, social security numbers, or dates in certain formats).<br><strong>• Data Consistency:</strong> Ensuring consistent formatting for numbers that need to appear in a specific way (e.g., with a certain number of decimal places or in a certain format).<br><strong>• Non-numeric Operations:</strong> When the data in the column is intended for non-numeric operations, such as concatenation or pattern matching.<br><strong>• User Input:</strong> Data entry forms might capture input as text to avoid conversion errors, especially if the input is not strictly numeric.<br><strong>• Preservation of Original Data:</strong> To prevent the alteration of data that might occur during numeric conversion, such as very large numbers losing precision or scientific notation issues.<br><strong>• Ease of Processing:</strong> Sometimes it’s easier to handle the data as text, especially if it will be processed by scripts or software that treat it as such.<br><strong>• Database Design:</strong> The database schema might have been designed with text fields to provide flexi￾bility in data entry.<br><strong>• Avoiding Mathematical Operations:</strong> To ensure that numbers are not mistakenly used in calcula￾tions, they are stored as text.</li></ul><p>While the above reasons are valid, it is better to convert it into a number format for all the statistical calculations. Below is the given image as to how we can convert the datatype from text to a decimal or whole number.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*69uEcts71ldyuJmbDSGJMA.png" /><figcaption>Figure 5: Right clicking on the desired column and following the options given on the image will lead to<br>changing the data type</figcaption></figure><h4>Removing columns</h4><p>We all know that some columns which don’t serve a purpose to visualization or any calculation can be considered redundant so we delete them. Again Power Bi also has very simple way of doing it</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1F79zbTmLnXND9kqJfgl_A.png" /><figcaption>Figure 6: Again we have to right click on the column and delete it if it serves no purpose</figcaption></figure><h4>Merging Columns</h4><p>When we have multiple columns which look useless but potentially be useful when we merge it then there is an option of merging multiple columns. For example when you have a column named ’First Name’ and<br>another named ’Last Name’, then these two can be merged to make a full name.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HoW93SX18Ov1rFC-Oeb0hg.png" /><figcaption>Merging Columns</figcaption></figure><p>We first press and hold left ctrl key and using the mouse we select the columns. then press on merge columns as given in the picture. Then there are a few options as to how they need to be merged. Then press<br>on apply and then Bam! Its done!</p><h3>Conclusion</h3><p>Data cleaning and understanding your data are fundamental steps in the data analysis process. Power BI<br>offers a variety of tools and features to help you efficiently clean and preview your data, ensuring that your<br>analysis is based on accurate and well-prepared data.<br>For more detailed tutorials and guides on using Power BI, visit the official <a href="https://learn.microsoft.com/en-us/power-bi/"><strong>Power BI documentation.</strong></a> <a href="https://www.microsoft.com/en-us/power-platform/products/power-bi/downloads"><strong>Power BI installation.</strong></a> <a href="https://online.hbs.edu/blog/post/data-visualization-tools"><strong>Power BI tools for help.</strong></a> <a href="https://learn.microsoft.com/en-us/power-bi/transform-model/desktop-external-tools"><strong>Another article for help.</strong></a> <a href="https://visme.co/blog/data-visualization-tools/"><strong>Some more tools like Power Bi for visualization.</strong></a><br>I have shared the .pbix file and the sample dataset in my GitHub. You can find it right here <a href="https://github.com/Icedsledgehammer/Power-Bi-practice"><strong>My work</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=254915dda7a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>