{"id":135608,"date":"2026-09-01T22:31:59","date_gmt":"2026-09-01T17:01:59","guid":{"rendered":"https:\/\/www.guvi.in\/blog\/?p=135608"},"modified":"2026-09-01T22:32:02","modified_gmt":"2026-09-01T17:02:02","slug":"a-b-testing-for-machine-learning-models","status":"publish","type":"post","link":"https:\/\/www.guvi.in\/blog\/a-b-testing-for-machine-learning-models\/","title":{"rendered":"A\/B Testing for Machine Learning Models"},"content":{"rendered":"\n<p><strong>A\/B Testing ML Models<\/strong> is a controlled production experiment that compares two model versions using real traffic. Model A is usually the current production model, while Model B is the new candidate. Users are assigned randomly, and both models are evaluated against predefined business, technical, and model-quality metrics. The winning model is rolled out only after statistical and operational checks.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>TL;DR Summary<\/strong><\/h3>\n\n\n\n<ul>\n<li>A\/B testing compares an existing model with a new candidate model on live traffic.<\/li>\n\n\n\n<li>Random assignment helps reduce selection bias.<\/li>\n\n\n\n<li>Business metrics, model metrics, and guardrail metrics should all be tracked.<\/li>\n\n\n\n<li>Statistical significance does not automatically mean business value.<\/li>\n\n\n\n<li>Start with a small rollout and increase traffic gradually.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What Is A\/B Testing ML Models?<\/strong><\/h2>\n\n\n\n<p><strong>A\/B testing <\/strong><a href=\"https:\/\/www.guvi.in\/blog\/introduction-to-machine-learning\/\" target=\"_blank\" rel=\"noreferrer noopener\"><strong>ML <\/strong><\/a>models means running two versions of a machine learning model at the same time and comparing their outcomes.<\/p>\n\n\n\n<p>The existing model is called the <strong>control<\/strong> or <strong>Model A<\/strong>. The new version is the <strong>treatment<\/strong> or <strong>Model B<\/strong>. Traffic is split between them, and each model\u2019s performance is measured using the same evaluation criteria.<\/p>\n\n\n\n<p>For example, an online store may compare:<\/p>\n\n\n\n<ul>\n<li>Model A: the current recommendation model.<\/li>\n\n\n\n<li>Model B: a new recommendation model using improved embeddings.<\/li>\n\n\n\n<li>Primary metric: purchase conversion rate.<\/li>\n\n\n\n<li>Guardrail metrics: page latency, revenue per visitor, and unsubscribe rate.<\/li>\n<\/ul>\n\n\n\n<p>A\/B testing is different from offline validation. Offline metrics tell you how a model performs on historical <a href=\"https:\/\/www.guvi.in\/blog\/guide-to-data-exploration\/\" target=\"_blank\" rel=\"noreferrer noopener\">data<\/a>. A\/B testing shows how users and real systems respond to it in production.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Why Is A\/B Testing Important for ML?<\/strong><\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"940\" height=\"529\" src=\"https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/image-542.png\" alt=\"Why Is A\/B Testing Important for ML?\" class=\"wp-image-135610\" srcset=\"https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/image-542.png 940w, https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/image-542-300x169.png 300w, https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/image-542-768x432.png 768w, https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/image-542-150x84.png 150w\" sizes=\"(max-width: 940px) 100vw, 940px\" title=\"\"><\/figure>\n\n\n\n<p>A model can perform better on a test dataset but produce worse business results after deployment.<\/p>\n\n\n\n<p>This happens because:<\/p>\n\n\n\n<ul>\n<li>Production data may differ from training data.<\/li>\n\n\n\n<li>Users may react differently to new predictions.<\/li>\n\n\n\n<li>Small accuracy gains may not change user behavior.<\/li>\n\n\n\n<li>A more accurate model may be slower or more expensive.<\/li>\n\n\n\n<li>Improvements for one user group may hurt another group.<\/li>\n<\/ul>\n\n\n\n<p>A\/B testing connects model quality with real-world impact. It helps teams make deployment decisions using evidence instead of assumptions.<\/p>\n\n\n\n<p><strong>Example Box<\/strong><\/p>\n\n\n\n<p>An advertising model may improve offline click-through prediction but reduce actual revenue because it recommends low-value clicks. A\/B testing exposes this difference.<\/p>\n\n\n\n<p><em>A\/B testing for ML models runs two (or more) model variants in parallel on randomized user traffic and compares real-world metrics like CTR, revenue, or latency to decide which performs better with statistical confidence. Learn AI &amp; ML with HCL GUVI\u2019s<\/em><a href=\"https:\/\/www.guvi.in\/mlp\/artificial-intelligence-and-machine-learning?utm_source=A-B-Testing-for-Machine-Learning-Models\" target=\"_blank\" rel=\"noreferrer noopener\"><em> Artificial Intelligence and Machine Learning course<\/em><\/a><em>.&nbsp;<\/em><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How Does A\/B Testing for ML Models Work?<\/strong><\/h2>\n\n\n\n<p>A typical experiment follows these steps:<\/p>\n\n\n\n<ol>\n<li>Define the business and model objective.<\/li>\n\n\n\n<li>Select primary and guardrail metrics.<\/li>\n\n\n\n<li>Deploy Model A and Model B.<\/li>\n\n\n\n<li>Randomly assign users or requests.<\/li>\n\n\n\n<li>Log predictions, outcomes, and system behavior.<\/li>\n\n\n\n<li>Run the test for a predefined period.<\/li>\n\n\n\n<li>Analyze the results.<\/li>\n\n\n\n<li>Gradually roll out the better model.<\/li>\n<\/ol>\n\n\n\n<p>AWS describes a similar production process: deploy multiple model variants, route traffic between them, evaluate performance, and increase traffic to the selected variant.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Control and Treatment Groups<\/strong><\/h2>\n\n\n\n<p>The control group receives the current production model. The treatment group receives the candidate model.<\/p>\n\n\n\n<p>A common split is:<\/p>\n\n\n\n<ul>\n<li>90% Model A and 10% Model B for an initial canary test.<\/li>\n\n\n\n<li>50% Model A and 50% Model B for a full comparison.<\/li>\n<\/ul>\n\n\n\n<p>Assign users consistently. If possible, assign at the user or account level rather than randomly changing the model on every request.<\/p>\n\n\n\n<p>This prevents a user from receiving inconsistent experiences and reduces cross-contamination between groups.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Choosing the Right Metrics<\/strong><\/h2>\n\n\n\n<p>A strong A\/B test uses three metric categories.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>1. Primary Metric<\/strong><\/h3>\n\n\n\n<p>This is the main outcome used to determine success.<\/p>\n\n\n\n<p>Examples include:<\/p>\n\n\n\n<ul>\n<li>Conversion rate.<\/li>\n\n\n\n<li>Click-through rate.<\/li>\n\n\n\n<li>Revenue per user.<\/li>\n\n\n\n<li>Fraud detection recall.<\/li>\n\n\n\n<li>Customer retention.<\/li>\n\n\n\n<li>Recommendation engagement.<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>2. Model Metrics<\/strong><\/h3>\n\n\n\n<p>These measure prediction quality.<\/p>\n\n\n\n<p>Examples include:<\/p>\n\n\n\n<ul>\n<li>Accuracy.<\/li>\n\n\n\n<li>Precision.<\/li>\n\n\n\n<li>Recall.<\/li>\n\n\n\n<li>F1 score.<\/li>\n\n\n\n<li><a href=\"https:\/\/www.guvi.in\/blog\/read-logs-and-work-with-linux\/\" target=\"_blank\" rel=\"noreferrer noopener\">Log <\/a>loss.<\/li>\n\n\n\n<li>Area under the ROC curve.<\/li>\n\n\n\n<li>Calibration error.<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>3. Guardrail Metrics<\/strong><\/h3>\n\n\n\n<p>Guardrails help detect unintended damage.<\/p>\n\n\n\n<p>Examples include:<\/p>\n\n\n\n<ul>\n<li>Latency.<\/li>\n\n\n\n<li>Error rate.<\/li>\n\n\n\n<li>Infrastructure cost.<\/li>\n\n\n\n<li>Customer complaints.<\/li>\n\n\n\n<li>Churn.<\/li>\n\n\n\n<li>Fairness metrics.<\/li>\n\n\n\n<li>Safety violations.<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-pullquote\"><blockquote><p><br><strong>Best Practice: <\/strong>Define one primary metric before the experiment starts. Treat secondary metrics as supporting evidence rather than repeatedly searching for whichever metric looks best.<\/p><\/blockquote><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How Do You Measure Improvement?<\/strong><\/h2>\n\n\n\n<p>For a metric where higher values are better, relative improvement can be calculated as follows:<br><strong>Formula Box: Relative Improvement<\/strong><strong><br><\/strong>Relative Improvement=MB\u2212MAMA\u00d7100\\text{Relative Improvement} = \\frac{M_B-M_A}{M_A} \\times 100Relative Improvement=MA\u200bMB\u200b\u2212MA\u200b\u200b\u00d7100<br>Where:<\/p>\n\n\n\n<ul>\n<li>MAM_AMA\u200b is the metric for Model A.<\/li>\n\n\n\n<li>MBM_BMB\u200b is the metric for Model B.<\/li>\n<\/ul>\n\n\n\n<p>For example, if Model A has a 5% conversion rate and Model B has a 5.5% conversion rate:<br>5.5\u22125.05.0\u00d7100=10%\\frac{5.5-5.0}{5.0}\\times100 = 10\\%5.05.5\u22125.0\u200b\u00d7100=10%<br>The absolute improvement is 0.5 percentage points, while the relative improvement is 10%. Always report both because they communicate different things. For error metrics where lower is better, reverse the comparison:<br><strong>Formula Box: Error Reduction<\/strong><\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>Error Reduction=EA\u2212EBEA\u00d7100\\text{Error Reduction} = \\frac{E_A-E_B}{E_A} \\times 100Error Reduction=EA\u200bEA\u200b\u2212EB\u200b\u200b\u00d7100<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How Long Should an A\/B Test Run?<\/strong><\/h2>\n\n\n\n<p>There is no universal test duration. It depends on traffic volume, conversion frequency, expected effect size, and the amount of natural variation.<\/p>\n\n\n\n<p>Run the test long enough to capture:<\/p>\n\n\n\n<ul>\n<li>Weekday and weekend behavior.<\/li>\n\n\n\n<li>Normal traffic variation.<\/li>\n\n\n\n<li>Delayed outcomes.<\/li>\n\n\n\n<li>Seasonal or campaign effects.<\/li>\n\n\n\n<li>Enough conversions or events for analysis.<\/li>\n<\/ul>\n\n\n\n<p>Do not stop the experiment immediately after seeing a temporary improvement. Repeatedly checking the result and stopping early can increase false-positive conclusions.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>A\/B Testing Workflow for Production<\/strong><\/h2>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 1: Define the Objective<\/strong><\/h3>\n\n\n\n<p>Write a specific hypothesis:<\/p>\n\n\n\n<p>\u201cModel B will increase completed purchases without increasing page latency by more than 50 milliseconds.\u201d<\/p>\n\n\n\n<p>This is better than saying, \u201cModel B should perform better.\u201d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 2: Estimate Sample Size<\/strong><\/h3>\n\n\n\n<p>Use baseline performance, expected improvement, significance level, and statistical power to estimate the required sample.<\/p>\n\n\n\n<p>This helps prevent underpowered experiments that cannot distinguish meaningful improvement from random variation.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 3: Implement Stable Routing<\/strong><\/h3>\n\n\n\n<p>Use a feature flag or routing layer to assign users consistently.<\/p>\n\n\n\n<p>Log:<\/p>\n\n\n\n<ul>\n<li>User or experiment ID.<\/li>\n\n\n\n<li>Model version.<\/li>\n\n\n\n<li>Input timestamp.<\/li>\n\n\n\n<li>Prediction.<\/li>\n\n\n\n<li>Outcome.<\/li>\n\n\n\n<li>Latency.<\/li>\n\n\n\n<li>Errors.<\/li>\n\n\n\n<li>Relevant context features.<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 4: Start with a Canary<\/strong><\/h3>\n\n\n\n<p>Send a small percentage of traffic to Model B. Watch for crashes, latency increases, unusual predictions, and serious regressions.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 5: Increase Traffic Gradually<\/strong><\/h3>\n\n\n\n<p>If the canary is healthy, increase traffic to 25%, 50%, and eventually 100%, depending on the results.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Step 6: Analyze by Segment<\/strong><\/h3>\n\n\n\n<p>Overall averages can hide problems. Compare performance by:<\/p>\n\n\n\n<ul>\n<li>Device type.<\/li>\n\n\n\n<li>Geography.<\/li>\n\n\n\n<li><a href=\"https:\/\/businessmodelcanvastemplate.com\/blogs\/target-market\/guvi-target-market?srsltid=AfmBOor2pM0HuvCXzJ5qMucRj4LIG_b5k26B3D2FTvcJx8NbJOfZlu-O\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Customer <\/a>segment.<\/li>\n\n\n\n<li>New versus returning users.<\/li>\n\n\n\n<li>Language.<\/li>\n\n\n\n<li>Data quality.<\/li>\n\n\n\n<li>High- and low-volume groups.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>A\/B Testing vs Other Evaluation Methods<\/strong><\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table><tbody><tr><td><strong>Method<\/strong><\/td><td><strong>Main purpose<\/strong><\/td><td><strong>Uses live traffic?<\/strong><\/td><td><strong>Best use<\/strong><\/td><\/tr><tr><td>Offline evaluation<\/td><td>Test on historical data<\/td><td>No<\/td><td>Fast model comparison<\/td><\/tr><tr><td>Cross-validation<\/td><td>Estimate generalization<\/td><td>No<\/td><td>Model selection during development<\/td><\/tr><tr><td>Shadow deployment<\/td><td>Run a model without affecting users<\/td><td>Yes<\/td><td>Check infrastructure and predictions<\/td><\/tr><tr><td>Canary release<\/td><td>Expose a small group to a new model<\/td><td>Yes<\/td><td>Detect production failures<\/td><\/tr><tr><td>A\/B testing<\/td><td>Compare business and model outcomes<\/td><td>Yes<\/td><td>Controlled decision-making<\/td><\/tr><tr><td>Multivariate testing<\/td><td>Test multiple changes together<\/td><td>Yes<\/td><td>Complex product experiments<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p>A\/B testing is powerful, but it should not be the first evaluation stage. Use offline validation and shadow testing to catch obvious problems before exposing users to the candidate model.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Common Mistakes to Avoid<\/strong><\/h2>\n\n\n\n<ul>\n<li>Splitting traffic non-randomly.<\/li>\n\n\n\n<li>Changing the primary metric after seeing the results.<\/li>\n\n\n\n<li>Testing for too short a period.<\/li>\n\n\n\n<li>Stopping as soon as a metric becomes significant.<\/li>\n\n\n\n<li>Ignoring latency and infrastructure cost.<\/li>\n\n\n\n<li>Measuring only accuracy instead of business outcomes.<\/li>\n\n\n\n<li>Assigning users to different models on every request.<\/li>\n\n\n\n<li>Ignoring delayed conversions or long-term effects.<\/li>\n\n\n\n<li>Running many experiments without correcting for multiple comparisons.<\/li>\n\n\n\n<li>Deploying a winner without monitoring after rollout.<\/li>\n<\/ul>\n\n\n\n<div style=\"background-color: #099f4e; border: 3px solid #110053; border-radius: 12px; padding: 18px 22px; color: #FFFFFF; font-size: 18px; font-family: Montserrat, Helvetica, sans-serif; line-height: 1.6; box-shadow: 0 4px 12px rgba(0, 0, 0, 0.15); max-width: 750px;\"> \n  <strong style=\"font-size: 22px; color: #FFFFFF;\">\ud83d\udca1 Did You Know?<\/strong> \n  <br \/><br \/> \nA model with higher offline accuracy can produce lower revenue or engagement in production if its predictions change user behavior.\nA\/B testing should measure not only whether Model B is statistically better, but whether its improvement is large enough to justify deployment.\n<\/div>\n\n\n\n<p><em>A\/B testing for ML models runs two (or more) model variants in parallel on randomized user traffic and compares real-world metrics like CTR, revenue, or latency to decide which performs better with statistical confidence. Learn AI &amp; ML with HCL GUVI\u2019s<\/em><a href=\"https:\/\/www.guvi.in\/mlp\/artificial-intelligence-and-machine-learning?utm_source=A-B-Testing-for-Machine-Learning-Models\" target=\"_blank\" rel=\"noreferrer noopener\"><em> Artificial Intelligence and Machine Learning course<\/em><\/a><em>.&nbsp;<\/em><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Conclusion<\/strong><\/h2>\n\n\n\n<p><strong>A\/B Testing ML Models<\/strong> provides a reliable way to compare a production model with a new candidate using real traffic and measurable outcomes. It reveals whether a model improvement survives contact with real users, production systems, and business constraints.<\/p>\n\n\n\n<p>The strongest experiments combine random assignment, clear metrics, statistical analysis, guardrails, and gradual rollout. Do not choose a model simply because it has a better offline score. Choose it because it creates meaningful improvement without introducing unacceptable cost, latency, or user impact.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>FAQs<\/strong><\/h2>\n\n\n<div id=\"rank-math-faq\" class=\"rank-math-block\">\n<div class=\"rank-math-list \">\n<div id=\"faq-question-1787723485934\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>What is A\/B testing in machine learning?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>A\/B testing in machine learning compares two model versions on real traffic. The current model acts as the control, while the new model acts as the treatment.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723491002\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>What is the difference between offline testing and A\/B testing?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Offline testing uses historical or held-out data. A\/B testing evaluates models with live users or production requests and measures real-world outcomes.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723497236\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>How should traffic be split between two ML models?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Start with a small percentage for the new model, such as 5% or 10%. If the system remains stable, increase traffic gradually for a stronger comparison.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723505435\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>Which metrics should be used in an ML A\/B test?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Use a primary business metric, model-quality metrics, and guardrails such as latency, error rate, cost, fairness, and user complaints.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723513216\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>How do you know whether an A\/B test result is reliable?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Use a sufficiently large sample, predefined success criteria, confidence intervals, statistical testing, and a test duration that captures normal traffic variation.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723519968\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>Can a model have better accuracy but worse business performance?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Yes. Accuracy may not reflect user behavior, business value, latency, or revenue. That is why production metrics should be included in the experiment.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723526408\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>What is a canary rollout?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>A canary rollout sends a small percentage of traffic to a new model before increasing exposure. It helps detect failures and harmful regressions early.<\/p>\n\n<\/div>\n<\/div>\n<div id=\"faq-question-1787723540546\" class=\"rank-math-list-item\">\n<h3 class=\"rank-math-question \"><strong>Should every new ML model undergo A\/B testing?<\/strong><\/h3>\n<div class=\"rank-math-answer \">\n\n<p>Not necessarily. Offline testing or shadow deployment may be enough for internal systems or low-impact changes. A\/B testing is especially useful when the model directly affects user behavior or business outcomes.<\/p>\n\n<\/div>\n<\/div>\n<\/div>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>A\/B Testing ML Models is a controlled production experiment that compares two model versions using real traffic. Model A is usually the current production model, while Model B is the new candidate. Users are assigned randomly, and both models are evaluated against predefined business, technical, and model-quality metrics. The winning model is rolled out only [&hellip;]<\/p>\n","protected":false},"author":7,"featured_media":135962,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[933],"tags":[],"views":"20","authorinfo":{"name":"HCL GUVI","url":"https:\/\/www.guvi.in\/blog\/author\/guvipr\/"},"thumbnailURL":"https:\/\/www.guvi.in\/blog\/wp-content\/uploads\/2026\/08\/AB-Testing-for-Machine-Learning-Models-300x116.webp","_links":{"self":[{"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/posts\/135608"}],"collection":[{"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/comments?post=135608"}],"version-history":[{"count":4,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/posts\/135608\/revisions"}],"predecessor-version":[{"id":136468,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/posts\/135608\/revisions\/136468"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/media\/135962"}],"wp:attachment":[{"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/media?parent=135608"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/categories?post=135608"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.guvi.in\/blog\/wp-json\/wp\/v2\/tags?post=135608"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}