<?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 Savankumar Gudaas on Medium]]></title>
        <description><![CDATA[Stories by Savankumar Gudaas on Medium]]></description>
        <link>https://medium.com/@gsavan?source=rss-4a35b1bbba04------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*flwUgfah7Z3ItrBlOMrIzw.jpeg</url>
            <title>Stories by Savankumar Gudaas on Medium</title>
            <link>https://medium.com/@gsavan?source=rss-4a35b1bbba04------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 16:36:26 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gsavan/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[Amazon DynamoDB Standard-IA vs DynamoDB Standard]]></title>
            <link>https://medium.com/@gsavan/amazon-dynamodb-standard-ia-vs-dynamodb-standard-78f81605bbdb?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/78f81605bbdb</guid>
            <category><![CDATA[dynamodb]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Wed, 09 Feb 2022 00:12:00 GMT</pubDate>
            <atom:updated>2022-06-10T01:25:50.626Z</atom:updated>
            <content:encoded><![CDATA[<p>Recently Amazon DynamoDB Standard Infrequent Access (DynamoDB Standard-IA) table class, which <em>AWS claims</em> helps you to reduce your DynamoDB costs by up to 60% for tables that store data that is infrequently accessed.</p><p>In this article let me give you the price dynamics of DynamoDB Standard-IA</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/568/0*izNPjd052G5hJVaH.png" /></figure><p>What DynamoDB costs break down</p><ol><li>Data storage</li><li>Write traffic</li><li>Read traffic</li><li>Other Costs are as below</li></ol><ul><li>On-demand backup</li><li>Continuous (PITR) backup</li><li>Table restore</li><li>Change data capture for Kinesis Data Streams</li><li>Data export to Amazon S3</li><li>DynamoDB Accelerator (DAX)</li><li>DynamoDB Streams</li></ul><p>Let’s just focus on data storage costs, which is a recurring cost, that grows as your storage grows.</p><p>DynamoDB storage: 20 Terrabytes</p><p>Cost per 20 Terrabytes: 20000 * $0.25 = $5000 per month</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*zzdl9aly5xupTGIq" /></figure><p>The cost slop is steep for DynamoDB Standard storage. If you do not use <a href="https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/TTL.html">the TTL function</a> of DyanmoDB you and if your data size grows steadily, you end up paying a steady storage cost.</p><p>If you are going with <em>DynamoDB Standard-IA, </em>you should be aware of the difference between Writes &amp; Read. Writes and reads are 25% more expensive than DyanamoDB Standard storage.</p><p>On-Demand Throughput Type</p><p>PriceWrite Request Units (WRU): $1.25 per million write request units</p><p>Read Request Units (RRU): $0.25 per million read request units</p><p>PriceWrite Request Units (WRU): $1.56 per million write request units <br>Read Request Units (RRU): $0.31 per million read request units <strong><em>DynamoDB Standard-IA</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/0*2xuJrc4C9st_8v7R.png" /></figure><p>Conclusion: Have a prediction of data usage and storage prediction before choosing DynamoDB Standard-IA</p><p><em>Originally published at </em><a href="https://bluecloudes8.blogspot.com/2022/06/amazon-dynamodb-standard-ia.html"><em>https://bluecloudes8.blogspot.com</em></a><em> on February 9, 2022.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=78f81605bbdb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AWS S3 Bucket limitations]]></title>
            <link>https://medium.com/@gsavan/aws-s3-bucket-limitations-6ad183b7faed?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/6ad183b7faed</guid>
            <category><![CDATA[s3]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[s3-bucket]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Tue, 11 Jan 2022 23:36:00 GMT</pubDate>
            <atom:updated>2022-06-08T00:54:17.612Z</atom:updated>
            <content:encoded><![CDATA[<p>Software systems always have some limits, reasons for limitations can varies. Hard limits boils down to limitations of underlying hardware resources. Soft limits can be to avoid noisy neighbor problems, that’s where rate-limiting, service quotas comes into place.</p><p>Let’s explore on limitations of AWS S3 bucket</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*81Huz3WaM6JBbDOg" /></figure><h3>Number of S3 buckets in a AWS Account</h3><p>By default, you can create up to 100 buckets in each of your AWS accounts. If you need additional buckets, you can increase your account bucket limit to a maximum of 1,000 buckets by submitting a service limit increase. There is no difference in performance whether you use many buckets or just a few.</p><p>TL;DR; 1000 buckets per account</p><h3>Cannot change S3 Bucket name or Region</h3><p>When you create a bucket, you choose its name and the AWS Region to create it in. After you create a bucket, you can’t change its name or Region.</p><h3>Reuse S3 bucket name</h3><p>You can reuse S3 bucket name, but.. when you delete the bucket and the name becomes available for reuse, another AWS account might create a bucket with that name. In addition, some time might pass before you can reuse the name of a deleted bucket.</p><p>TL;DR; If you want to reuse bucket name, safe option is don’t delete it ;)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6ad183b7faed" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Linux limits on threads per process]]></title>
            <link>https://medium.com/@gsavan/linux-limits-on-threads-per-process-642e9183de34?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/642e9183de34</guid>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[operating-systems]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Sun, 16 Feb 2020 09:53:00 GMT</pubDate>
            <atom:updated>2022-05-29T11:06:35.694Z</atom:updated>
            <content:encoded><![CDATA[<p>I came across an interesting learning on number of threads a process can hold.</p><p># of threads per process = total virtual memory / (stack size*1024*1024)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/0*w4ayiiZBZmB-kmlg.jpg" /></figure><ul><li>Increase in virtual memory can directly leads to increase in # threads per process. Virtual memory can be at most swap memory</li><li>Increase in stack size decreases # of threads per process</li></ul><p>Check Virtual Memory:</p><p>ulimit -v (default is unlimited, thus you need to increase swap memory to increase this)</p><p>Total Stack Size:</p><p>ulimit -s (default is 8Mb)</p><p>Command to update these values:</p><p>ulimit -s newvalue</p><p>ulimit -v newvalue</p><p><em>Originally published at </em><a href="https://bluecloudes8.blogspot.com/2020/02/linux-limits-on-threads-per-process.html"><em>https://bluecloudes8.blogspot.com</em></a><em> on February 16, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=642e9183de34" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gradient descent algorithm in Machine learning]]></title>
            <link>https://medium.com/@gsavan/gradient-descent-algorithm-in-machine-learning-88b3a28abc43?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/88b3a28abc43</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gradient-descent]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Sat, 25 Jan 2020 08:00:00 GMT</pubDate>
            <atom:updated>2022-05-29T10:33:36.823Z</atom:updated>
            <content:encoded><![CDATA[<p>Machine Learning and Deep Learning commonly uses Gradient descent (GD) to find a local minimum/maximum of a given function. Gradient descent minimises a cost/loss function (e.g. in a linear regression).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/0*FWbqjWKvrM2ASQnx.png" /></figure><p>In mathematical terms, Gradient descent is a way to minimize an objective function J(θ) parameterized by a model’s parameters θ ∈ Rd by updating the parameters in the opposite direction of the gradient of the objective function ∇θJ(θ) w.r.t. to the parameters. The learning rate η determines the size of the steps we take to reach a (local) minimum. In other words, we follow the direction of the slope of the surface created by the objective function downhill until we reach a valley.</p><p>There are three variants of gradient descent, which differ in how much data we use to compute the gradient of the objective function. Choice of variant depends on accuracy and time taken to compute parameters.</p><h3>Batch Gradient descent</h3><p>A batch gradient descent, computes the gradient of the cost function w.r.t. to the parameters θ for the entire training dataset:</p><p>θ = θ − η · ∇θJ(θ)</p><p>Batch gradient descent is guaranteed to converge to the global minimum for convex error surfaces and to a local minimum for non-convex surfaces.</p><ul><li>Needs whole dataset to perform just one update, so it can be slow</li><li>If given_data doesn’t fits to memory, computation can be expensive</li><li>Not applicable for online dataset</li></ul><h3>Stochastic gradient descent</h3><p>Stochastic gradient descent (SGD) in contrast performs a parameter update <em>for each training example</em> x (i) and label</p><p>y (i) : θ = θ − η · ∇θJ(θ; x (i) ; y (i) )</p><pre><strong>for</strong> i in range (nb_epochs): np.random.shuffle(given_data)<strong> for</strong> example in data : params_grad = calc_slope(loss_function , example , params )<br> params = params - learning_rate * params_grad</pre><p>Batch gradient descent performs redundant computations for large datasets, as it recomputes gradients for similar examples before each parameter update. SGD does away with this redundancy by performing one update at a time.</p><ul><li>Works with online dataset</li><li>Performant effective</li><li>Parameters can be fluctuating, it can help to jump to new and potentially better local minima.</li></ul><h3>Mini-batch gradient descent</h3><p>Mini-batch gradient descent finally takes the best of both worlds and performs an update for every mini-batch of n training examples:</p><p>θ = θ − η · ∇θJ(θ; x (i:i+n) ; y (i:i+n) )</p><p>In large-scale applications, the training data can have on order of millions of examples. Hence, it seems wasteful to compute the full loss function over the entire training set in order to perform only a single parameter update. A very common approach to addressing this challenge is to compute the gradient over batches of the training data.</p><ul><li>Reduces the variance of the parameter updates, which can lead to more stable convergence;</li><li>Can make use of highly optimized matrix optimisations</li></ul><p><em>Originally published at </em><a href="https://bluecloudes8.blogspot.com/2020/01/gradient-descent-alg-in-machine-learning.html"><em>https://bluecloudes8.blogspot.com</em></a><em> on January 25, 2020.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=88b3a28abc43" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Supervised learning and unsupervised learning]]></title>
            <link>https://medium.com/@gsavan/supervised-learning-and-unsupervised-learning-b8e8665b1dfe?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/b8e8665b1dfe</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Fri, 20 Dec 2019 01:32:00 GMT</pubDate>
            <atom:updated>2022-05-29T02:58:08.783Z</atom:updated>
            <content:encoded><![CDATA[<p>This article explains about basic Machine Learning concepts of Supervised learning and Unsupervised learnings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/320/0*JSy-sxP2JS48dfIg.jpg" /></figure><p>Human learning happens with external sense objects. You can now read English because you have learned English, but if some text is in a language which you don’t know you cannot understand it. Same applies to data driven algorithms which utilizes the data to judge are loosely called Machine Learning algorithms.</p><p>Let’s say, you want to develop Machine Learning algorithm to detect whether car exists in a image</p><h3>Supervised learning</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*extD7hYHgyvzAIWB.gif" /></figure><p>When you complete a <a href="https://bluecloudes8.blogspot.com/2019/12/supervised-learning-and-unsupervised.html">Captcha</a> and identify the images that have cars, you’re labeling images! A supervised machine learning algorithm can now use those pictures that you’ve tagged to make it’s car-image predictor more accurate.</p><h3>Mathematical definition</h3><p>In the supervised learning paradigm, the goal is to infer a function f : X → Y , the classifier, from a sample data or training set An composed of pairs of (input, output) points, xi belonging to some feature set X , and yi ∈ Y :</p><p>An = ((x1,y1),…,(xn,yn)) ∈ (X ×Y )n.</p><p>Typically X ⊂ IRd , and yi ∈ IR for regression problems, and yi is discrete for classification problems. We will often use examples with yi ∈ {−1, +1} for binary classification. <br>Unsupervised learning</p><p>In an unsupervised algorithm your example images are not labeled, i.e. you don’t say anything. Of course, in such a case the algorithm itself cannot “invent” what a car is, but it can try to cluster the data into different groups, e.g. it can distinguish that cars are very different from animals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*AuZwS3kja0PtJwX1.gif" /></figure><p>References:</p><p><a href="https://bluecloudes8.blogspot.com/2019/12/supervised-learning-and-unsupervised.html">https://link.springer.com/chapter/10.1007/978-3-540-75171-7_2</a> <br><a href="https://www.researchgate.net/publication/342121950_What_is_Unsupervised_Learning/link/5ee30fbe299bf1faac4e78f1/download">https://www.researchgate.net/publication/342121950_What_is_Unsupervised_Learning/link/5ee30fbe299bf1faac4e78f1/download</a></p><p><a href="https://www.codecademy.com/article/machine-learning-supervised-vs-unsupervised">https://www.codecademy.com/article/machine-learning-supervised-vs-unsupervised</a></p><p><em>Originally published at </em><a href="https://bluecloudes8.blogspot.com/2019/12/supervised-learning-and-unsupervised.html"><em>https://bluecloudes8.blogspot.com</em></a><em> on December 20, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b8e8665b1dfe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Applications of Artificial Neural Network]]></title>
            <link>https://medium.com/@gsavan/applications-of-artificial-neural-network-46a26f33cc7a?source=rss-4a35b1bbba04------2</link>
            <guid isPermaLink="false">https://medium.com/p/46a26f33cc7a</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[applications-of-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Savankumar Gudaas]]></dc:creator>
            <pubDate>Tue, 28 May 2019 03:11:00 GMT</pubDate>
            <atom:updated>2022-05-29T03:40:11.971Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*tXgAulTVz0Gg4NMn.jpg" /></figure><p>Artificial Neural Networks have capability successfully applied in speech or communication recognition. In the past decades, ML algorithms have applied widely in areas like acoustic modeling and automatic</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*05ZfUazFRrU2R_pi.png" /></figure><p>Computer vision aims at making computers to accurately understand and process visual data efficiently like videos and images. Computer vision alludes to the logical control which studies how to</p><p>separate data from images in artificial frameworks. Sub domains of computer vision include object detection and object recognition, object estimation, object position, event detection, scene reconstruction, image restoration, image editing, video enhancement, and statistical learning. Hence, in computer vision, ANN models are very useful.</p><h3>Pattern recognition</h3><p>The recent improvement in deep learning models has given novel ways to deal with the issue in recognition of a pattern or pattern recognition (PR). PR is a scientific area that focus in identification of sequence in each input. PR is a general concept that surrounds various subdomains such as speech tagging, regression, sequence labeling and classification. There are rapidly increasing needs for information processing and output, due to industrial development, that has new trend and challenges to PR.</p><h3>Face alignment</h3><p>Face alignment plays a role that is significant in diverse visual applications. In recent times ANNs has claimed successes in face alignment and face recognition and other models have shown successes. Interestingly DL techniques can be applying to explain genetic variants to identify pathogenic variants. Usually, combined annotation dependent depletion algorithm is popularly applied to interpret the coding and non-coding variants.</p><h3>Detection</h3><p>Detection in medical diagnosis, security, image objects, financial irregularity, a fault in a system, are being enhanced through ANNs application. Thus, ANN plays an essential role in the detection, particularly when applied to breast cancer. The performance of ANN can be relatively compared with other approaches in crime detection such as DNA and activity profiling and the use of big data for financial crime detection</p><p><em>Originally published at </em><a href="https://bluecloudes8.blogspot.com/2019/05/applications-of-artificial-neural.html"><em>https://bluecloudes8.blogspot.com</em></a><em> on May 28, 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=46a26f33cc7a" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>