<?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 Hitansh K Doshi on Medium]]></title>
        <description><![CDATA[Stories by Hitansh K Doshi on Medium]]></description>
        <link>https://medium.com/@hkd159?source=rss-13fc7b6cd1aa------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Hitansh K Doshi on Medium</title>
            <link>https://medium.com/@hkd159?source=rss-13fc7b6cd1aa------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 10:04:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@hkd159/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[What is the Internet??]]></title>
            <link>https://medium.com/@hkd159/what-is-the-internet-e258c99be883?source=rss-13fc7b6cd1aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/e258c99be883</guid>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[internet]]></category>
            <category><![CDATA[basics]]></category>
            <dc:creator><![CDATA[Hitansh K Doshi]]></dc:creator>
            <pubDate>Sun, 29 Jan 2023 16:10:25 GMT</pubDate>
            <atom:updated>2023-01-29T16:10:25.706Z</atom:updated>
            <content:encoded><![CDATA[<p>It is something that has the answer to everything.<br>No, It is interconnected computers duh.<br>Le Google: a global computer network providing various information and communication facilities, consisting of interconnected networks using standardized communication protocols.</p><p>My friend, let me tell you how I see the internet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/220/0*QvC8WCF1xbkaegKH.JPG" /><figcaption><em>The Internet Messenger</em> by Buky Schwartz, located in Holon, Israel. (credit: Wikipedia)</figcaption></figure><p>1960 ARPA thought of a system where users can share computer timing. Computer timing means the use of multiple computer resources at once by multiple users by the means of multi-processing or multi-programming.<br>so… they Did it! 😆 That is the nerdy part of how it started.</p><h3>How I see the Internet</h3><p>Let us say that our computer is an island. And for anything to come in and out of it, we need a port. There are in total 65, 535 ports. Well all of them are closed initially. then we open some of the ports to receive ship (information from other places).</p><p>Now let us say you have a very funny meme you want to ship to your friend&#39;s island(to his computer). you can send it with the help of someone else’s ship. for eg: you send it using WhatsApp&#39;s ship. (you send a meme as a WhatsApp msg). Now you need to put your parcel on the ship in the correct way so that it goes to the right person. Lucky for you WhatsApp application will take care of that for you. <br>But wait a minute! what if your meme was opened by WhatsApp or someone else before it reaches your friend’s island? <br>For that, we seal it using a special lock. This lock can now only be opened by your friend. So it is safe.</p><p>Here islands are computer or mobile devices. A ship is a packet. The lock is encryption.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/622/1*QkC6A1ZdV5-kDTt2Row0Gw.png" /><figcaption>An island with ports. (credit: generated using stable diffusion)</figcaption></figure><p>Now you think I want my ship. (you want to host your server) and you can get your ship fairly easily. You just need to open your port 80. now you put a program in charge of looking into what is coming and where it should be directed. It checks if the given address is indeed on the island. The parcel is then passed to the address.</p><p>Here getting one ship means getting your server working. port 80 is commonly used for HTTPS servers. the port master or manager is your server code which handles all the incoming requests.</p><p>Now imagine you get a lot of ships from different islands. The port manager will be confused about checking if a parcel is at the correct place or not. That is why we have a standardized way to put the parcel on the ship. The way of placing the parcel is known as the protocol.</p><p>We also lock our parcels. Just in case, our ship gets into the hands of pirates our secrets remain safe.</p><p>So let us summarize what we have looked through till now.</p><p>Devices are like an island they communicate through ships(packets) and ports. We can use our ship or another’s ship. In either case, we need to protect our parcel. So, what we send we put a lock(encryption) on it. we make sure we follow the proper rules(protocols) for sending and packaging the parcel so that it is not lost.</p><p>Hope you enjoyed reading this analogy! 😃</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e258c99be883" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I automated the title finalization]]></title>
            <link>https://medium.com/@hkd159/i-automated-the-title-finalization-1ba0ec28e163?source=rss-13fc7b6cd1aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ba0ec28e163</guid>
            <category><![CDATA[pandas]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Hitansh K Doshi]]></dc:creator>
            <pubDate>Mon, 12 Dec 2022 15:49:58 GMT</pubDate>
            <atom:updated>2022-12-12T15:49:58.834Z</atom:updated>
            <content:encoded><![CDATA[<p>One fine day one faculty called me into the staffroom. She told me about the process of selecting the final year project. Then doubt was can this manual work be automated? I put on my coder glove and got started.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1006/1*RAfvj3mcx6TdN44sE4l_uw.png" /><figcaption>coder glove from if programming was an anime credit: Joma Tech</figcaption></figure><h3>Process :</h3><ol><li>Students are provided with a list of titles. They have to select two of them as their 1st and 2nd preference. 3rd one is the student’s own idea.</li><li>These selections are put into a google form. Responses are put into Google Sheets.</li><li>Their responses are then manually analyzed and titles are provided to the students.</li></ol><p>The First 2 steps are obvious and nothing of a problem. The 3rd one takes most of the time. Faculty first sorts data by 1st title. The unique titles are assigned to the teams that filled form first. If the first is already given then 2nd title and so on.</p><p>This is where I apply my skills and automate.</p><ol><li>We should sort wrt time, not by titles. Simply, because the person to put the title first will be getting it.</li><li>Make a list of titles assigned. and list of teams.</li><li>Output the list in style 😉</li></ol><p>Now let’s get started!!</p><p>Lets load up the data.</p><pre>df = pd.read_excel(&quot;/Project Title and Group List (Responses).xlsx&quot;)</pre><p>In this data, we have the following columns</p><pre>df.columns</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wT0jVbKvx128upsKdcVP2w.png" /><figcaption>columns in dataframe</figcaption></figure><p>Sort by time</p><pre>df.sort_values(&quot;Timestamp&quot;, inplace= True)</pre><p>Let&#39;s make a class to save the data in the structure.</p><pre>class Team:<br>  def __init__(self) -&gt; None:<br>    <br>    self.lead = {}<br>    self.member1 = {}<br>    self.member2 = {}<br>    self.member3 = {}<br>    self.titles = {<br>        &quot;title1&quot; : {&quot;name&quot; : &quot;&quot; , &quot;selected&quot;: False},<br>        &quot;title2&quot; : {&quot;name&quot; : &quot;&quot; , &quot;selected&quot;: False},<br>        &quot;title3&quot; : {&quot;name&quot; : &quot;&quot; , &quot;selected&quot;: False}<br>    }<br>    self.allocated_title = &quot;&quot;  <br>    self.guide = &quot;&quot;<br>    self.domain = &quot;&quot;</pre><p>Let&#39;s loop over the data and save them in class. While looping over this we can make a list of teams and allocated titles.</p><pre>titles = {}<br>teams = []<br>unallocated_teams = []<br><br>for idx in df.index:<br>  # initalizing class team and filling data into it <br>  team = Team()<br>  team.lead = {&quot;name&quot;: df[&quot;Name of Group Leader&quot;][idx], &quot;roll&quot;:  df[&#39;Roll no.of Group Leader&#39;][idx]}<br>  team.member1 = {&quot;name&quot;: df[&#39;Name of Member 1&#39;][idx], &quot;roll&quot;: df[&#39;Roll no.of Member 1&#39;][idx]}<br>  team.member2 = {&quot;name&quot;: df[&#39;Name of Member 2&#39;][idx], &quot;roll&quot;: df[&#39;Roll no.of Member 2&#39;][idx]}<br>  team.member3 = {&quot;name&quot;: df[&#39;Name of Member 3&#39;][idx],&quot;roll&quot;: df[&#39;Roll no.of Member 3&#39;][idx]}<br>  team.guide = df[&quot;Choice of Guides(Tentative)&quot;][idx]<br>  team.titles[&quot;title1&quot;][&quot;name&quot;] = df[&quot;Choice of Title 1 (Ideas Given by Faculty)&quot;][idx]<br>  team.titles[&quot;title2&quot;][&quot;name&quot;] = df[&quot;Choice of Title 2 (Ideas Given by Faculty)&quot;][idx]<br>  team.titles[&quot;title3&quot;][&quot;name&quot;] = df[&quot;Choice of Title 3 (It can be your Own Idea)&quot;][idx]<br><br>  # making list of teams <br>  title = &quot;&quot;<br>  if not titles.get(team.titles[&quot;title1&quot;][&quot;name&quot;]):<br>    team.titles[&quot;title1&quot;][&quot;selected&quot;] = True<br>    title = team.titles[&quot;title1&quot;][&quot;name&quot;]  <br>  elif not titles.get(team.titles[&quot;title2&quot;][&quot;name&quot;]):<br>    team.titles[&quot;title2&quot;][&quot;selected&quot;] = True<br>    title = team.titles[&quot;title2&quot;][&quot;name&quot;]  <br>  <br>  elif not titles.get(team.titles[&quot;title3&quot;][&quot;name&quot;]):<br>    team.titles[&quot;title3&quot;][&quot;selected&quot;] = True<br>    title = team.titles[&quot;title3&quot;][&quot;name&quot;]  <br>  <br>  team.allocated_title = title<br>  titles[title] = True<br><br>  if title == &quot;&quot;:<br>    unallocated_teams.append(team)<br><br>  teams.append(team)<br>  del team</pre><p>We did it! 🎇 <br>Now let&#39;s output this in style 😎</p><p>We can do this using rich text lib from python.</p><p>We first import the pprint which is pretty print and the decorator that allows us to structure the output from pprint. We will add this decorator and magic function to the team class.</p><pre>import rich.repr<br>from rich.pretty import pprint<br><br>@rich.repr.auto<br>class Team:<br>  def __init__(self) -&gt; None:<br>    ...<br><br>  def __rich_repr__(self) -&gt; rich.repr.Result:<br>    yield &quot;lead&quot;, self.lead<br>    yield &quot;Allocated Topic&quot;, self.allocated_title<br>    yield &quot;Topics&quot;, self.titles<br>    yield &quot;Members&quot;, [self.member1, self.member2, self.member3]</pre><p>Now its all set. Lets print it out!!</p><pre>print(f&quot;number of Unallocated teams: {len(unallocated_teams)}&quot;)<br>for i in range(len(unallocated_teams)): <br>    pprint(unallocated_teams[i], expand_all=True)<br><br>print(&quot;All teams: &quot;)<br>for i in range(len(teams)):<br>    pprint(teams[i], expand_all=True)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/1*sZUrfbgYgwKY-Ba8TuxHkw.png" /><figcaption>This is a dummy data output</figcaption></figure><h3>Conclusion:</h3><p>A simple task of sorting and assigning made a day&#39;s worth of work into a minute’s task. As the thumb rule of automation time to code must be lesser than the time used in doing tasks manually. That is achieved, So the mission is successful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ba0ec28e163" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Experience as an Intern at JP Morgan & Chase]]></title>
            <link>https://medium.com/@hkd159/my-experience-as-an-intern-at-jp-morgan-chase-ff12d0b65e87?source=rss-13fc7b6cd1aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/ff12d0b65e87</guid>
            <category><![CDATA[internship-experience]]></category>
            <category><![CDATA[learning-and-development]]></category>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[jpmorgan-chase]]></category>
            <dc:creator><![CDATA[Hitansh K Doshi]]></dc:creator>
            <pubDate>Sun, 14 Aug 2022 06:54:19 GMT</pubDate>
            <atom:updated>2022-12-03T05:42:29.566Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>“Lucky me!!”</strong> I thought when I read the acceptance mail from JP Morgan &amp; Chase</p><p>My name is Hitansh Doshi and this is my experience at JPMC.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*To0Zy22mm4iFE0rGk_7r5A.jpeg" /><figcaption>This beautiful looking building is actually the office!! No kidding it is good looking</figcaption></figure><p>Before starting the internship, we had to fill out forms regarding our preferred location and role, and our strengths in experience in different Technology. This info is used to place us in a Team. Around 2 weeks before my internship I got a call from my Manager. She asked me about my strengths and my tech stack. She selected a task for my internship task based on this information.</p><p>The first week was mainly occupied with onboarding tasks and getting to know the firm better. Who knew they would give us membership of O&#39;Reilly! For those who don’t know what O’Reilly is, Google it! It is one of my favorite publishers. Aside from onboarding tasks I got to know what my team works on. Meeting the team, setting up the workspace, and raising access requests sums up my first week. Also, I got 2 monitors for work! Slowly I started working on my tasks from the second week. It was my first time being part of the agile process. We had daily standup, frequent retrospect, and story sizing for the next sprint. In the end, I had to present my work to different teams.</p><p>To elaborate a bit on my work let me tell you about what part of the company I worked in. I was on the credit team. I made an internal tool that supports a critical service. It was a support tool for Dashboard Search. I was given what is their expectations and use case. Based on that I designed the frontend for that and the backend. Then I had to make test cases. I have worked with test cases in the backend but writing test cases for the front end was a whole new experience. Apart from this, I got some spare time to read and learn about Data pipelines and streaming data through O’Reilly.</p><p>From the second week onwards, I got to attend a lot of seminars from senior leaders. They shared their experience and learnings. Once at lunchtime, I got to meet a very experience employee, He shared his experience at ISRO as a research engineer. The team was super supportive. My manager was great. She even helps me prepare for my presentation.</p><p>One day while heading back home, I met some new joiners on the bus. We played dum sheras on the bus. It was so fun. In the end, my team gave me a farewell party at a nearby restaurant. It was a delightful experience. <br>To sum up my experience, bit nervous, byte of fun, MB of learning, and a ton of chocolate milk 😜.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff12d0b65e87" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Summary of VGG paper]]></title>
            <link>https://medium.com/@hkd159/summary-of-vgg-paper-40a14ee98beb?source=rss-13fc7b6cd1aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/40a14ee98beb</guid>
            <category><![CDATA[computer-vision]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Hitansh K Doshi]]></dc:creator>
            <pubDate>Fri, 12 Nov 2021 14:33:12 GMT</pubDate>
            <atom:updated>2022-01-28T09:45:18.880Z</atom:updated>
            <content:encoded><![CDATA[<h3>5 Minute Summary of VGG paper</h3><h3>Title</h3><ul><li>Very deep convolutional networks for large scale image recognition</li></ul><p>result:</p><ul><li>Large Scale image recognition</li></ul><p>Method:</p><ul><li>Very deep convolution</li></ul><h3>Abstract</h3><ul><li>Why / aim: To investigate the effect of the depth on the accuracy of recognition</li><li>Description/method: By increasing the depth using an architecture with a very small (3*3) convolution filter increasing depth up to 16–19 weight layers</li><li>Result: a model that generalizes well on different datasets</li></ul><h3>Introduction</h3><ul><li>Background: Convolution networks have seen great success in large scale image and video recognition this is possible due to:</li></ul><ol><li>Large public image datasets</li><li>high-performance systems</li><li>large scale distributed clusters</li></ol><ul><li>Intention: This paper address an important aspect of ConvNet architecture design i.e. depth. In this paper, the authors keep other aspects constant while steadily increasing the depth of the network by adding 3*3 filter convolution layers</li></ul><h3>Data</h3><ul><li>Dataset<br>ILSVRC-2012 dataset<br>It includes 1000 classes and 1.45 Million labeled images which are divided as follows:</li></ul><ol><li>Training 1.3 Million</li><li>Validation 50 Thousand</li><li>testing 100 Thousand</li></ol><ul><li>Performance Evaluation is done in 2 ways</li></ul><ol><li>Top — 1: Multi-class classification error: the problem of classifying instances into one of three or more classes</li><li>Top — 5: Computed as the proportion of images such that the ground truth category is outside the top 5 predicted categories</li></ol><p>Preprocessing:</p><ul><li>Isotropically rescaling (i.e. keeping image ratio same while scaling) to 224</li><li>Subtracting mean RGB value</li><li>Random cropping one crop per image per SGD iteration</li><li>Random horizontal flips</li><li>Random RGB shift</li></ul><h3>Model Architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/475/0*Hy4f-f-zdhnMcmCd.png" /><figcaption>Figure from VGG 2015 <a href="https://arxiv.org/abs/1409.1556">arXiv:1409.1556</a></figcaption></figure><p>Input is a fixed size RGB image of 224*224 px.followed by the preproccessing steps.<br>Given architecture is a stack of convolution layers with the filters of smallreceptive field: 3*3 or 1*1 in some cases<br>1*1 convolution layers can be treated as linear transformation followed by the non-linear transformation.<br>Stride is fixed to 1 px . Padding is such that the size is maintained.<br>pooling is done by max pooling layers with 2*2 window and the stride of 2.</p><p>After this 3 Fully Connected layers are used.<br>2 layes have 4096 channels eachand the last one is 1000 way soft-max layer.</p><p>All hidden layers used ReLU as activation function.<br>also, some models use Local response Normalization. But the use of Local response normalization did not improve the performance while increasing memory consumption and computation time.</p><p>There are 6 models based on this generic design</p><p>named as A, B, C, D, E, A-LRN also referred to as A-E, A-LRN in the paper.</p><p>A-LRN is A model with local response Normalization.</p><p>A-E 5 models differ only in depth from 11 to 19 layers.<br>With a width of convolution layers starting from 64 to 512 increasing in factor of 2.</p><h3>Input and Output</h3><ul><li>Input:fixed size 224*224 RGB image</li><li>Output: class</li></ul><h3>New techniques</h3><p>An important part of this paper is, why author selected 3*3 window field ?<br>Explanation by author:<br>2 conv layer of window 3 * 3 = 1 window of 5 * 5<br>3 conv layer of window 3 * 3 = 1 window of 7 * 7</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/582/1*cc4G0aPLRq2Nwc7118iZtQ.png" /><figcaption>Appling 3*3 filter on 5*5 image twice (generated using NN SVG)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/447/1*8W_q-nHYzua-Dlm5GuuqHg.png" /><figcaption>Appling 5*5 filter on 5*5 image (generated using NN SVG)</figcaption></figure><p>As one can see from the above images that two3*3 convolution layer is equivalent to one 5*5 convolution layer.</p><p>The question might arise why we should use 3*3 and not 5*5 or 7*7?</p><ol><li>using 3 convo layers means we get to involve 3 non-linear rectification layers. Which makes decision-making more discriminated.</li><li>It decreases the number of parameters</li></ol><p>Let us assume there are C number of channels then,<br>Number of parameters for 3 Convolution layers with a window of 3 * 3 <br>= 3(3² * C² ) = 27C²</p><p>Number of parameters for 1 Convolution layer with window of 7 * 7 <br>= 7² * C² = 49C²</p><p>One can also use a 1 * 1 window to increase the non-linearity of the decision function.<br>This small window technique works best if net is deep.</p><h3>Loss</h3><p>The input image is isotropically rescaled to a pre-defined smallest image size.<br>The fully connected layers are converted to convolution layers. First FC is converted to a 7 * 7 convolution layer, last two layers are converted to a 1 * 1 convolution layer.</p><p>Now, on perdition we get class score map i.e. we get confidence for each class.<br>to obtain the fixed size vector of class scores for the image, the class score map is spatially averaged (sum-pooled).<br>The final score is the average score from the original image and its horizontal flip.</p><p>Multi crop was drop as it doesn&#39;t justify the computation time required for the accuracy growth it showed.</p><h3>Model training</h3><p>Technique: mini batch gradient descent<br>Batch size: 256<br>Optimizer: multinomial logistic regression<br>Momentum: 0.9<br>Regularization: L2<br>Regularization multiplier: 5*10^-4<br>Dropout: for first 2 FC 0.5<br>Learning rate: 10^-2 decreased by factor of 10 when validation accuracy stopped improving<br>(learning rate was totally decreased 3 times)<br>learning stopped after 370k iterations 74 epochs</p><p>Model converge in fewer epochs due to :</p><ol><li>Implicit regularization imposed by the greater depth and smaller filter size</li><li>pre-initialization of a certain layer</li></ol><p>weights are initialized without pre-training by using Xavier weights</p><p>Two training settings were used</p><ol><li>Fixed scale</li><li>Multiscale</li></ol><p>First Approach (Fixed Scale)</p><p>Two fix sizes were considered S = 256, 384<br>The first network was trained with a size of 256.<br>Then the same weight is used to initialize the network for a size of 384. and the learning rate was set to 10^-3.</p><p>Second approach(Multiscale)</p><p>Set input image size to S where S from the certain range [256, 512]<br>this can also be seen as training set augmentation by scale jittering. Where a single model is trained on recognizing objects over a wide range of scales.<br>This approach was speed up by fine-tuning weights from the first approach</p><p>This was implemented with c++ Caffe toolbox with a number of significant modifications.To make use of multiple GPUs.<br>Images were trained in batches parallelly in GPUs.<br>Results from each GPUs was averaged to obtained graident for full batch.</p><p>4 NVIDIA Titan BLACK GPUs were used to give 3.75 times fast training then then single GPU.<br>It took 2-3 weeks for single net to train.</p><h3>Result</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/0*95RAfl3JNkj0p1u7.png" /><figcaption>Figure from VGG 2015 <a href="https://arxiv.org/abs/1409.1556">arXiv:1409.1556</a></figcaption></figure><h3>Summary</h3><p>In very deep convolution networks for large-scale image, classification depth is beneficial for classification accuracy.</p><h3>Data and code available</h3><pre>Dataset: https://image-net.org/challenges/LSVRC/index.php <br>Code: <br>    By author: https://worksheets.codalab.org/worksheets/0xe2ac460eee7443438d5ab9f43824a819<br>    Tensorflow implementation: <a href="https://github.com/tensorflow/models/blob">https://github.com/tensorflow/models/blob</a></pre><p>List Of References:</p><ol><li>[2015 ICLR] [VGGNet]<br><a href="https://arxiv.org/pdf/1409.1556">Very Deep Convolutional Networks for Large-Scale Image Recognition</a></li><li>Official implementation: <a href="https://worksheets.codalab.org/worksheets/0xe2ac460eee7443438d5ab9f43824a819">https://worksheets.codalab.org/worksheets/0xe2ac460eee7443438d5ab9f43824a819</a></li><li>TensorFlow implementation: <a href="https://github.com/tensorflow/models">https://github.com/tensorflow/models</a></li><li>Dataset: <a href="https://image-net.org/challenges/LSVRC/index.php">https://image-net.org/challenges/LSVRC/index.php</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40a14ee98beb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Git workflow]]></title>
            <link>https://medium.com/@hkd159/git-workflow-b2e4a784c53d?source=rss-13fc7b6cd1aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2e4a784c53d</guid>
            <category><![CDATA[beginner]]></category>
            <category><![CDATA[git]]></category>
            <dc:creator><![CDATA[Hitansh K Doshi]]></dc:creator>
            <pubDate>Tue, 27 Jul 2021 15:48:02 GMT</pubDate>
            <atom:updated>2021-07-27T15:48:02.011Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*RWWUeL6m69poubXB" /><figcaption>Photo by <a href="https://unsplash.com/@yancymin?utm_source=medium&amp;utm_medium=referral">Yancy Min</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Getting started</h3><p>Go to the folder where you want to put our project</p><p>Start git bash/ CMD/ Terminal</p><pre>git clone &lt;link-to-repo&gt;<br>cd &lt;repo-name&gt;</pre><h3>Making changes</h3><p>Making new branch and changing your branch for eg xyz with command</p><pre>git checkout -b xyz</pre><p>changing your branch to for examplexyz with command (for already existing branch)</p><pre>git checkout xyz</pre><p>Make the changes you want to make</p><p>If Their are a lot of changes make sure you commit after every big change you would like to save</p><p>To check your changes you can use</p><p>git diff</p><p>To commit your changes</p><p>If one file the use</p><pre>git add &lt;filename&gt;</pre><p>for all files use</p><pre>git add .</pre><p>Then</p><pre>git commit</pre><p>after commit your changes you can push it to the repo.</p><h3>Pushing changes</h3><p>git push &lt;repo&gt;</p><p>After pushing your changes you will be able to make a PR</p><p>While making a PR make sure you are merging</p><p>Team-75 - &lt;branch-intended-to-receive-changes&gt;&lt;- Team-75 - &lt;your-branch&gt;</p><h3>Pulling changes</h3><p>To pull code from main branch<br>git pull &lt;link-to-repo&gt;</p><p>To pull code from specific branch<br>git pull &lt;link-to-repo&gt; &lt;branch-name&gt;</p><p>Hope this is helpful! 😃</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b2e4a784c53d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>