<?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 Ronak Agrawal on Medium]]></title>
        <description><![CDATA[Stories by Ronak Agrawal on Medium]]></description>
        <link>https://medium.com/@ronakagrawal2006?source=rss-19c9bd324a60------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*vj-VP_6YucHDG8CpXhDD7A.jpeg</url>
            <title>Stories by Ronak Agrawal on Medium</title>
            <link>https://medium.com/@ronakagrawal2006?source=rss-19c9bd324a60------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 03:56:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ronakagrawal2006/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[HiPlot — Better Parallel Coordinates and Super Easy Quick Start w/ EDA]]></title>
            <link>https://medium.com/@ronakagrawal2006/hiplot-better-parallel-coordinates-and-super-easy-quick-start-w-eda-10d981217f8?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/10d981217f8</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[exploratory-data-analysis]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Sun, 05 Dec 2021 11:28:59 GMT</pubDate>
            <atom:updated>2021-12-05T11:32:02.784Z</atom:updated>
            <content:encoded><![CDATA[<h3>HiPlot — Better Parallel Coordinates and Super Easy Quick Start w/ EDA</h3><h3>Overview</h3><p><a href="https://pypi.org/project/hiplot/">HiPlot</a> is a <strong>very-light interactive</strong> visualisation tool by <strong>Facebook AI Research</strong> that allows you to view n-dimensional relationships in a 2 D view and was released very recently.</p><p>While parallel coordinates is not a new methodology, and has been around figuratively for more than a century — <a href="https://en.wikipedia.org/wiki/Parallel_coordinates">1885</a>, the Key Benefits w/ the HiPlot implementation are listed as below.</p><ol><li><strong>Interactivity</strong> — highly customisable and fast</li></ol><ul><li>Able to visually transform scales to log-scale and percentile scales</li><li>Select color axis</li><li>Reorder columns in column axis (should help in discovery of relationships/corr. b/w variables)</li></ul><p>2. <strong>Scale</strong> — Able to handle interactivity w/ around <strong>half a million data points w/o getting sluggish</strong></p><p>3.<strong> Looks pretty</strong> and not dull!!! /\</p><p>In summary, <strong>discovering the story</strong> (assuming you have all the characters) or <strong>EDA</strong> became a lot easier</p><p>Let us build a narrative together and explore the <a href="https://www.rdocumentation.org/packages/datasets/versions/3.6.2/topics/ChickWeight">chicken weight dataset</a> to just see what all can we do.</p><p>The installation instructions are pretty standard and covered here. A quick command is as below:</p><pre>pip install -U hiplot # for pip users<br>conda install -c conda-forge hiplot # for conda users:</pre><h3>First look at the data in a jupyter notebook</h3><p>The following code allows one to load and display the dataset.</p><pre>chicks = pd.read_csv(&#39;chickweight.csv&#39;) # get the chicks :D import hiplot as hip data = chicks.to_dict(orient = &#39;records&#39;) # this transforms the chicks into a data dictionary.. hip.Experiment.from_iterable(data).display() #</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/779/1*79laaC26_KamCRnpTGe6ZA.png" /><figcaption>First look at the data</figcaption></figure><h3>Stating the obvious</h3><ul><li><strong>About the dataset</strong>: dataset that has weight measures of chicks for 21 days after they are born given a specific diet</li><li>The columns of the <strong>data frame</strong> are the vertical axes, and given that these axes representing coordinates are parallel to each other, ergo,<strong>parallel coordinates</strong>.</li><li>We’ll just rearrange the dataset so that we understand this a tad bit better and know what, how and where to look for.</li><li>We move the Chick column as the first column (this is <strong>chick_id</strong>), followed by Diet , as this is a different feed to the chickens, and our <strong>target variable</strong> is the weight of the chicks, which comes to the last, which leaves the time columns as the third one. This gives us the following ordering.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/951/1*Rw6TwbW9tUX3107Hd4gmlQ.png" /><figcaption>First view when we load the data</figcaption></figure><h3>Stating the Not so obvious</h3><ul><li>The first <strong>numerical</strong> <strong>column</strong> in the dataset is used for <strong>coloring</strong>. For our case, it was weight, which is also our target variable, so we don&#39;t really need to explicitly set it.</li><li>This parallel coordinates view would have automatically handled <strong>categorical</strong> variable of Diet. You can easily mark a variable categorical if need be(just right-click on the column text at top).</li></ul><h3>Free Stuff</h3><h3>Outlier Detection(Multi-dimensional, yes!)</h3><ul><li>At a glance, one can see that the extreme weights of chicks are in the <strong>range &gt;300</strong>.</li><li>Selecting records will highlight the row and associated values of different columns.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tVm7xwyjBBTInXYVj01i5A.gif" /><figcaption>Select and Exclude records from the interactive plot</figcaption></figure><ul><li>An added advantage is, you could see the multi-dimensional outliers in a glance here.</li><li>You can <strong>exclude</strong> them, by just clicking on the <strong>Exclude</strong> button and the plot would <strong>re-scale</strong> if needed</li></ul><h3>Helping us ask the questions</h3><p>It does not take a very fine eye to note the following for the last 2 variables — <strong>Time</strong> and <strong>Weight</strong></p><ul><li>The <strong>weight</strong> of the chickens do seem to <strong>increase w/ time</strong> (given the lines look parallel to each other w.r.t time) and the change in color</li><li>We also observe there is an <strong>increase in variance</strong> as the time increases. We could confirm this by looking at time vs weight plot. Guess what, you can do it right here!!</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8mcj_haTGmzRleHyuVQlUQ.gif" /><figcaption>Create a scatter plot for the variables of interest below.</figcaption></figure><h3>MultiVariate Relationships</h3><ul><li>A very powerful thing that we can do is, select a window of one or more column and observe the association w/ target variable.</li><li>Here, we obviously ask, for a set of 2 different diets say <strong>Diet 2</strong>, and <strong>Diet 3</strong>, <strong>how does the weight of chickens change with respect to time window of days?</strong>.</li><li>Here, we first <strong>color by Diet</strong> variable to get a sense of differentiation b/w two diets.</li><li>Select the Diet we wish to compare, and simply, select the rolling timeframe</li><li>We move the <strong>weight axes now in the center</strong>, to observe the growth of chicken w/ time.</li><li>The following view, then allows us to think in the direction that <strong>Diet 3</strong> fosters more growth than <strong>Diet 2</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EhbqpK6ttV-K3rpOjbMd4w.gif" /><figcaption>Observe the change in target w.r.t. 2 variables</figcaption></figure><h3>Additional Features</h3><ul><li><strong>Export as interactive html</strong>: One can easily export this as a standalone html to explore the dataset using just the csv. You can even share the html offline. The command for the same is</li></ul><pre>hiplot-render chickweight.csv &gt; chickweight.html</pre><ul><li><strong>DataFrame Access</strong>: It’s easy to get lost in plots and charts, and a lot of us tend to stick to tables. There is <strong>searchable dataframe</strong> right below the plot to enable access the entire dataframe content right at the bottom.</li><li><strong>Variable Transformation</strong> :Hi-Plot visual allows us to do few very useful transformations on the fly, including b/w <strong>types</strong>(categorical, numerical, timestamp) and <strong>Scale</strong> — (Log, Percentiles)</li><li>If you took a note above(Multivariate — relationships), we did <strong>center</strong> the weight axis and had the <strong>Time</strong> and <strong>Diet</strong> around it. This was for the purpose that it is <strong>easier to identify relationships b/w adjoining columns</strong>, and it gets <strong>difficult as we move away</strong> from the columns.</li><li>This view will definitely help us to <strong>get started w/ discovering the story</strong>, but might not necessarily help us have an exhaustive view. A good example of that is that <strong>we did miss the fact that some chickens in the above dataset ** prematurely died**</strong></li><li>Another thing to note, that <strong>discovering the order in which to view data might need a few iterations</strong> and might not be very obvious for a dataset having a lot of features. Do give a read to <a href="https://en.wikipedia.org/wiki/Parallel_coordinates">wiki</a> for better usage/inferences from the parallel coordinates view.</li></ul><h3>Sources</h3><ul><li><a href="https://pypi.org/project/hiplot/">HiPlot</a></li><li><a href="https://en.wikipedia.org/wiki/Parallel_coordinates">Parallel Coordinates Wiki</a></li><li><a href="https://www.rdocumentation.org/packages/datasets/versions/3.6.2/topics/ChickWeight">Chicken Dataset</a></li><li><a href="https://calmcode.io/datasets/chickweight.html">Chicken Dataset for Python</a></li><li>(Warmerdam (2021, Sept. 2). koaning.io: Bad Labels. Retrieved from <a href="https://koaning./">https://koaning.</a> io/posts/labels/)</li></ul><p>That was it for today. Happy learning :)</p><p><em>Originally published at</em></p><p><a href="https://theunalivepool.github.io/Active%20Threads/Tools%20and%20Workflows/HiPlot/">https://theunalivepool.github.io/Active%20Threads/Tools%20and%20Workflows/HiPlot/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=10d981217f8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Breaking needed Bad!]]></title>
            <link>https://medium.com/@ronakagrawal2006/breaking-needed-bad-8aeb6d28db5c?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/8aeb6d28db5c</guid>
            <category><![CDATA[stress]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Sat, 13 Jul 2019 09:08:05 GMT</pubDate>
            <atom:updated>2019-07-13T09:08:34.937Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/259/1*PVqHpd3VvosaqNC6pD5LcA.jpeg" /></figure><p>Ambition drives lots of us. A lot of us feel there are miles to go ahead, and there are. And given the only factor we can control most of the times is effort, and the fact that we are sincere by nature, we put in a lot of effort. This is good as far as it gets.</p><p>Later in the cycle, We start compromising on a lot of facets in life, with a specific goal in mind, with/without realisation. Slowly, that becomes a part of our routine. We develop a compulsive need to do something productive. This is the stage where ambition becomes a continuous source of stress. We might not realise it, but we are burnt out and are as dull as a blunt pencil or a flat tyre. No matter the effort put in, the movement in the right direction is really small.</p><p>The core reason behind this act, I believe, is the belief that <em>I am supposed to be taking an active effort</em> , and <em>If I don’t make a conscious effort, I l loose on the opportunity</em>. Unconsciously/consciously, one is trying to build an image of self into giving maximum effort. The intent is really genuine, but the effectiveness of the act decreases exponentially.</p><p>The supposed to do acts takes a toll on one’s mind as it needs a conscious effort. One could go as far as forcing hobbies(yes, talk about irony here.). Like, reading a book, watching a certain segment of videos/lectures, completing a certain set of courses just to get a sense of progress to self, that one is moving towards the right direction. This arises from the need of external validation(you can choose to disagree but it is what it is). And this is where all sense starts flushing down the the drain.</p><p>The objective of the journey becomes generating artifacts and not learning. An example is, when we start a course, we set timelines as to by when would one finish the course. This is good if this is augmented by curiosity to skim thru content and an excitement that one would be able to view the world in a more refined lens.</p><p>Bottom line, you are stressed and the efficiency has stalled. One also does meditation because, one feels they are supposed to do it!</p><p>What a sad situation to be in when we did nothing wrong. Working hard for something that we want to achieve, ain’t nothing wrong with that. But the stress, that it carries is a dead weight that one is definitely better without.</p><p>The answer to it?</p><p>It’s really simple.</p><p>Ask, Is it really that bad If I take a break? What is the worst case that would happen if I take a break of a day/week? The impact would be a day/week’s delay in your goal times(although the increase in efficiency would do better than to cover it up).</p><p>And most of the times, the delay is acceptable. Take a break, not because you are suppose to take one, but you need one, or rather want one.</p><p>Remember, breaks are where you do what you want to do, no judgments attached, no validations required from anyone else or from self for that matter.</p><p>There are not good or bad hobbies! It’s a hobby, don’t try to make a resume out of it. Live for yourself at least during breaks.</p><p>A shorter way of arriving at it would be asking to self, What would I do If I had no responsibilities/commitments? If I achieved my goal, what would I do post that?</p><p>Tell yourself, that during that break, you will be yourself. And if it pleases you to just lie around bed for the whole day, that’s it. That is what you do. Just be honest to self and you will be able to break the stress barriers. It’s really relieving. And that is what it is like when you are stress free.</p><p>I have learnt the importance of these things the hard way. A couple of folks did tell me to do so, but my reaction was, it’s easy to say something when you have already achieved your goals. Later in my experience, I have come to realise that, the journey need not be that stressful. It needs effort, no doubt. But one can enjoy it. And a week’s delay is just fine.</p><p>Hope this helps motivates someone to take a break and be more efficient :)</p><p>Credits: <a href="https://timesofindia.indiatimes.com/city/thane/hsc-ssc-students-replace-backpacks-with-travel-gear/articleshow/68568317.cms">Image Source</a></p><p><em>This post was originally published on </em><a href="https://www.linkedin.com/pulse/breaking-needed-bad-ronak-agrawal/"><em>here</em></a><em> on 13th July 2019.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8aeb6d28db5c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Mystic Stats — Pilot]]></title>
            <link>https://medium.com/@ronakagrawal2006/the-mystic-stats-pilot-d24e4a7dd11a?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/d24e4a7dd11a</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Sun, 25 Nov 2018 22:45:06 GMT</pubDate>
            <atom:updated>2018-11-25T23:11:36.596Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Mystic Stats — Pilot</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/630/0*BzyfHNxCTw37qan_" /></figure><p>Recently, I have seen many folks wanting to start with data science, study concepts/tests, but often cannot intuitively infer where to use a particular test, validation method when working with data.</p><p>So, under this series, I’ll be starting with a real-time problem at hand, and we’ll derive the intuition behind concepts and define the usage. We won’t even know the fancy names until the very end of it! Hence, the mystery.. Let’s see how many masters of mystic stats can we breed here.</p><p>With that context set up, let’s get started.</p><p>A client(say <em>John</em>) that we were engaging with had built a feature (Add to favourite) in their commercial website. The revenue model was based on registrations. He had to justify senior stakeholders that the feature was indeed valuable and generating more registrations($$$).</p><p>So, the business ask was:</p><p><strong>Is my feature(Add to favourite) generating more revenue?</strong></p><p>Well, how do we measure this?</p><p>Follow along to see.<br>The data collection part consisted of identifying a unique user, if the user engaged with the feature and finally if the user registered.</p><p>A simple way of answering the above question is to split users into two groups. Expose the new feature to them. The other half is oblivious to the new feature’s existence in the <strong>same timeframe</strong>. Now, just get the number of registrations for each.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oaMfQbg4WZsOflBv-fbqfA.png" /></figure><p>Guess what, this intuitive idea is called <strong>A/B Testing.</strong> This is ideal to have. But often, features are not built to allow this flow. Welcome to real world!</p><p>Given, how common this problem is in industry, we only have some users who had accessed the website without feature before, and some users who have accessed the website recently with feature in recent past. Well, now John was in a soup, as he had data in the following format.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b5z_GBJmwTBNoaLjLpAgzA.png" /></figure><p>Now, this time one could conclude that there is a 10 % increase in registrations. Hence the feature is adding value.</p><p>But WAIT, is that really the case???</p><p>One could question that the number of registrations is not constant and is changing. This increase could just be random.</p><p>Well, well…you don’t like being questioned(you should not, but that’s for another time), but can’t deny the fact that <strong>it is a valid question</strong>. So, you need a method/insight that lets you measure the chances of this changes happening randomly.</p><p>Let’s rearrange data first in a format that allows some insights. Below are the numbers we got.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T9-zKfPVGAV9TS5OOYXIzw.png" /></figure><p>Take a minute and understand the table above.</p><p>What we did above is just summarise the total information in a table. Let’s call this TABLE 1.<br>Now, we can see 2 set of outcomes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J8FlISP8ZDpWhnPxF8IWJQ.png" /></figure><p>Now the question I have ‘<em>Is ENGAGEMENT having an effect on REGISTRATION?</em>‘ or to rephrase, <em>Is there a relationship between these two variables</em>? or to rephrase again, ‘<em>Does Registration depends on Engagement?</em>‘ or rephrase finally, ‘<em>Are REGISTRATION and ENGAGEMENT INDEPENDENT?’</em></p><p>We want to know, what would be the user number if this feature had no impact.<br>Consider only the data in total rows. We can derive the insight that there is <strong>25% chance that user would register( 2500/10000)</strong>.</p><p>Let’s get the numbers in cells based on the chance we have. So, for the first cell, if we had 3000 users visit, and 25% of them engage, then the expected number is <strong>3000 x 25%= 750</strong>. Similarly, if we had 7000 users who visit and 25% of them engage, then expected non-registered users is <strong>7000 * 25% = 1750</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X7KnLRDiSm8ZaBuJgFH0VQ.png" /></figure><p>Let’s call this TABLE 2.</p><p>The above numbers give us the values if there was no relationship between the two variables under consideration. The condition we framed above ‘<em>if there was no relationship between the two variables under consideration</em>‘ really not an interesting one. It adds no further conclusion if this turns out to be true. Lets call this CONDITION 1.</p><p>Okay, we got this numbers if there was not a relation between the two variables.</p><p>Now, what we need is a probability of getting the values we observed if the values we expected follow a random normal distribution.</p><p>We identify the residual for each cell, square and normalise with the expected value(for obvious reasons), and have a sum across both the variables.</p><p>A typical cell would look like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/884/1*hz-BFcOM_J3mjxcTxBPfeQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/962/1*Be07T_hOga_GhrdE2ub18g.png" /></figure><p>The above value is the residue for the current criteria i.e. Registered and Engaged. We follow similar calculations for others. Now we add these across all cells.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E5lEE32BgTftE6vsq1KU2A.png" /></figure><p>This should be the ‘Total Residue’. Assuming that the error is normally distributed(ASSUMPTION 1), and we have to identify the chances of getting this error in the corresponding PDF.</p><p>Now, we just need to find the error value under this PDF and we have the answer to ‘What are the chances of getting observed value totally at random if the variables are independent?’.</p><p>Note that, if we sample from a standard normal random variable, and square the values(which we did before — to deal with negative values), the resulting distribution is an interesting one and has been studied for long. Let’s call this DISTRIBUTION OF INTEREST. We leverage the study here and look up the table to get that value.</p><h3>The moment of truth:</h3><p>Let’s take a pause here for a good news so far. Let me just tell you, we just covered the following:</p><ol><li>Contingency Tables.</li><li>Expected Frequencies.</li><li>Null Hypothesis(Hypothesis testing).</li><li>Chi-square statistic.</li><li>Assumptions of chi-square application(application of central limit theorem).</li><li>Chi-square distribution.</li><li>P-Values.</li></ol><p>Don’t believe it.Let’s just see how…</p><ol><li>TABLE 1 is a Contingency table.</li><li>TABLE 2 is Expected frequencies.</li><li>CONDITION 1 is NULL HYPOTHESIS.(<em>In summary, the uninteresting condition of your variables would be the ‘null hypothesis’.)</em></li><li>RESIDUAL VALUE is the CHI-SQUARE Statistic</li><li>ASSUMPTION 1 is the one of the assumptions when applying chi-square test, also an application of Central limit theorem.</li><li>DISTRIBUTION OF INTEREST is CHI SQUARE DISTRIBUTION.</li><li>To be honest, we did not cover it yet. It is below.</li></ol><p>The value we have from the chi-square table is ‘<strong>&lt;0.00001</strong>‘. This is the p-value.<br>That’s it. What we can conclude is, ‘<strong><em>The chances of getting a result at least this extreme are less than 0.001% if the variables are totally independent</em></strong>.’</p><p>Hence, we can reject the null hypothesis, that engagement is not really impacting registrations. John can have a peaceful night! So can you :).</p><p><em>Moral of the story: It’s all really intuitive. We don’t really need to recite formulas. Just need to have a moment of reflection and done!</em></p><p>That’s it folks in today’s episode of mystic stats!</p><p>Happy Machine Learning!!!</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2018/11/26/the-mystic-stats-pilot/"><em>ronakbits.tech</em></a><em> on November 25, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d24e4a7dd11a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[LSTM Cell — with a magnifier!]]></title>
            <link>https://medium.com/@ronakagrawal2006/lstm-cell-with-a-magnifier-42656dd6ce0d?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/42656dd6ce0d</guid>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[lstm]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Mon, 23 Apr 2018 18:06:47 GMT</pubDate>
            <atom:updated>2018-04-24T03:23:33.295Z</atom:updated>
            <content:encoded><![CDATA[<p>Hello!</p><p>Today, we’ll be talking about LSTM Networks QnA Style. The motivation for this is again as I have seen often when people read about LSTM’s, they have more questions than they have answers for. So, here I will try to give a gist of LSTM networks in comparison to FFN(Feed Forward Network or a regular NN).</p><p>So, Lets begin.</p><p>Q. How is LSTM different from a FFN cell(or a Feed Forward Network cell or a Node in hidden layer w.r.t. FFN)?</p><p><em>&gt;The fundamental difference between a LSTM Cell and a FFN cell is that the LSTM cell itself is a combination of 4 networks. Yes, you read that right.</em></p><p>Q. Why do we have 4 networks in a single cell?</p><p><em>&gt;Because we need memory or ‘previous state information’. And one way of achieving that is the implemented Networks.</em></p><p>Q. What are these 4 networks?</p><p><em>&gt;We’l go into the details shortly. Refer to section: ‘</em><strong><em>The Worthy Networks</em></strong><em>!’</em></p><p>Q. Why do we need ‘previous state information’?</p><p><em>&gt;Because we don’t want to loose context, and we believe that knowing the context could help better predict/classify the target.</em></p><p>Q. What if you are wrong, in the sense, previous state information is not required?</p><p><em>&gt;Well, we could be. Generally, in applications where ‘sequence’ is of importance, LSTM’s do well. They essentially capture the sequence information(previous state) in the ‘memory’ cells.</em></p><p>Q. Wait, you said <strong>Sequence</strong>. CNN’s(Convolutional Neural Networks) are good with sequences. I could get away by using CNN and not use LSTM?</p><p><em>&gt;Again, you could, depending on you application. One could say that there is a similarity between these two work as opposed to a FFN. They both take sequence(batch) of inputs. But that is where the similarity ends.</em></p><p><em>Once the data is in the hidden layers, for a CNN, the hidden state of the first image window(batch) will not be supplied to next image window; for a LSTM, this will be the case. I know this looks short, but this topic deserves an article on its own; I’ll try to come up with one soon.</em></p><p>Q. Okay, so you say that a single LSTM cell is a cell with 4 networks, then for a hidden layer in LSTM implementation, do I have multiple LSTM cells?</p><p><em>&gt;Well, Let me put it this way. A single LSTM cell is equivalent to a hidden layer in FFN. Again, that IS true. Similar to what you have number of nodes in a hidden layer, LSTM cell takes the number of hidden nodes as a parameter. This is the number of nodes in each of these 4 layers. As far as stacking goes, you can stack multiple LSTM cells, which will be similar to stacking of hidden layers when compared to FFN.</em></p><p>Okay, Let’s take a stab at the working of LSTM. The cell has following components.</p><ol><li>Long Term Memory</li><li>Short Term Memory</li><li>Input/Event</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/466/0*YhsfJFcQz_etLvEU." /></figure><p>The above diagram is read from left to right, bottom to top. So, as in case of FFN, you take input and predict output. Also, you use short term memory and long term memory to make a prediction. Also, we need to update the respective memories.</p><p><strong>The Worthy Networks!</strong></p><ol><li>The Learn Network.</li><li>The Forget Network</li><li>The Remember Network.</li><li>The Use Network.</li></ol><p>Lets take a look at each network in a bit detail.</p><ol><li>Learn Network: This network combines the Event Data and the Short term memory, learns new information from the event, and forgets what is not required.</li><li>Forget Network: The long term memory goes here and forgets what is not required. It needs the Short term memory as well to identify what is not required.</li><li>Remember Network: We combine the Long term memory and short term memory to get the updated Long term memory</li><li>Use Network: As the name suggests, it takes into account, whatever we already know (long term memory) and whatever we learnt recently(short term memory), and combine them to predict the output. The output becomes the new short term memory and prediction.</li></ol><p><strong>Learn Network:</strong></p><p>There is nothing to worry about.</p><p>We had 2 operations, Combine and Ignore for Short Term memory. And we are doing nothing more here.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/524/1*8jocBI7L7yh6GdaBcK_yNg.jpeg" /></figure><p>Why <em>tanh </em>for <strong>Combine</strong>, and <em>sigmoid </em>for <strong>Ignore</strong>? Because, it works! I am sorry here. If someone has a better understanding, please share. The documents I found had empirical conclusions, and not something that I could derive intuition from.</p><p><strong>Forget Network.</strong></p><p>Here, for Long Term memory, we need to forget what is not required. So, we multiply LTM by a forget factor. How do we get forget factor? Another network to rescue!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/541/1*eVqbcSkldXilxAHmwuMpJw.jpeg" /></figure><p><strong>Remember Network</strong></p><p>Here, we update the long term memory. How do we do it? Combine Long Term (whatever is important) with Short Term(updated with latest event data).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/666/1*_YIBHjG29hbEs2Vbo6U_IQ.jpeg" /></figure><p>The Learn Network and Forget Network have exactly what we need. So we just add them.</p><p><strong>Use Network</strong></p><p>Here, we take what is useful from Long term memory(Forget Network); take the most recent Short Term memory, combine, and that is going to be our new short term memory.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/656/1*ulTUPsImvUvr1U1tuJgiKw.jpeg" /></figure><p>One catch here is the <em>tanh </em>function. If we only wanted the updated LTM, we could just have used output of forget network directly. Empirically, the <em>tanh </em>is proved to be improving performance of networks under test. Again, here, if someone has a better understanding, please share inputs, and I will update the article to be more intuitive.</p><p><strong>THE MOMENT OF TRUTH!</strong></p><p>Putting it all together:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/788/0*I4657zsf4Wp4nkbj." /></figure><p>This is the famous LSTM Cell. Now if you plugin the individual networks in the respective cells, you should see the scary LSTM cell, that one generally sees. It should not look scary now. Try that as an exercise. It is actually fun.</p><p>Thats it guys.</p><p>Cheers. Happy Learning</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2018/04/23/lstm-cell-with-a-magnifier/"><em>ronakbits.tech</em></a><em> on April 23, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=42656dd6ce0d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Causing the COST Function]]></title>
            <link>https://medium.com/@ronakagrawal2006/causing-the-cost-function-9b0b4000cfac?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/9b0b4000cfac</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[statistics]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Wed, 07 Mar 2018 17:47:52 GMT</pubDate>
            <atom:updated>2018-03-07T17:51:22.940Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*nKZ9ySxGz1hNVCwz." /></figure><p>Have you ever wondered why does the cost function under gradient descent really looks like the way it does? If yes, today we will be building this up the right way. This is going to be a little longer, but totally worth it. So grab your coffee.</p><p>A little Kickstarter:</p><p>You have a model to build, which you are hoping to improve iteratively(in the true sense of the word). Let’s take the simplest problem of classifying 2 points as Red and Blue.</p><p>So, the task at hand: <strong>Classify the Red point as Red and the Blue point as Blue.</strong></p><p>To ‘improve’ model, you need to compare one model with another; so in effect, you select the ‘better’ one. How do you do that? A good model is one that does what we want with greater confidence. In our case, Classifying Red point as the red and Blue point as Blue.</p><p>To get confidence: We make the model predict output classes(Red and Blue) in probability.</p><p>For example: P(Red)=0.4 meaning that I have 40% confidence that the given point is Red. It also means for the same point that P(Blue)=0.6 given there are only 2 classes.</p><p>So far so good? Cool.</p><p>Now, for all points classified by my model, I need a number which has following property:</p><p><strong>If this is greater, we have correctly classified the points and if this is smaller, we have some mislabelled classifications.</strong></p><p>Then we <strong>maximize</strong> this number to get the best model.</p><p>All we have now is the probabilities of each points belonging to its actual class. Let’s take the product of these as the events are independent.</p><p>Say, we got for RED point: P(R)=0.6 and for Blue point: P(B)=0.7</p><blockquote>The number we have is product of P(R) for red points and P(B) for blue points. i.e P(R) *P(B) = 0.6 x 0.7 = 0.42</blockquote><p>Note that if it had a lower confidence of predicting the correct class, the number would go down. e.g if P(R) for red point was 0.3 instead of 0.6, the resultant would be 0.3 x 0.7 = 0.21 which is less than what we had before. So we do have the above property satisfied.</p><p>Just a little more, hang in there.</p><p>This could very well be our ‘model comparator’, but notice something, as we are taking products, and we have small numbers(in the range 0–1), the results are getting smaller. We would want to <strong>avoid products also because a slight change in one could have a significant impact.</strong></p><p>So we have Product is BAD. But sure, SUM could work. Let’s see. We want to convert PRODUCT →SUM? <strong>Log</strong> to the rescue, as</p><blockquote>log(a * b) = log(a)+log(b)</blockquote><p>So our model comparator becomes :</p><blockquote>Output = ln(P(R)) + ln(P(B))</blockquote><p>Well, we do have another problem here. Log of a number less than 1 is negative. We do need a positive number, so we add minus sign. Also, instead of maximizing it, we’ll have to minimize it, given we added a negative sign. Updating we have:</p><blockquote>Loss = — ln(P(R)) — ln(P(B))</blockquote><p>The ‘Output’ here is the information you have extracted.</p><blockquote>The function looks like — ln(P(y)) for the event occurred.</blockquote><p>If we have 2 classes the function looks like :</p><blockquote>Loss = — {y * ln(p(y))} — {(1-y) * ln(p(1-y))}</blockquote><p>To take the information when a particular event actually occurs, we have a representative term ‘y’ or ‘(1-y)’ to represent the occurrence of that event(There is no information gained when that event did not occur.)</p><p>Congratulations, you have built your own Loss function which appeared out of nowhere first. This is formally called the <strong>Cross Entropy Loss Function. </strong>Fancy, right?</p><p>Now, the above function is true for 2 output classes, i.e. Binary Classification.</p><p>For more output classes, let’s guess? Changes:</p><blockquote>The values will not be <strong>y</strong> and <strong>(1-y)</strong> instead y`, y“ …. y“` up to the number of output classes, with a product of log of probability of that event happening.</blockquote><p>Representing the ‘+’ for all elements with a summation would give you the formal definition of Cross Entropy Loss function which you can just search over. Il leave the rest to you.</p><p>That’s it for today, folks. Happy Machine Learning!</p><p>P.S.: As always, feedback welcome!</p><p>This entry was posted in <a href="https://ronakbits.tech/category/uncategorized/">Uncategorized</a> on <a href="https://ronakbits.tech/2018/03/07/causing-the-cost-function/">March 7, 2018</a> by <a href="https://ronakbits.tech/author/ronaktechbits/">Ronak Agrawal</a>.</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2018/03/07/causing-the-cost-function/"><em>ronakbits.tech</em></a><em> on March 7, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b0b4000cfac" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Demystifying Regularisation!]]></title>
            <link>https://medium.com/@ronakagrawal2006/regularization-what-why-b6fe48ef8e5e?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6fe48ef8e5e</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Thu, 01 Feb 2018 02:27:34 GMT</pubDate>
            <atom:updated>2018-02-21T14:11:53.477Z</atom:updated>
            <content:encoded><![CDATA[<p>The motivation for this specific topic?<br> Often when I connect around, the answer to ‘When to use Regularization’ is ‘To prevent overfitting of model’. While that is true, it is important to understand how it works and get a sense to have an overall understanding of your model. So, as suggested by one of my colleague at ThoughtWorks, here’s my attempt at intuiting it.</p><p>Prerequisites:<br> Gradient Descent(some basics)</p><p>Let’s get started.</p><p>One way to think of your model is a curve which separates some outputs from other. Now, this curve is defined by a function(could be a line, for example, is defined by function y=mx+c).</p><p>The function will have a slope at any given point in time. This slope gets steeper as the constants associated(weights of coefficients) with them increase rapidly.</p><p>To do gradient descent properly, we need a model which has a gentle slope and not a steep slope. For intuition, one can say, that the model with steep or very low slopes are very certain and gives little room for gradient descent.</p><p>Also, if the certain model makes errors in classification, the errors generated will be large and will be difficult to tune that model to correct it.</p><p>The intuition here is we say that larger coefficients could probably lead you into overfitting.<br>So we would want to avoid large coefficients.</p><p>Having established above, what do we do?<br>We punish high coefficients. And how do we do it?<br>Given that we minimize error using error function, and we want to stay away from high coefficients; we add a penalty term to the error function that will increase error when the coefficients are high, and will not contribute much to error when the coefficients are small.</p><p>Now, all we need is some way to get the measure of the coefficients(weights), and to this measure, we can apply our penalty term.</p><p>2 approaches to measure coefficients:<br>a. Simply, add the absolute values of weights(given the weights can be negative and we don’t want them canceling the positive weights) and multiply by our penalty term lambda.<br>b. Sum of squares of weights and multiply by our penalty term lambda.</p><p>The first one is L1-Regularization, and the latter is L2-Regularization.</p><p>To summarize:<br>a. Larger coefficients could well probably lead to overfitting<br>b. Need to punish them<br>c.Updatefunctionunctoin to add a term that does the second.</p><p>Yes, that was all to it. You should have an idea of what exactly are we doing when we say, ‘regularize model’. That’s it for today.</p><p>Oh yes, if you have feedback, please share.</p><p>Happy Learning folks.</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2018/02/01/regularizationwhat-why/"><em>ronakbits.tech</em></a><em> on February 1, 2018.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6fe48ef8e5e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning : How NOT to get started with Machine Learning.]]></title>
            <link>https://medium.com/@ronakagrawal2006/machine-learning-how-not-to-get-started-with-machine-learning-99228becb24a?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/99228becb24a</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Fri, 27 Oct 2017 12:17:58 GMT</pubDate>
            <atom:updated>2017-10-28T18:02:19.572Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/250/0*Sv7_-WP-t7IvlDW8." /></figure><p>Over past year, I have seen quite a number of folks start with Data Science. And there are plenty of articles indicating the surface area of the entire domain. Many start, but few continue. Here, I try to list some traps that could stall an aspiring Data Scientist’s progress. As always, feel free to share any feedback you have.</p><p>A typical journey on ML looks like below:</p><ol><li>Get fascinated by all the hype and aim to become a data scientist.</li><li>Get started with Andrew Ng’s ML course.</li><li>Don’t understand what’s going on for 3 weeks, and wonder when will we start ‘actual’ deep learning.</li><li>Switch to the <em>fast.ai</em> course.</li><li>Get a feel of learning but deep down still not understanding how this works.</li><li>Feel hollow and abandon all hope.</li></ol><p>My journey was a little different but let me share my learnings over past year. This will definitely be the same structure of bullet points, but a little more verbose.</p><ul><li>Don’t be in a hurry to finish things up. I’ve learned this lesson the hard way and it cost me ~3 months of effort. Even after completing some courses and solving assignments, I forgot all about it. Before moving to next part, pause for a moment, ask yourself, ‘<em>Did I understand what was done?</em>‘. If the answer is ‘YES’, pause again and ask ‘<em>Explain WHY?</em>‘. If you have a hiccup there, don’t hesitate to go back, re-validate your understanding.</li><li>Don’t just know the ‘HOW’ and ‘WHAT’ part of it, but focus more on the ‘WHY’ part. Do understand steps to do logistic regression but also understand the rationale behind on why do we need each step. Trust me, it will come back to haunt you in not so distant future if you have not understood it. If you need to go and look for variance and standard deviation, go and look for it.</li><li>Don’t be paranoid if your progress is slow. ‘Getting there right’ is far more important than ‘Getting there fast’.</li><li>Do give a shot at <a href="https://www.kaggle.com/">kaggle</a> competitions. No amount of course experience can come close to exposure shared by a kaggle competition.</li><li>Know that you will have to understand your data to solve an ML problem. It’s not an option. You cannot just try models and hope that a Neural Network will work magically(well, it will do that to some extent.). But for it to be of any practical use, you will have to understand the nitty-gritty details.</li><li>You will have to care about numerical, categorical variables, measurements et all. You have feature engineering an integral part of an ML solution.</li><li>Know that ML is Applied Statistics. And that is what Mr. Andrew Ng was trying to teach you all through Octave.</li><li>Understand visualizations and by that I mean, understand when would you plot a certain graph instead of getting snsplot or ggplot right(that too is important but understand the intent first. ) e.g. If you have to confirm around association between two variables, scatter plot would make sense whereas if no association is required, a box plot should suffice. The ‘HOW’ part can always follow once you have the ‘WHY’ bit.</li><li>Lastly, ‘Gut’ feeling is definitely important but visualizations will help you develop and direct your intuition. Don’t shy away from them. Knowing your data via plots can be better than skimming thru a large csv.</li></ul><p>That’s it, folks. Happy Machine Learning!</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2017/10/27/machine-learning-how-not-to-get-started-with-machine-learning/"><em>ronakbits.tech</em></a><em> on October 27, 2017.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=99228becb24a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Re-Structuring Machine Learning Execution]]></title>
            <link>https://medium.com/@ronakagrawal2006/re-structuring-machine-learning-execution-772883823020?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/772883823020</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[tips]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Thu, 12 Oct 2017 02:28:40 GMT</pubDate>
            <atom:updated>2017-10-23T07:53:09.410Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*Bo6D2oZmjgPEe6YN." /></figure><p>Recently, I gained some insight on Structuring Machine Learning projects. How I wish I had this insight when we did some experiments in ML domain in not so distant past. Anyways, I wouldn’t want anybody else to get hit by the same stones, so below is a crux of what I think I have understood. Feel free to share your inputs if you are not totally onboard or have a different insight.</p><p><em>[3 mins read]</em></p><p>A typical ML project pathway would have followed as the ToDo list:</p><ol><li>Fit Training set well.</li><li>Fit Dev Set well.</li><li>Fit Test Set well.</li><li>Perform well in real world.</li></ol><p>Here is a list of guidelines that could help a team to work efficiently. I know these are very subtle inputs, but I think, when working iteratively, it could have a significant impact.</p><p><strong>For Overall Model:</strong></p><ul><li>Define a Single Calculation metrics to know which model is better.e.g. Instead of having Precision and Recall as 2 values, get F1-Score as a single metric.</li><li>To Split dataset, Identify the right split between your train/dev/test set. I know it was a standard practice to have a 60–20–20 split. If you have a good dataset(of the order ~100K), you could do well with 98–1–1 split(yes, you read that right). The objective of having a Dev set is to have a representative sample is that it gives you the confidence of your model performance. If 10k examples could do it, why go for an overkill.</li></ul><p><strong>For Dev/Test set:</strong></p><ul><li>Have dev set and test set from the same distribution. Spending some hours on getting this right could help a lot, as we can definitely rely more on our dev set metric. Failing to do so, could lead you spending efforts on tuning a product that client did not ask for.</li><li>Make sure the data in these sets reflect data you expect to get in future and consider important to do well on. e.g. If you care more about HD images, it might NOT be a good idea to optimize your model for some random quality of images downloaded from the web. These are the cracks which can cause a huge rework later by collapsing your foundation.</li></ul><p><strong>Overall Strategy:</strong></p><p><strong>Orthogonality in Execution:</strong> This is by far one of the most important thought processes which were too obvious yet only after knowing how to execute it, I realize we missed it.</p><p>Often when working on ML project, we think about too many things at once. And everybody ends up working on all the problems. Yes, it is as bad as it sounds.</p><p>For example, Identifying a Single Calculation Metric and Improving It are totally different problems. Similarly, getting a bigger dataset and extracting appropriate features out are totally different problems. Worry about them separately.</p><p><strong>Error Analysis:</strong></p><p>A good thing to do when doing error analysis would be to:</p><ul><li>Get the mis-labeled dev set examples(where you got incorrect classification).</li><li>Get the distribution of failures (basically, categorize them why this could have been misclassified and count them).</li><li>Tag them with the complexity to fix those against the gain in the improvement of the model.</li></ul><p>Now, you have a better sense of what to pursue.</p><p>As a bonus, you could split teams to fix those issues orthogonally.</p><p>Another important fact to be aware of at all moments is to maintain same distributions of dev set and test set if you are altering dev sets. An example could be, if you add more HD images to your dev set, make sure you update your test set to reflect the same distribution.</p><p><strong>Human level performance:</strong></p><p>One can use intuitions from folks around if we are struggling to achieve human-level accuracy.It might be obvious but I still mention these, as one might miss these with too many moving parts.</p><ul><li>Why did the human get it right?</li><li>Better analysis of Bias/Variance</li><li>Get Labelled data from humans.</li></ul><p><strong>Revisiting Metrics:</strong></p><p>It’s a good idea to revisit your metrics after a certain period of time and revalidate them. E.g. If doing well on Dev set + Single Evaluation Metrics does not reflect doing well on your product, REVISIT and CHANGE the metric as it no longer serves its purpose.</p><p>That’s it, folks! I am not sure if I did justice to whatever intuition I gained, and I won’t know unless you share feedback.</p><p>Together we learn, share and discover.</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2017/10/12/re-structuring-machine-learning-execution/"><em>ronakbits.tech</em></a><em> on October 12, 2017.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=772883823020" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Before starting Andrew Ng’s ML Course…]]></title>
            <link>https://medium.com/@ronakagrawal2006/before-starting-andrew-ngs-ml-course-767b6f8cf06c?source=rss-19c9bd324a60------2</link>
            <guid isPermaLink="false">https://medium.com/p/767b6f8cf06c</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[tips]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Ronak Agrawal]]></dc:creator>
            <pubDate>Wed, 11 Oct 2017 14:37:12 GMT</pubDate>
            <atom:updated>2017-10-23T07:53:25.903Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/460/0*svInfqOR_-ecvUXk." /></figure><p>If you are thinking of starting ML, without a doubt <a href="https://www.coursera.org/learn/machine-learning">Andrew Ng’s Course on Coursera</a>. is the best place to start.</p><p>However, a couple of things below that should ease your journey.</p><ol><li>Make sure you complete at least 4 Weeks. The first 2 assignments are the mountain that you must scale before witnessing the beautiful horizons of ML.</li><li>Do not skip any videos/lectures (I tried to act smart and tried; Let’s just say, not one of my brightest ideas…).</li><li>Do not shy away from watching videos again and again(and again), if required. Everything you need is right there, in the videos.</li><li>Use Emacs or Sublime Text as an editor (I spent quite a bit of time setting up the ‘ideal environment’ only to later use Sublime Text. (In the event that you discover a better alternative, please share.).</li><li>If you think you lack the fundamentals to get this right, NOW is the time to get them (Looking back, Sleeping through that lecture on probability was not cool). The good news is you are smarter now with more resources at your disposal. It only takes 10 minutes to get each of the fundamentals correct.</li><li>It is OKAY to reverse engineer for the first 2 weeks and then try again. There were times when I did not get to the solution directly. I saw the solutions on <a href="https://github.com/merwan/ml-class">Github</a>. With that as a reference, reverse engineered it. (Just make sure you re-implement them later on and are able to explain what has been done.)</li><li>Octave-CLI is your ally, trust it; use it. (You’ll know this once you complete installation in Week 2)</li><li>Get peers!! Get someone to discuss. With so many equations messing around, I can’t stress enough the importance of having someone to discuss them with. Start in pairs (get someone in the same boat as yourself); that should really help reinforce your learning.</li><li>If the speed of videos is slow for you, watch videos at 1.25x or 1.5x.</li><li>Don’t hush that inner voice that whispers, <em>‘You really did not get that, did you?’</em>. Instead, embrace it.</li><li>If you feel lost, you are on the right track(that implies you understand a tiny bit of it and are questioning the rest). The transition from an expert programmer to a grad student surely takes a toll.</li><li>This is the most fundamental course there is which completely covers basics of ML (AFAIK). So, yes, you’ll have to get through it.</li><li>Discuss, Explain and Talk ML. Nothing will concretize your understanding other than explaining it to someone else.</li><li>If at some point while watching lectures, you feel that you are not following, PAUSE the video right there, go back and start over. Don’t, seriously DON’T, be in a hurry to finish the video up.<strong> Take your time and really understand what’s going on in there.</strong> Or it will come back and bite you later.</li><li>Take pen and paper! Solve the algorithm with a very small dataset manually on paper. It really helps in understanding whats happening to the data and how the algorithm is working.</li></ol><p>That’s it, folks.</p><p>Happy Machine Learning :).</p><p>[Image subject to copyright by Coursera.]</p><p><em>Originally published at </em><a href="https://ronakbits.tech/2017/10/11/before-starting-andrew-ngs-ml-course/"><em>ronakbits.tech</em></a><em> on October 11, 2017.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=767b6f8cf06c" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>