<?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[EKbana - Medium]]></title>
        <description><![CDATA[EKbana&#39;s blog spot for our latest works, our developer showcases and Office Culture. Ekbana.com - Medium]]></description>
        <link>https://medium.com/ekbana?source=rss----8a692552e928---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>EKbana - Medium</title>
            <link>https://medium.com/ekbana?source=rss----8a692552e928---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 11:58:26 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/ekbana" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Audio Signal Processing]]></title>
            <link>https://medium.com/ekbana/audio-signal-processing-f7e86d415489?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/f7e86d415489</guid>
            <category><![CDATA[amplitude]]></category>
            <category><![CDATA[audio-signal-processing]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[digital-signal-processing]]></category>
            <category><![CDATA[signal]]></category>
            <dc:creator><![CDATA[Pema Gurung]]></dc:creator>
            <pubDate>Wed, 29 Apr 2020 09:26:04 GMT</pubDate>
            <atom:updated>2022-03-21T11:02:48.620Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/550/1*HbN4YcNTnlGp9a30cKbfUA.jpeg" /></figure><h4>Overview of Audio Signal Processing for Beginners</h4><p>According to Wikipedia,<strong> “Audio signal processing </strong>and<strong> Digital Signal Processing </strong>is a subfield of <strong>signal processing</strong> that is concerned with the electronic manipulation of <strong>audio signals.” </strong>Now for those who are totally new to Audio Signal Processing, might be wondering what is signal processing and audio signals?</p><h4><strong>Signal Processing</strong></h4><p>Again taking the definition from Wikipedia, “Signal processing is an electrical engineering<strong> </strong>subfield that focuses on analyzing, modifying, and synthesizing signals such as <strong>sound</strong>, <strong>images</strong>, and <strong>biological measurements</strong>.” In short, it turns signals into useful information.</p><h4><strong>Audio Signals</strong></h4><p>An audio signal is a representation of <strong>sound</strong>, typically using a level of electrical voltage for <strong>analog signals</strong>, and a series of binary numbers for <strong>digital signals</strong>. Audio signals have frequencies in the audio frequency range of roughly 20 to 20,000 Hz, which corresponds to the lower and upper limits of human hearing.[2] There are two types of <strong>Audio</strong> signals, <strong>Digital</strong> and <strong>Analog. </strong>Difference between them? In simple form explained by <strong>Michael Bauers:</strong> “<em>Digital means that that the source of the audio is a string of numbers called samples. There’s a set number of samples per second. Each sample has what’s called a bit depth, which is simply the number of bits. For example, the digital audio could be something like 16-bit audio, at 44.1 kHz ( 44,100 samples per second.)” and “Analog generally implies that the source audio is analog. It’s a continuously varying signal. For example, a vinyl record is analog.” [10]</em></p><h4><strong>Digital Signal Processors (DSP)</strong></h4><p>Digital Signal Processors (DSP) take real-world signals like voice, audio, video, temperature, pressure, or position that have been digitized and then mathematically manipulate them. Signals need to be processed so that the information that they contain can be displayed, analyzed, or converted to another type of signal that may be of use.[8]</p><p><strong>Now let’s jump into sound since audio signals are the representation of a sound!</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/308/1*WgbJPKXKm6wdVxDO8KlRhQ.png" /></figure><p>Sound is a vibration that travels through the air or another medium and can be heard when they reach a person&#39;s ear. All vibrations, including sound, have a frequency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/798/1*IAMGwFGyzugUYR_ubYRU4A.png" /><figcaption>The human ear can hear between frequencies of about 20 Hz to 20,000 Hz</figcaption></figure><p>To work with the sound, you should know about a <strong>wave</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/459/1*A-SXZCamTDC1sjtGvPbPxg.png" /><figcaption>wave</figcaption></figure><p>A <strong>wave</strong> is a movement in time in a medium. The energy is transferred due to this movement in the medium. In such a case, it is called a wave. For eg: Electromagnetic waves, light waves. While a <strong>signal</strong> is the one that carries information in the wave. the basic function of a sine wave is</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/303/1*oh1kmOlh58OuZuA7XNXTeg.png" /><figcaption>Source: Wikipedia</figcaption></figure><p>where A is the amplitude, f is the Frequency, w is (2*pi*f), and phase is often used to shift the starting point of the waves.</p><p>In python, it can be written as</p><pre>y = amplitude*np.sin( 2 * np.pi *frequency * t)</pre><p><strong>Plotting a simple sine wave</strong> by applying the sine wave function to understand a wave better.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*o-zgJqyiTOIfOK_gCZHfEA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/410/1*FndLlzL9XsHjnDNtTMn5xw.png" /></figure><p>As you can see, it has created one cycle withing the given audio length.</p><h4><strong>Amplitude</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/406/1*Tib38RkYie8UhFCxvQIuNA.png" /></figure><p>The <strong>amplitude of a wave</strong> refers to the maximum amount of displacement of a particle on the medium from its rest position. In a sense, the <strong>amplitude</strong> is the distance from rest to crest. Similarly, the <strong>amplitude</strong> can be measured from the rest position to the trough position. [9] (A <strong>crest</strong> is a point on a wave with the maximum value upward displacement within a cycle. A <strong>trough</strong> is a minimum or lowest point in a cycle. [Reffer to the Wave image for crest and trough])</p><p><strong>Plotting a simple sine wave </strong>with different amplitude for comparison</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/462/1*3uRxTWk0lz5gBGtQVUDsUg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/377/1*iYQzBrmgHdiRaPOo-PhvQQ.png" /><figcaption>with amplitude = 10</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/395/1*YSc8-kfjgPxEovoZXjk4FQ.png" /><figcaption>with amplitude = 20</figcaption></figure><p>as you can see that the <strong>crest and trough </strong>are 10, -10 respectively when the amplitude is given as 10 and similarly for Amplitude=20, <strong>crest and trough </strong>are 20,-20 respectively.</p><h4><strong>Frequency</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/585/1*SKchKo1UZPIUMIW9fP3LmQ.png" /></figure><p>Frequency is a measure of the number of wave crests that pass a fixed point per second. At higher frequencies, more wave crests will pass the point each second; for this to happen, the wavelengths must be short. As frequency decreases, fewer waves will pass the fixed point per second, so wavelengths will be relatively longer.</p><p><strong>Plotting a simple sine wave </strong>with a different frequency to compare</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*79ukcAsjXA6FesElGoE_eg.png" /><figcaption>with Frequency = 10</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/557/1*KkPLSsegDZfNMmGiGrWzDQ.png" /><figcaption>with Frequency = 20</figcaption></figure><p>As you can see that the number of frequencies that we give, it generates that number of cycles in the given length. Like when the frequency is 10, there are 10 cycles in the given time and similarly, when the frequency is 20, there are 20 cycles in one second timeline. To plot and listen to it, you can check out the <a href="https://github.com/ekbanasolutions/AudioAnalysis/blob/master/Audio%20Basics.ipynb">notebook</a></p><h4><strong>Sampling Rate</strong></h4><p>In a Sound, the <strong>sampling rate</strong> (or sample frequency) is the number of samples per second. A sampling rate of 44,1 kHz (44.100 Hertz) means that the audio signal has been sampled with 44100 samples per second and if it&#39;s for 20 seconds, it will 44100 *20 = 882000 samples.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*kRjYDQfoQV1hqCUAZT4ZwQ.png" /><figcaption>sinusoidal wave (in black) and red lines represent measurements that have been done to sample that audio signal</figcaption></figure><p><strong>Plotting a simple sine wave</strong> with different sample rate to compare</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/436/1*B1O33n7sVlb70k9SWQ-FbA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/419/1*4qGVlQGxT9u8lf87vOi3DA.png" /><figcaption>sample rate = 50</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/407/1*M0h-8acEdWMjSMn9If__Fg.png" /><figcaption>sample rate = 100</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/429/1*sEfwtIBE3wzwYMItPmDP_g.png" /><figcaption>sample rate = 40</figcaption></figure><p>Along with the various sample rate plotting, you can see that it generates the number of samples for one second. Like for sample rate = 40, there are 40 samples taken from the one-second audio.</p><p>👏 if you liked my First Article on Audio Processing. 😄</p><p><strong>Notebook link for the code and execution can be found </strong><a href="https://github.com/ekbanasolutions/AudioBasic/blob/master/Audio%20Basics.ipynb"><strong>here</strong></a></p><p>And also two Amazing blogs that you can read</p><p>[1] <a href="https://medium.com/@stephenhartzell/digital-signal-processing-in-one-lesson-a2f4f64305bf">https://medium.com/@stephenhartzell/digital-signal-processing-in-one-lesson-a2f4f64305bf</a></p><p>[2] <a href="https://towardsdatascience.com/understanding-audio-data-fourier-transform-fft-spectrogram-and-speech-recognition-a4072d228520">https://towardsdatascience.com/understanding-audio-data-fourier-transform-fft-spectrogram-and-speech-recognition-a4072d228520</a></p><p>Other References:</p><p>[1] <a href="https://en.wikipedia.org/wiki/Sine_wave">https://en.wikipedia.org/wiki/Sine_wave</a></p><p>[2] <a href="https://en.wikipedia.org/wiki/Audio_signal">https://en.wikipedia.org/wiki/Audio_signal</a></p><p>[2] <a href="https://en.wikipedia.org/wiki/Audio_signal_processing">https://en.wikipedia.org/wiki/Audio_signal_processing</a></p><p>[3] <a href="https://kids.britannica.com/students/assembly/view/223513">https://kids.britannica.com/students/assembly/view/223513</a></p><p>[4] <a href="https://www.mixcorner.com/What-is-Sample-Rate-in-Audio?lang=fr">https://www.mixcorner.com/What-is-Sample-Rate-in-Audio?lang=fr</a></p><p>[5] <a href="https://blogs.arubanetworks.com/industries/frequency-cycle-wavelength-amplitude-and-phase/">https://blogs.arubanetworks.com/industries/frequency-cycle-wavelength-amplitude-and-phase/</a></p><p>[6] <a href="https://en.wikipedia.org/wiki/Audio_signal_processing">https://en.wikipedia.org/wiki/Audio_signal_processing</a></p><p>[7] <a href="https://en.wikipedia.org/wiki/Signal_processing">https://en.wikipedia.org/wiki/Signal_processing</a></p><p>[8] <a href="https://www.analog.com/en/design-center/landing-pages/001/beginners-guide-to-dsp.html">https://www.analog.com/en/design-center/landing-pages/001/beginners-guide-to-dsp.html</a></p><p>[9] <a href="https://www.physicsclassroom.com/class/waves/Lesson-2/The-Anatomy-of-a-Wave">https://www.physicsclassroom.com/class/waves/Lesson-2/The-Anatomy-of-a-Wave</a></p><p>[10] <a href="https://www.quora.com/What-are-the-differences-between-an-analog-and-digital-audio">https://www.quora.com/What-are-the-differences-between-an-analog-and-digital-audio</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f7e86d415489" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/audio-signal-processing-f7e86d415489">Audio Signal Processing</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Planar and Spherical Projections of a Point Cloud]]></title>
            <link>https://medium.com/ekbana/planar-and-spherical-projections-of-a-point-cloud-d796db76563e?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/d796db76563e</guid>
            <category><![CDATA[open3d]]></category>
            <category><![CDATA[python3]]></category>
            <category><![CDATA[numpy]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[geometry]]></category>
            <dc:creator><![CDATA[Aashutosh Pyakurel]]></dc:creator>
            <pubDate>Wed, 29 Apr 2020 09:22:55 GMT</pubDate>
            <atom:updated>2020-04-29T12:30:55.079Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iwcIa-MHN6mBTG3cZCA-AA.png" /></figure><h3>Planar and Spherical Projections of a Point Cloud (Using Open3D)</h3><p>April 29, 2020</p><p>A point cloud is a collection of points in 3-dimensional coordinate space represented by x, y, and z axes. These points may just form a geometric shape or more complex structures such as a 3-dimensional representation of the real world. Various libraries, even with open-sources, such as the <a href="http://pointclouds.org/">Point Cloud Library (PCL)</a> and Intel’s <a href="http://www.open3d.org/">Open3D</a>, provide native datatype for a point cloud. These libraries not only allow simple geometric transformations of the cloud but also provide sophisticated functionalities such as the reconstruction of a 3D scene from various images.</p><p>However, this article discusses a simple geometric approach to achieve the reverse of 3D reconstruction, i.e generate an image from a point cloud. We will also see how to project the point cloud onto other non-planar geometries such as a sphere. Hence, our objective is the following</p><ul><li>Project a point cloud from a certain perspective to a given plane, then store the projection as an image; and</li><li>Project the point cloud onto the surface of a given sphere.</li></ul><p>The sample implementation included in this article was run in python3.6. The following additional libraries for python must be available</p><ul><li>Open3D (tested with v0.9)</li><li>NumPy</li></ul><h3>Projection onto a Plane</h3><p>We will try to achieve the perspective projection of a point cloud (i.e. from a fixed optical center) onto a plane given by the general equation:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/171/1*GCsuWc8BsFSNLCn61Pq00g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/0*aPaEY4MRpiRzyosA.png" /></figure><p>The above demonstration shows the perspective transform of a set of points onto the plane P from the optical center of the camera, with arbitrary co-ordinates (x₀, y₀, z₀). We can see that the projection (x, y, z)of a point (x₁, y₁, z₁) on the plane P is nothing but the point of intersection between the said plane and a line L passing through the optical center and the given point in the cloud.</p><p>The line L in 3-dimensional space can be represented as follows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/83/1*T0ibRK5PRm_g3Se1Rtr6iQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/79/1*BNyz1gt4TNLEBXN76UGrZQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/78/1*dEPieoP7O5rBXJnGQzuEnQ.png" /></figure><p>where a, b and c are the direction cosines of the line L, and t is known as the direction ratio.</p><p>With the given optical center and the set of points in the cloud, we can compute the direction cosines for the lines passing through each of the points.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/357/1*OwpZZ4FJkEdAs6y4i_vmGg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/355/1*kq_DbFMHTWV2xt49FJU8MQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/354/1*9XkmaTiKNjAzSWj01AsY3g.png" /></figure><p>We can substitute the equations of the line in the equation of the plane P since we are solving for the point of intersection between the two.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/330/1*xl0n1dqEmvpC8k6QHWgPhw.png" /></figure><p>Solving for t, we can deduce the above equation as follows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/202/1*aYAJQXf2sckiY59s5T9rwg.png" /></figure><p>Finally, we can obtain the projection by substituting the value of t into the equations of the line L.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/254/1*d7OlJ73JyzqRYl_DZMVcXQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/1*2FlW-mVIp-VD0TIzFIIHIQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/250/1*gwuRjfenhW0v9qRI4lFBmA.png" /></figure><p>The above computation should be performed for each point in the Point Cloud to obtain the complete projection on the given plane.</p><h3>Python Implementation</h3><p>The following function takes an Open3D PointCloud, equation of a plane (A, B, C, and D) and the optical center and returns a planar Open3D PointCloud Geometry.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/909827766900175a5049bda765deb659/href">https://medium.com/media/909827766900175a5049bda765deb659/href</a></iframe><h3>Output</h3><p>The resulting planar Point Cloud can be seen as follows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/945/0*_xtyO-7LfDFGGD-8.png" /></figure><h3>Projection onto a Sphere</h3><p>A Point Cloud maybe projected onto geometrical surfaces other than a plane, such as a sphere. The spherical projection in particular maybe used to approximate a 360 image of the Point Cloud since those are spherical or cylindrical photographs of a scene.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/615/0*t4Ewaio1S25qW3JW.png" /></figure><p>The figure above shows the spherical projection of a point onto the surface of a sphere with radius r₁ and with its center at the point(x₀, y₀ , z₀). Such a sphere can be represented by the following equation</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/269/1*NVFF9pgg6cjBR8KrjguKeg.png" /></figure><p>Similar to the projection on a plane, the spherical projection of a point is the point of intersection between the given sphere S and the line passing through the center of the sphere and the point.</p><p>Again, the line L in 3-dimensional space can be represented as follows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/83/1*T0ibRK5PRm_g3Se1Rtr6iQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/79/1*BNyz1gt4TNLEBXN76UGrZQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/78/1*dEPieoP7O5rBXJnGQzuEnQ.png" /></figure><p>where a, b and c are the direction cosines of the line, and t is the direction ratio. With the given center of the sphere and the set of points in the cloud, we can compute the direction cosines for the lines passing through each of the points and the center.</p><p>Here,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/305/1*IUtLeDK5crDTDBMZZKURAw.png" /></figure><p>is the distance between the center of the sphere and the point in the Point Cloud.</p><p>Now, we can substitute the equations of the line into the equations of the sphere to solve for t.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/390/1*uCPsRmnJIhp1vZlTbmz2cw.png" /></figure><p>Solving this equation for t, we obtain the following relation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/76/1*GVPFcxoo0XTpYrDy08Ys5w.png" /></figure><p>Finally, substituting this value of t and the direction cosines into the equations of the line, we can obtain the projected points.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/259/1*4qrmNujkMR29KCuMQ_yKPA.png" /></figure><p>or,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/151/1*OatLmkGWs6DolIqCJSVT7g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/147/1*OkCPUtg_KdDG1wGWZf-cEg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/145/1*CufznVIAvdOutGuoj9jPRQ.png" /></figure><h3>Python Implementation</h3><p>The following python method takes an Open3D PointCloud geometry, and the radius and center of the sphere to project on.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/f368aba4487833abdf164f6e5b3594a1/href">https://medium.com/media/f368aba4487833abdf164f6e5b3594a1/href</a></iframe><h3>Output</h3><p>The resulting spherical projection of the cloud is as follows</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8XAID3-BBep124kp.png" /></figure><h3>References</h3><p>The point of intersection between a line and a plane. <a href="http://www.ambrsoft.com/TrigoCalc/Plan3D/PlaneLineIntersection_.htm">http://www.ambrsoft.com/TrigoCalc/Plan3D/PlaneLineIntersection_.htm</a></p><p>Aashutosh is an engineer currently researching and engineering at Ekbana.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d796db76563e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/planar-and-spherical-projections-of-a-point-cloud-d796db76563e">Planar and Spherical Projections of a Point Cloud</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Write a python binding, for your C++ code, using pybind11 library]]></title>
            <link>https://medium.com/ekbana/write-a-python-binding-for-your-c-code-using-pybind11-library-ef0992d4b68?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/ef0992d4b68</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[dev-talk]]></category>
            <category><![CDATA[cpp]]></category>
            <category><![CDATA[pybind11]]></category>
            <dc:creator><![CDATA[Aashutosh Pyakurel]]></dc:creator>
            <pubDate>Fri, 03 Apr 2020 10:08:45 GMT</pubDate>
            <atom:updated>2020-04-04T12:39:12.313Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LexbvyS-6QUjrZJQJ-jkEw.png" /></figure><p>Programmers have increasingly inclined towards using python for various complex applications over the years. The easy to understand code, and the wide adaptation of various libraries makes it convenient for both new as well as established developers to easily build the program of their will. However, the convenience of coding with high-level language such as python comes at the cost of performance degradation, this is because the more human readable that a programming language is the more computational cost is required to convert it to machine readable format. This is why heavy libraries such as NumPy, although written for python, are actually written in C++ with access in python with the help of bindings, hence providing structured and quick computation at low-level yet highly convenient to the user.</p><p>This post is intended to be a beginner’s guide to writing their own python binding using pybind11 library, which maps various core C++ functionalities to python, some of which will be discussed and used in this post. We will build our own python binding for a matrix class that provides various methods for construction and operations.</p><h3>Step 1: Build the pybind11 library</h3><p>The building process discussed here was tested in Ubuntu 18.04.4 LTS Operating System.</p><h3>Dependencies</h3><p>The following dependencies must be met before building the pybind11 library.</p><ul><li>Clang / LLVM Compiler with C++11 support, GCC 4.8 or newer</li><li>Python3, python3-dev (tested with v3.6.9)</li><li>Cmake (tested with v3.13)</li></ul><h3>Build from Source</h3><p>Pybind11 source can be cloned from the repository.</p><pre>$ git clone <a href="https://github.com/pybind/pybind11">https://github.com/pybind/pybind11</a></pre><p>Then the following code can be executed to build and install the pybind11 library.</p><pre>$ cd pybind11<br>$ mkdir build<br>$ cd build<br>$ cmake ..<br>$ make -j4<br>$ sudo make install</pre><p>Note: The make install command copies the pybind11 library to a global system location which has to be added as a environment variable. This can be avoided by adding pybind11 library as sub-module in required projects.</p><h3>Step 2: Write the code for Matrix class and its python binding</h3><p>We are using a simple class named Matrix with multiple constructors and multiple overloaded operators to demonstrate python binding. The actual binding code is included in a separate header file named “mainpy.cpp”, this file is used to generate the module that can be imported into python to use the bindings created using pybind11.</p><h3>Matrix implementation subject to binding</h3><p>The C++ Matrix class has basic constructors and overloaded operators declared in “matrix.hpp” as follows</p><pre>#include &lt;iostream&gt;<br>#include &lt;vector&gt;<br>class Matrix{<br>	int nr;<br>	int nc;<br>	public:<br>	enum INIT_PARAM{ONES = 1,ZEROS = 0};<br>	std::vector&lt;std::vector&lt;int&gt;&gt; mat;<br>	Matrix();<br>	Matrix(int r, int c, INIT_PARAM ini = ZEROS);<br>	void setValues(std::vector&lt;int&gt; l);<br>	Matrix operator +(const Matrix&amp;) const;<br>	Matrix operator -( const Matrix&amp;)const;<br>	Matrix operator *( const Matrix &amp;mat1)const;<br>	int rows() const;<br>	int cols() const;<br>	void print();<br>};</pre><p>We are also using a custom exception class to handle errors in matrix operations. It is declared in “matexp.hpp” as follows</p><pre>#include &lt;exception&gt;</pre><pre>class MatExp: public std::exception{<br>	public:<br>	virtual const char* what() const throw();<br>};</pre><h3>Python Binding Implementation</h3><p>We implement python binding by using pybind11 header-only library to map each constructor, method and operation to python. This has been performed in the code as follows, included in “mainpy.cpp” file.</p><pre>#include &quot;matrix.hpp&quot;<br>#include &quot;matexp.hpp&quot;<br>#include &lt;pybind11/pybind11.h&gt;<br>#include &lt;pybind11/operators.h&gt;<br>namespace py = pybind11;<br>PYBIND11_MODULE(matrix, m){<br>	py::class_&lt;Matrix&gt;(m, &quot;Matrix&quot;)<br>		.def(py::init([] () {return new Matrix(); }))<br>		.def(py::init([] (int r, int c) {return new Matrix(r,c); }))<br>		.def(py::init([] (int r, int c, Matrix::INIT_PARAM ini){ return new Matrix(r,c,ini); }))<br>		.def(py::init([] (int r, int c, py::list l){<br>			Matrix t(r,c);<br>			std::vector&lt;int&gt; temp;<br>			for (py::handle i : l){<br>				temp.push_back(i.cast&lt;int&gt;());<br>			}<br>			t.setValues(temp);<br>			return t;}))<br>		.def(py::self + py::self)<br>		.def(py::self - py::self)<br>		.def(py::self * py::self)<br>		.def(&quot;rows&quot;, &amp;Matrix::rows)<br>		.def(&quot;cols&quot;,&amp;Matrix::cols)<br>		.def(&quot;print&quot;,&amp;Matrix::print);<br>	<br>	py::enum_&lt;Matrix::INIT_PARAM&gt; (m, &quot;INIT_PARAM&quot;)<br>		.value(&quot;ZEROS&quot;,Matrix::INIT_PARAM::ZEROS)<br>		.value(&quot;ONES&quot;,Matrix::INIT_PARAM::ONES)<br>		.export_values();</pre><pre>	py::exception&lt;MatExp&gt; exc(m, &quot;MatrixError&quot;);<br>}</pre><p>This code includes the Matrix and exception (MatExp) classes declared in “matrix.hpp” and “matexp.hpp” files respectively. The pre-defined pybind11 types such as class_, init (constructor) and enum_ are defined in “pybind11” namespace (accessed as “py”) in “pybind/pybind.h” header file. Similarly, operators such as “self”, used for accessing an object implicitly for purpose of operator overloading, are declared in the same “pybind11” namespace in “pybind/operators.h” header file.</p><p>Module</p><p>The module can be named by using PYBIND11_MODULE(matrix,m){} wrapper, where the first argument “matrix” is the name to be used while importing the module in python, and the second argument “m” is the name used to access the module within this wrapper. All the classes, variables, methods, enumerations, exceptions and other types to use binding must be defined within this wrapper.</p><p>Class</p><p>The method class&lt;typename&gt;(access_name, method_class_name) can be used to create binding for a class. Every member of the class that requires binding has to be mentioned by using .def() attribute of this method.</p><p>Constructors</p><p>The constructors can be mapped into init() method of pybind11 within .def() attribute of the class to be constructed, as shown in the code. init() method may call any constructor defined by the given class.</p><p>Overloading</p><p>Operator overloading can be bound by defining the operation on implicit operator self, used to access the current class object, in pybind11 namespace.</p><p>Other Methods</p><p>All other methods can be mapped by using .def() attribute of the class by passing two arguments: name of the method to be used for calling in python and the reference to the function to be called (using &amp; operator).</p><p>Enumerations</p><p>The enumerations of a class can be provided python binding in a module by calling enum_&lt;typename&gt;(reference_names) method and its attribute value() to define each value to be wrapped using the enumeration. The values must be exported using export_values() attribute of the enum_ method.</p><p>Exceptions</p><p>The standard (built-in) and custom exceptions in C++ implementation can be mapped into python by using pybind11::exception&lt;typename&gt; exc(reference_names) method. A custom translator can also be registered for by calling the following wrapper.</p><pre>py::register_exception_translator([] (std::exception\_ptr p){}</pre><p>Python Objects (PyObject) and Handle</p><p>Sometimes python objects such as list and dict defined in python might be needed to be accessed in C++, these can be translated into PyObject and its inherited classes defined by cpython. pybind11 also supports the following python objects (inheriting from PyObject class) that can be handled from C++</p><pre>bool_, buffer, bytes, capsule, dict, dtype, exception&lt;type&gt;,<br>float_, function, generic_type, int_, iterable, iterator, list,<br>memoryview, module, none, sequence, set, slice, staticmethod, str,<br>tuple, weakref</pre><p>These objects may be used to take inputs from a python object or modify the python object from within C++.</p><p>The python object (such as pybind11::list as shown in example) can be iterated by using Handle class that can be cast into desired C++ type using cast&lt;typename&gt;() method.</p><h3>Step 3: Building the Binding</h3><p>The python module created by using pybind11 library can be built manually or by using Cmake.</p><h3>Using CMake</h3><p>The following “CMakeLists.txt” is used to build the pybind11 module that can be imported into python.</p><pre>cmake_minimum_required(VERSION 3.1.0)<br>project (matrix_operations)<br>find_package(pybind11 REQUIRED)<br>pybind11_add_module(matrix mainpy.cpp matrix.cpp)<br>add_executable (matrixop main.cpp matrix.cpp)<br>pybind11_add_module(module_name SOURCES) is used to build the python<br>module using the sources listed.</pre><p>Note: find_package(pybind11 REQUIRED) may be used only when the pybind11 library has been built globally. If the library is included as a sub-module, add_subdirectory(pybind11) should be used instead.</p><h3>Manual Build</h3><p>The module may also be built manually as follows.</p><pre>$ c++ -O3 -Wall -shared -std=c++11 -undefined dynamic_lookup `python3 -m pybind11 --includes` mainpy.cpp -o  matrix`python3-config --extension-suffix`</pre><h3>Step 4: Output</h3><p>The built module can be imported by invoking python3 in the same directory as the built shared-object (.so) file. The output is as follows</p><pre>aashutosh@aashutosh-msi:~/ekbana/internship/pybind-matrix/build$ python3<br>Python 3.6.9 (default, Nov 7 2019, 10:44:02)<br>[GCC 8.3.0] on linux<br>Type &quot;help&quot;, &quot;copyright&quot;, &quot;credits&quot; or &quot;license&quot; for more information.<br>&gt;&gt;&gt;import matrix<br>&gt;&gt;&gt; m = matrix.Matrix(3,2,\[1,2,3,4,5,6\])<br>&gt;&gt;&gt; n = matrix.Matrix(2,3,\[1,-1,-1,1,1,-1\])<br>&gt;&gt;&gt; o = m\*n<br>&gt;&gt;&gt; o.print()<br>3 1 -3<br>7 1 -7<br>11 1 -11<br>&gt;&gt;&gt;</pre><p>Looks Good!</p><p>Further documentation and references can be found here <a href="https://pybind11.readthedocs.io/en/stable/intro.html"><em>https://pybind11.readthedocs.io/en/stable/intro.html</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ef0992d4b68" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/write-a-python-binding-for-your-c-code-using-pybind11-library-ef0992d4b68">Write a python binding, for your C++ code, using pybind11 library</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ekbana Sports Day 2019]]></title>
            <link>https://medium.com/ekbana/ekbana-sports-day-2019-2777af308dc2?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/2777af308dc2</guid>
            <category><![CDATA[office-culture]]></category>
            <category><![CDATA[it-companies]]></category>
            <category><![CDATA[life-at-ekbana]]></category>
            <category><![CDATA[futsal]]></category>
            <category><![CDATA[nepal]]></category>
            <dc:creator><![CDATA[EKbana]]></dc:creator>
            <pubDate>Mon, 30 Dec 2019 11:10:50 GMT</pubDate>
            <atom:updated>2019-12-30T11:10:50.044Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*vRE6tv1F8gqqWJFzpR9oJA.jpeg" /></figure><h4>30th Dec 2019</h4><p>Another year, another sports day for us EKbanians. With the growth of our company and the increase of our team members, we saw a larger sports day with 8 premade teams competing for sports day gold.</p><p>The games and sports competition we saw throughout the event were as follows, Futsal, FIFA, Table Tennis, the Three-Legged race, a Spoon race, Sack race, Thigh Balloon blast, a Tug of War and a Food eating challenge.</p><p>The event took place on the 27th of December and carried on to the 28th, at 7 am the 8 teams were raring to go at the futsal ground as the competition started with the Futsal competition. Tough competition and 8 matches fulfilled, the early group stages of the futsal tournament were completed as the rest of the day taking place at the company grounds. Breakfast and lunch were provided for the competitors by Ekbana. The men and women Table Tennis matches went underway and so did the FIFA tournament. Tough competition was seen as all teams gave it their all and the winners emerged for the two competitions.</p><p>This was followed by the food eating challenge where the competitors had to finish a 1.2kg burger. The winner was crowned and the next competition on the agenda was the thigh balloon blast where a team of three took part. A pair of them giving their attention to blowing up the balloons with the power of their lungs, while one was designated to blast the balloons using their thighs. The competition was fierce yet fun, the winners emerged and the day’s events were completed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*c7Pk0P3kq1uHwEXmeX9aYg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*lau-StUgs0mWRmL2QpU1gg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*wSUk8d2QOrkqSgwJuvy6Uw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ElKpfIOhBJ4m1F-cKrxH_A.jpeg" /></figure><p>The 2nd day took place at 9 am sharp at the futsal ground where the competition started with the ongoing futsal competition for both men and women. Futsal was halted before the finals and the other competition took place. The spoon race was the first in line as both men and women competed for their respective teams. The event continued to the Tug of war competition huffing and heaving away we saw the emergence of the strongest teams of the day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ohP7rdR8Tdhf03uayD0VOw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4kKF7L-61YFiCUXej1gCAQ.jpeg" /></figure><p>Continued with the Three-legged race for both men and women, a fill the glass competition where you fill an empty glass of water while transferring water with a spoon over a distance, a Sack race for both men and women followed by the futsal competition finals. The day finally over, the tired and haggard-looking ladies and gentlemen of Ekbana headed for a much-needed intake of food at a restaurant venue hosted by Ekbana.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oT_B7UX3AtYS-zmlxK4gxA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yezzJvi4B7VmRWYrVvcUNQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FmY6oQu6JG9n9FDShPCjtA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-aZVhx_rZX5AyY5OAvxJvg.jpeg" /></figure><p>Before the meal was had, the prizes were distributed to the respective winners and runners up of the individual competitions. With a surprise overall winners being tied for 1st place, Team Niruta and Team Danny, Congratulations! The medals and cups were distributed and the mood was set for a hearty meal with friends and colleagues.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*FZa8QKfUQzmQx8xnR2cpeQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*9IH9WMWhZfthDhF8iTGJCg.jpeg" /><figcaption>The winners!</figcaption></figure><p>The spirit of competition and teamwork defines our company. Every year we do this and every year we see the bonding of teams like never before. We as Ekbanians feel proud of days like this and the environment and friendships that it creates. Thank you to all the wonderful organizers of the event, it was a huge hit and as always, See you next year!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2777af308dc2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/ekbana-sports-day-2019-2777af308dc2">Ekbana Sports Day 2019</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Customer Churn Prediction]]></title>
            <link>https://medium.com/ekbana/customer-churn-prediction-4094fc24656b?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/4094fc24656b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[predictive-algorithms]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[customer-churn]]></category>
            <category><![CDATA[ektech]]></category>
            <dc:creator><![CDATA[EKbana]]></dc:creator>
            <pubDate>Thu, 14 Nov 2019 08:39:22 GMT</pubDate>
            <atom:updated>2019-11-14T08:39:21.951Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*d58iZ6esNNcfntQ7" /></figure><h4>14 November 2019</h4><h3>Customer Churn</h3><p>Customer churn refers to when a customer (player, subscriber, user, etc.) ceases his or her relationship with a company. Online businesses typically treat a customer as churned once a particular amount of time has elapsed since the customer’s last interaction with the site or service. The full cost of customer churn includes both lost revenue and the marketing costs involved with replacing those customers with new ones. Reducing customer churn is a key business goal of every online business.</p><h3>The Importance of Predicting Customer Churn</h3><p>The ability to predict that a particular customer is at a high risk of churning, while there is still time to do something about it, represents a huge additional potential revenue source for every online business. Besides the direct loss of revenue that results from a customer abandoning the business, the costs of initially acquiring that customer may not have already been covered by the customer’s spending to date. (In other words, acquiring that customer may have actually been a losing investment.) Furthermore, it is always more difficult and expensive to acquire a new customer than it is to retain current paying customers.</p><h3>Steps of Churn Prediction</h3><ul><li>Data processing</li><li>Feature selection &amp; engineering</li><li>Modelling</li><li>Insights and Actions</li></ul><h3>Data Processing</h3><p>The data were taken from 2 years of transaction data from every store. This data was the transaction details of customers in-store at a specific time. The customers’ data were the transaction data with bill date and other transaction details. The train set and validation set data were taken from the overall customer data of date 2019–02–01 and the test set was taken from the overall customer data of date 2019–03–01.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/534/0*1lkSmYhJeYfAzKwC" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/746/0*Ey_adxra3uw8iR6u" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/449/0*YgkeMWMbD7Ikp-M6" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/429/0*eOvEwjY9hYwKrdbY" /></figure><p>Figure a Figure b</p><p>The above ‘Figure-a’ is the amount of churn and non-churn customers in February 2019. ‘Figure-b’ is the amount of churned customers in February (‘Feb’) who were also new customers joined from February 2019. ‘Past’ is the number of churned customers who were not new customers joined from February.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/580/0*OMtjVHyG1wUBD4G8" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/527/0*SVcM3Q1fF3baJsI_" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/539/0*c6PcbarjUMaAIcUD" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/574/0*QajYYcNzaPyyOHp1" /></figure><p>The figures above are the analysis and visualization of the dataset and features in 3D space.</p><h3>Feature Selection &amp; Engineering</h3><p>Features were more complicated to extract from this dataset. The features that were extracted were the past transaction behaviour of a specific customer.</p><p><strong>Features 1</strong>:</p><ul><li>7 months attendance of customer</li><li>Origin (Value of started time of customer started to)</li><li>Visit ( Average visit per month)</li></ul><p><strong>Features 2</strong>:</p><ul><li>Average Basket</li><li>Visit</li><li>Weekday</li><li>Weekend</li><li>Five Hundred</li><li>Thousand</li><li>Fifteen hundred</li><li>Three Thousand</li><li>Five Thousand</li><li>Ten Thousand</li><li>Above</li></ul><h4>Correlation matrix:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/642/0*Yu1T9z5t4HsGGaRJ" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/625/0*GZOM28-DOT4nx2es" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/565/0*g_OM3gITkYLAcTwH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/981/0*474UL9WqxiJHwxfh" /></figure><h4>Feature Importance:</h4><p>Feature importance using random forest</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/802/0*DXXyTnBibDVTJL9F" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/475/0*O64YkGl-nKhdScuD" /></figure><h3>Modelling</h3><p><strong>Train Set</strong>:</p><p>Active customers in February month were taken as train set where the new customers who also were churned customers were excluded from the training data. Those types of customers were excluded because they have very fewer data and such data can affect our predictive model. Among the remaining customers, 2500 were churned customers and 3000 non-churned customers were taken as the training set for the predictive model.</p><p><strong>Test Set</strong>:</p><p>Active customers in March month were taken as test set where the new customers who also were churned customers were excluded from the test data. Among the remaining customers 43,000 customers were taken as test set where 3300 were Churned customers and 39,000 non-Churned customers.</p><h4>Model Building</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/606/0*JN3JDvK4vQPHW6Lv" /></figure><h3>Result</h3><p>The test accuracy of the predictive model was around 90%. Where 648 churned customers were wrongly predicted as non-churned customers and 3877 non-churned customers were wrongly predicted as churned customers. Where 39225 non-churned customers were correctly predicted as non-churned and 2672 churned customers were correctly predicted as churned customers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/599/0*e263Q2VaGEXlQ541" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/483/0*ORaSNnKf55YGVCZh" /></figure><h3>Conclusion</h3><p>Churn Prediction was done using a neural network (Sequential model). The dataset creation and feature engineering were the complicated part of this project. High accuracy cannot be achieved simply in this type of dataset. Training the previous month dataset now we can predict the next month’s customers’ status in churn prediction.</p><h3>Future Work</h3><ul><li>Analysis of the new customers who are going to churn from next month.</li><li>Finding other factors (features) that can help our model to give better accuracy.</li></ul><h3>References</h3><p><a href="https://towardsdatascience.com/explaining-feature-importance-by-example-of-a-random-forest-d9166011959e">https://towardsdatascience.com/explaining-feature-importance-by-example-of-a-random-forest-d9166011959e</a></p><p><a href="https://www.optimove.com/resources/learning-center/customer-churn-prediction-and-prevention">https://www.optimove.com/resources/learning-center/customer-churn-prediction-and-prevention</a></p><p><a href="https://plot.ly/python/3d-scatter-plots/">https://plot.ly/python/3d-scatter-plots/</a></p><p>This article was written by <a href="https://www.linkedin.com/in/ramesh-rawal-877848a8/">Ramesh Rawal</a>, Machine Learning Engineer at Ekbana.</p><p>Original Article: <a href="https://medium.com/@rawalrameshr2/churn-prediction-f21a0e1c198f">https://medium.com/@rawalrameshr2/churn-prediction-f21a0e1c198f</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4094fc24656b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/customer-churn-prediction-4094fc24656b">Customer Churn Prediction</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nepal Government and Blockchain]]></title>
            <link>https://medium.com/ekbana/nepal-government-and-blockchain-555050cdca32?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/555050cdca32</guid>
            <category><![CDATA[ektech]]></category>
            <category><![CDATA[hyperledger]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[nepal]]></category>
            <category><![CDATA[information-technology]]></category>
            <dc:creator><![CDATA[EKbana]]></dc:creator>
            <pubDate>Thu, 07 Nov 2019 08:04:20 GMT</pubDate>
            <atom:updated>2019-11-07T08:04:20.124Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*rsyPbLIAaYwJrJ6daqlApA.jpeg" /><figcaption>Image by <a href="https://pixabay.com/users/TheDigitalArtist-202249/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=3212312">Pete Linforth</a> from <a href="https://pixabay.com/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=3212312">Pixabay</a></figcaption></figure><h4>7th November 2019</h4><p>It was 1972 when the first computer was brought to Nepal as a lease for a population census calculation. Analyzing history, Nepal has been very slow to adopt digitization in the government sector. Even now many of the government offices are using hardware and software dating back 10–15 year. Sometimes we as citizens have to wait for days to have our work done due to manual work which could be easily automated. Corruption and downright negligence have been some of the main problems in the government sector as there aren’t any liability, transparency and liquidity.</p><p>As a tax-paying citizen, we all have the right to information since our constitution ensures that every individual can freely express ideas, opinions, and thoughts, it should be able to freely formulate them, hence to be very well informed. In the current situation, there is no way for normal citizens like us to keep a check on where and how our tax has been utilized because there are not good records or misplaced somewhere. There have been minor digitization on government sectors but still, many of us think it is not enough. For example, let&#39;s look at our driver license distribution, it will take about a year just to get a smart driver license today.</p><p>Blockchain had been introduced back in 1991. I won’t go into detail on what blockchain is because there is a separate article in which I’ve posted <a href="https://www.linkedin.com/pulse/blockchain-its-introduction-roshan-ranabhat/">here</a> in detail about blockchain. But the fundamental concept of blockchain is shared ledger protected by some sophisticated cryptography. The blockchain consists of the block where any kind of records can be kept as a cryptographic hash. All blocks are linked with each other. Any kind of action to change or tamper with data of one block will cause to invalidate the whole blockchain which will help to keep records immutable. People are mistaking blockchain as cryptocurrency, but cryptocurrency is merely only the implementation of blockchain.</p><p>Blockchain can be the way the government to be fully digitized. Many countries and financial sectors have begun to adopt blockchain in their day to day business. Countries like Dubai, the US, India, China, etc have already started to adopt blockchain technology. By 2020 Dubai is planning to integrate blockchain solutions into all its services like visa applications, renewal of the license, payment and many other services. Though in our government sector security has been a reason that have stood in the way of digitization because since blockchain is shared ledger and bringing data of millions of people onto digital platform may risk hacking of this information, but we need to understand that since the blockchain is virtually unhackable and changing or tampering data cannot be done it may be solution for governments to go fully digital. If every government body is digitized and brought to the blockchain every citizen can monitor those institutions and keep a check on if there have been any fraudulent which will also reduce corruption. We can also view how the budget has been utilized for development in our area. And if we need some kind of permission blockchain for the privacy of our data, there are many other implementations like Hyperledger Composer and Corda which gives an option to provide permission blockchain network for sensitive information.</p><p>Blockchain can be used as a distributed ledger platform to model and facilitate government processes and track all public sector assets and expenditures. It can also be used to track and accommodate government transactions with full transparency and in a short time. Blockchain can be implemented in the identity management of citizens where it can be used to enable us to send and request credentials, and securely manage personal data. The same information can be used as a KYC in financial institutions.</p><p>Lastly, even though there may be some hurdles to implement it, for which studies and researches have to be conducted on how to overcome it accordingly. Nepal has been a trending follower in the global IT market and had done some really good innovations. Great things are happening in IT scenes now. We hope our government utilizes that manpower and be on its way to digitize Nepal.</p><p><em>Written By </em><a href="https://www.linkedin.com/pulse/nepal-government-blockchain-roshan-ranabhat/"><em>Roshan Ranabhat</em></a><em>: Fullstack developer at Ekbana.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=555050cdca32" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/nepal-government-and-blockchain-555050cdca32">Nepal Government and Blockchain</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Nepali Part of Speech(POS) Tagging]]></title>
            <link>https://medium.com/ekbana/nepali-part-of-speech-pos-tagging-72eff56111c0?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/72eff56111c0</guid>
            <category><![CDATA[napali-part-of-speech]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[nepali]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Pema Gurung]]></dc:creator>
            <pubDate>Fri, 25 Oct 2019 05:05:52 GMT</pubDate>
            <atom:updated>2019-10-25T05:05:52.528Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/876/0*dyFXxANhtnYGb2MH.png" /></figure><p>The Nepali language is mostly spoken by the people of Nepal and also in some part of India. Nepali is actually written with the Devanagari alphabet and is an Indo-Aryan Language. The Devanagari script, which is generally known as Nagari, is written from left to right. The order of the letters made up of vowels and consonants is known as the “varnamala” which means the “garland of flowers.” In the Unicode Conventional, the Devanagari is constituted in three blocks. U+0900–U+097F comprises the Devanagari, U+1CD0–U+1CFF comprises the Devanagari Extended, and U+A8E0–U+A8FF comprises the Vedic Extension.</p><p>The paper, “Structure of Nepali Grammar” by Bal Krishna Bal has an awesome explanation of the grammar of Nepali [1] where he explains how each part of speech is used in Nepali. Asmita (Student of Bal Krishna Bal) has also done her degree project under the guidance of Bal Krishna Bal on “Part of Speech Tagger for Nepali Text using SVM” where she got an accuracy of 88% [2]. Tej Bahadur Shahi, Tank Nath Dhamala, and Bikash Balami also published a paper on “Support Vector Machines based Part of Speech Tagging for Nepali Text” where they got an accuracy of 90% on TNT and 90% on SVM, using 80000 training data size[3].</p><p>Nepali and Hindi are quite similar as they both follow the Devanagari script.</p><h4><strong>Example:</strong></h4><p>English: I will go home</p><p>Hindi: मे घर जाऊंगा</p><p>Nepali: म घर जानेछु</p><p>As we can see that in Nepali and Hindi, the word “home” is same i.e. “घर” and both gives the POS tag as “NN”. So, same way lets implement the Nepali POS Tagger using TNT model just like we did for H<a href="https://medium.com/@pemagrg/hindi-part-of-speech-pos-tagging-5c3b8a6302b4">indi POS</a>. Lets Start!</p><p><strong>Let’s say we have a text to tag</strong></p><pre>text = &quot;१० वर्षीया बालिका बलात्कारपछि हत्या गर्ने सार्वजनिक&quot;</pre><p><strong>Let’s use our own tagged data and train it.</strong></p><pre>from nltk.tag import tnt<br>from nltk.corpus import indian</pre><pre>train_data = indian.tagged_sents(‘nepali.pos’)<br>tnt_pos_tagger = tnt.TnT()<br>tnt_pos_tagger.train(train_data)</pre><p><strong>Let’s Tag the text now!</strong></p><pre>tagged_words = (tnt_pos_tagger.tag(nltk.word_tokenize(text)))<br>print(tagged_words)</pre><pre>[<strong>OUTPUT</strong>]:<br> [(&#39;१०&#39;, &#39;CD&#39;), (&#39;वर्षीया&#39;, &#39;JJ&#39;), (&#39;बालिका&#39;, &#39;NN&#39;), (&#39;बलात्कारपछि&#39;, &#39;IN&#39;), (&#39;हत्या&#39;, &#39;NN&#39;), (&#39;गर्ने&#39;, &#39;VBNE&#39;), (&#39;सार्वजनिक&#39;, &#39;JJ&#39;)]</pre><p><strong>easy wasn’t it?</strong></p><p>The main issue here is that its difficult to get Nepali tagged corpus. So we tend to get the tag “Unk” most of the time while tagging the words ex: (‘वाशिंग’, ‘Unk’), (‘मशीन’, ‘Unk’).</p><p><strong>How to overcome that?</strong></p><ol><li>we can add more tagged sentences to nepali.pos</li><li>we can also use Google translator to translate and get the tag</li></ol><p>I have tried using Google Translator API to handle the “Unk” tags by translating and getting the tags and then appending it to the Nepali Corpus which gave a pretty good result. You can check in my Github.</p><p><strong>Github Link: </strong><a href="https://github.com/pemagrg1/Nepali-POS-Tagging-and-Keyword-Extraction">https://github.com/pemagrg1/Nepali-POS-Tagging-and-Keyword-Extraction</a></p><h4>REF:</h4><ol><li><a href="http://www.panl10n.net/english/outputs/Working%20Papers/Nepal/Microsoft%20Word%20-%207_E_N_396.pdf">http://www.panl10n.net/english/outputs/Working%20Papers/Nepal/Microsoft%20Word%20-%207_E_N_396.pdf</a></li><li><a href="https://github.com/asmitasubedi/Nepali-Pos-Tagger/blob/master/AsmitaS_Parts%20of%20Speech%20Tagger%20for%20Nepali%20Text%20Using%20SVM.pdf">https://github.com/asmitasubedi/Nepali-Pos-Tagger/blob/master/AsmitaS_Parts%20of%20Speech%20Tagger%20for%20Nepali%20Text%20Using%20SVM.pdf</a></li><li><a href="https://pdfs.semanticscholar.org/b365/05276ef839d9b0cf193c6e65032ca5c73b37.pdf">https://pdfs.semanticscholar.org/b365/05276ef839d9b0cf193c6e65032ca5c73b37.pdf</a></li><li><a href="https://github.com/prashant-subedi/nepali-pos-tagger">https://github.com/prashant-subedi/nepali-pos-tagger</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72eff56111c0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/nepali-part-of-speech-pos-tagging-72eff56111c0">Nepali Part of Speech(POS) Tagging</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Fine-tuning BERT for Text Classification (20news group classification)]]></title>
            <link>https://medium.com/ekbana/fine-tuning-bert-for-text-classification-20news-group-classification-53a55dc09738?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/53a55dc09738</guid>
            <category><![CDATA[text-classification]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ektech]]></category>
            <category><![CDATA[nepal]]></category>
            <category><![CDATA[nlp]]></category>
            <dc:creator><![CDATA[EKbana]]></dc:creator>
            <pubDate>Fri, 25 Oct 2019 05:05:13 GMT</pubDate>
            <atom:updated>2019-10-25T05:05:13.230Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*YYP2tMrPwz8tZGONKnKsMg.jpeg" /></figure><p>2018 has been a break-through year in the field of NLP. Google’s BERT, deep bidirectional training using the transformer, gave state of the art results for many NLP tasks. Therefore, In this tutorial, we will be learning to use Keras-Bert on TPU (Google collab recommended) and in GPU. It’s easy and handy.</p><p>BERT provides a pre-trained model for English and Chinese language and we are hoping will be available for other languages as well. Furthermore, we will be finetuning the English pre-trained model in this tutorial to classify the 20news group dataset.</p><p><strong>What does finetuning means??</strong></p><p>It means using an already well-trained model to further perform a similar task by twitching it a little. For example:</p><p>Let’s consider we have to classify images into fruits. So, we have a model already trained to classify 1000s classes of image. This model can be used to classify fruits.</p><p>Last layer of this model is removed and trainable is set false for the lower rate. Since they are already trained layer (while training 1000s classes classification) So, the feature extraction part of an image is taken from the previous model. Now add a required layer to the model and compile it.</p><p>Once this has been done, other added layers in the model can be set as ‘trainable=True’ so that in further epochs their weights can be fine-tuned for the new task of classification.</p><p>Consequently, we want to classify text by finetuning BERT. We will be classifying using a layer of Bert to classify news. For that, we will be taking the 20newsgroup dataset.</p><p><strong>let’s finetune the BERT model using Keras!!!!!!</strong></p><p>Keras offers a very quick way to prototype state-of-the-art deep learning models, and is, therefore an important tool we use in our work. I have shared a Kaggle and collab link, both <a href="https://colab.research.google.com/drive/1YSfscbb-g92m1vkYxY4IOVMWMfgfLLJD">using TPU</a> and <a href="https://www.kaggle.com/sharmilaupadhyaya/20newsgroup-classification-using-keras-bert-in-gpu">using GPU</a>.</p><p>Remember, In Keras-Bert you got to set the variable TF_KERAS to 1.</p><p>Install Keras-Bert as well as Keras-rectified-Adam for finetuning. And download uncased large pre-trained model of Bert with WordPiece tokenization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*7g9N0pcIFbKqSOzs" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/859/0*Ygn3Sew0PUZ2hd3f.png" /></figure><p><strong>Importing Libraries.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*R031RIgsMFgeTIj3" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/857/0*-b6AEp29QHNdK4yl.png" /></figure><p><strong>Parameters needed for training.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*-DUtu5vDuYBssGOH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/865/0*JlpAfYR9xhXZhf8A.png" /></figure><p>SEQ_LEN is a number of lengths of the sequence after tokenizing. It is set to 128. BERT has worked on at max 512 sequence length.</p><p>LR is the learning rate.</p><p>Remember, for TPU batch size 128 works like a charm, fast and easily trainable. While for GPU it will throw exhaustion error. So, reduce batch size for GPU.</p><p>Note: BERT in CPU is heavy and slow and my code froze. So, better use Kaggle kernel or Google Collab.</p><p>Now, load the pre-trained model of BERT.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*cnM_JMr0_sZcJUUH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*k5DvAnfCAFvGtnlx.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*_M4f_FvY4dY7nXcG" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/872/0*3l65iEB7NSKg-FWU.png" /></figure><p>Extracting token dictionary from vocab of pre-trained model to refer for input we will be using.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*miQOxXLGjmu_SnoI" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/814/0*RP2CI5asR4DsEzd4.png" /></figure><p>Download the dataset and extract using keras get_file function.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*Hj4e1lWsUBWrYGpb" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/888/0*VZkbolObLvbHam2k.png" /></figure><p>Defining tokenizer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*xFaOJN13yKDKUAMy" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/867/0*oCzN4a29L2s8xutA.png" /></figure><p>Making Label, index pair.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*48XLBVirgreG_ljH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/861/0*mGf_4e2hiO1-4oss.png" /></figure><p>Splitting data into train, test preceded by tokenizing sentences and returning test train data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*NVv5NJOGYnhq6saQ" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/0*YEnDenb-b8t17jY2.png" /></figure><p>Extracting layer from pre-trained bert model and adding a layer with softmax function to classify 20 classes of news.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*RlLShaNiuH_HpKdp" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/0*i_Ggah4tckANjrem.png" /></figure><p>Initializing variables.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*29gdljeX0zAYCrWf" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/858/0*wW9eNGA-eSSk-vtN.png" /></figure><p>Finally, training the model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*-yWq7-JYr3JHm8KF" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*2abMndEIKyxLpmdB.png" /></figure><p>Making predictions for the test datasets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*w63zJhnbK_qXDfvR" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/889/0*dQv8csI-f0KwZg0x.png" /></figure><p>Calculating accuracy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*_UUD_qg7AOt70FgH" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/899/0*-mY-feJ8HiPI-Upu.png" /></figure><p>Now, lets test for single data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*lqhgdQzoaSXFUc4V" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/867/0*xbwjRDDjv0AjmYGK.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*FUOB_QqKkoLkNYd3" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/872/0*MzCns-9qCkFzSHh0.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/60/0*YPmPvKz8Pia1OZbZ" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/867/0*0puUpCUDAODtAjiX.png" /></figure><p>Please refer to the Kaggle link to get access to the code. In this way, classification problems can be addressed using BERT. In the next series, we will be doing a sequential problem. Thank you for going through this article.</p><p>LINKS:</p><ol><li><a href="https://www.kaggle.com/sharmilaupadhyaya/20newsgroup-classification-using-keras-bert-in-gpu/edit/run/20803530">https://www.kaggle.com/sharmilaupadhyaya/20newsgroup-classification-using-keras-bert-in-gpu/edit/run/20803530</a></li><li><a href="https://colab.research.google.com/drive/1YSfscbb-g92m1vkYxY4IOVMWMfgfLLJD">https://colab.research.google.com/drive/1YSfscbb-g92m1vkYxY4IOVMWMfgfLLJD</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=53a55dc09738" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/fine-tuning-bert-for-text-classification-20news-group-classification-53a55dc09738">Fine-tuning BERT for Text Classification (20news group classification)</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pre-Processing Text in Python]]></title>
            <link>https://medium.com/ekbana/pre-processing-text-in-python-ad13ea544dae?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/ad13ea544dae</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[text-preprocessing]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[text]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Pema Gurung]]></dc:creator>
            <pubDate>Fri, 25 Oct 2019 04:57:09 GMT</pubDate>
            <atom:updated>2020-02-22T02:50:41.871Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/576/1*Lb-R0j5wfsBcaPMH5G8GUQ.png" /><figcaption>Source-<a href="https://ontotext.com/top-5-semantic-technology-trends-2017/">https://ontotext.com/top-5-semantic-technology-trends-2017/</a></figcaption></figure><p>So are you planning to do research in text fields but not sure about how to start?</p><p>Well, why not start with pre-processing of text as it is very important while doing research in the text field and its easy! Cleaning the text helps you get quality output by removing all irrelevant text and getting the forms of the words etc.</p><h4>In this article, we will be covering:</h4><p>1. Converting text to lowercase</p><p>2. Contraction</p><p>3. Sentence tokenize</p><p>4. Word tokenize</p><p>5. Spell Check</p><p>6. Lemmatize</p><p>7. Stemming</p><p>8. Remove Tags</p><p>9. Remove numbers</p><p>10. Remove punctuation</p><p>11. Remove stopwords</p><h3>Let’s START!</h3><h4>Pre-requisites:</h4><blockquote>install <a href="https://www.python.org/downloads/">Python</a></blockquote><blockquote>install <a href="https://www.nltk.org/install.html">NLTK</a></blockquote><blockquote>pip install autocorrect</blockquote><p><strong><em>Done with the installations? okay! let’s start coding!</em></strong></p><h4>Convert text to lower case:</h4><blockquote>Converting text to lower case as in, converting “Hello” to “hello” or “HELLO” to “hello”.</blockquote><pre>import nltk</pre><pre>def to_lower(text):<br>    &quot;&quot;&quot;<br>    Converting text to lower case as in, converting &quot;Hello&quot; to  &quot;hello&quot; or &quot;HELLO&quot; to &quot;hello&quot;.<br>    &quot;&quot;&quot;<br>    return &#39; &#39;.join([w.lower() for w in word_tokenize(text)])</pre><pre>text = &quot;&quot;&quot;Harry Potter is the most miserable, lonely boy you can imagine. He&#39;s shunned by his relatives, the Dursley&#39;s, that have raised him since he was an infant. He&#39;s forced to live in the cupboard under the stairs, forced to wear his cousin Dudley&#39;s hand-me-down clothes, and forced to go to his neighbour&#39;s house when the rest of the family is doing something fun. Yes, he&#39;s just about as miserable as you can get.&quot;&quot;&quot;</pre><pre>print (to_lower(text))</pre><h4>Remove Tags</h4><blockquote>Removing html tags from the text like “&lt;head&gt;&lt;body&gt;” using regex.</blockquote><pre>import re</pre><pre>text = &quot;&quot;&quot;&lt;head&gt;&lt;body&gt;hello world!&lt;/body&gt;&lt;/head&gt;&quot;&quot;&quot;<br>cleaned_text = re.sub(&#39;&lt;[^&lt;]+?&gt;&#39;,&#39;&#39;, text)<br>print (cleaned_text)</pre><pre><strong>[OUTPUT]:</strong><br>hello world!</pre><h4>Remove Numbers</h4><blockquote>Removing numbers from the text like “1,2,3,4,5…” We usually remove numbers when we do text clustering or getting keyphrases as we numbers doesn’t give much importance to get the main words. To remove numbers, you can use: <strong>.isnumeric() </strong>else <strong>.isdigit()</strong></blockquote><pre>text = &quot;There was 200 people standing right next to me at 2pm.&quot;<br>output = &#39;&#39;.join(c for c in text if not c.isdigit())<br>print(output)</pre><pre><strong>[OUTPUT]:<br></strong>There was  people standing right next to me at pm.</pre><h4>Remove punctuation</h4><blockquote>Removing punctuation from the text like “.?!” and also the symbols like “@#$” .</blockquote><pre>from string import punctuation</pre><pre>def strip_punctuation(s):<br>    return &#39;&#39;.join(c for c in s if c not in punctuation)<br>text = &quot;Hello! how are you doing?&quot;</pre><pre>print (strip_punctuation(text))</pre><pre><strong>[OUTPUT]:<br></strong>Hello how are you doing</pre><h4>Lemmatize</h4><blockquote>Lemmatization is similar to stemming but it brings context to the words.So it links words with similar meaning to one word.lemmatization does morphological analysis of the words. In short, lemmatize the text so as to get its root form eg: functions,funtionality as function</blockquote><pre>import nltk<br>from nltk.corpus import stopwords<br>from nltk.stem import WordNetLemmatizer<br>#is based on The Porter Stemming Algorithm</pre><pre>stopword = stopwords.words(‘english’)<br>wordnet_lemmatizer = WordNetLemmatizer()</pre><pre>text = “the functions of this fan is great”<br>word_tokens = nltk.word_tokenize(text)<br>lemmatized_word = [wordnet_lemmatizer.lemmatize(word) for word in word_tokens]<br>print (lemmatized_word)</pre><pre><strong>[OUTPUT]</strong>: [&#39;the&#39;, &#39;function&#39;, &#39;of&#39;, &#39;this&#39;, &#39;fan&#39;, &#39;is&#39;, &#39;great&#39;]</pre><h4>Stemming</h4><blockquote>Stemming and Lemmatization are Text Normalization (or sometimes called Word Normalization)<br>It does not keep a lookup table for actual stems of the word but applies algorithmic rules to generate stems. It uses the rules to decide whether it is wise to strip a suffix.</blockquote><pre>import nltk<br>from nltk.corpus import stopwords<br>from nltk.stem import SnowballStemmer<br>#is based on The Porter Stemming Algorithm</pre><pre>stopword = stopwords.words(‘english’)<br>snowball_stemmer = SnowballStemmer(‘english’)</pre><pre>text = “This is a Demo Text for NLP using NLTK. Full form of NLTK is Natural Language Toolkit”<br>word_tokens = nltk.word_tokenize(text)<br>stemmed_word = [snowball_stemmer.stem(word) for word in word_tokens]<br>print (stemmed_word)</pre><pre><strong>[OUTPUT]</strong>: [&#39;this&#39;, &#39;is&#39;, &#39;a&#39;, &#39;demo&#39;, &#39;text&#39;, &#39;for&#39;, &#39;nlp&#39;, &#39;use&#39;, &#39;nltk&#39;, &#39;.&#39;, &#39;full&#39;, &#39;form&#39;, &#39;of&#39;, &#39;nltk&#39;, &#39;is&#39;, &#39;natur&#39;, &#39;languag&#39;, &#39;toolkit&#39;]</pre><h4>word tokenize</h4><blockquote>Tokenize words to get the tokens of the text i.e breaking the sentences into list of words.</blockquote><pre>text = “This is a Demo Text for NLP using NLTK. Full form of NLTK is Natural Language Toolkit”<br>word_tokens = nltk.word_tokenize(text)<br>print (word_tokens)</pre><pre><strong>[OUTPUT]</strong>: [&#39;This&#39;, &#39;is&#39;, &#39;a&#39;, &#39;Demo&#39;, &#39;Text&#39;, &#39;for&#39;, &#39;NLP&#39;, &#39;using&#39;, &#39;NLTK&#39;, &#39;.&#39;, &#39;Full&#39;, &#39;form&#39;, &#39;of&#39;, &#39;NLTK&#39;, &#39;is&#39;, &#39;Natural&#39;, &#39;Language&#39;, &#39;Toolkit&#39;]</pre><h4>sentence tokenize</h4><blockquote>If the there are more than 1 sentence split it i.e breaking the sentences to a list of sentence.</blockquote><pre>text = “This is a Demo Text for NLP using NLTK. Full form of NLTK is Natural Language Toolkit”<br>sent_token = nltk.sent_tokenize(text)<br>print (sent_token)</pre><h4>Stop words removal</h4><blockquote>Remove irrelevant words using nltk stop words like “is,the,a” etc from the sentences as they don’t carry any information.</blockquote><pre>import nltk<br>from nltk.corpus import stopwords<br>stopword = stopwords.words(‘english’)</pre><pre>text = “This is a Demo Text for NLP using NLTK. Full form of NLTK is Natural Language Toolkit”<br>word_tokens = nltk.word_tokenize(text)<br>removing_stopwords = [word for word in word_tokens if word not in stopword]<br>print (removing_stopwords)</pre><pre><strong>[OUTPUT]</strong>: [&#39;This&#39;, &#39;Demo&#39;, &#39;Text&#39;, &#39;NLP&#39;, &#39;using&#39;, &#39;NLTK&#39;, &#39;.&#39;, &#39;Full&#39;, &#39;form&#39;, &#39;NLTK&#39;, &#39;Natural&#39;, &#39;Language&#39;, &#39;Toolkit&#39;]</pre><h4>Contraction</h4><blockquote>Contraction helps to expand the word form like “ain’t”: “am not”. Contractions file has been created in my <a href="https://github.com/pemagrg1/Text-Pre-Processing-in-Python">github</a> which we will be importing to use it.</blockquote><pre># coding: utf-8<br>import re<br>import nltk<br>from contractions import contractions_dict</pre><pre>def expand_contractions(text, contractions_dict):<br>    contractions_pattern = re.compile(&#39;({})&#39;.format(&#39;|&#39;.join(contractions_dict.keys())),<br>                                      flags=re.IGNORECASE | re.DOTALL)</pre><pre>def expand_match(contraction):<br>        match = contraction.group(0)<br>        first_char = match[0]<br>        expanded_contraction = contractions_dict.get(match) \<br>            if contractions_dict.get(match) \<br>            else contractions_dict.get(match.lower())<br>        expanded_contraction = expanded_contraction<br>        return expanded_contraction</pre><pre>expanded_text = contractions_pattern.sub(expand_match, text)<br>    expanded_text = re.sub(&quot;&#39;&quot;, &quot;&quot;, expanded_text)<br>    return expanded_text</pre><pre>def main():<br>    text = &quot;&quot;&quot;I ain&#39;t going there. You&#39;ll have to go alone.&quot;&quot;&quot;<br>    <br>    text=expand_contractions(text,contractions_dict)<br>    sentences = nltk.sent_tokenize(text)<br>    tokenized_sentences = [nltk.word_tokenize(sentence) for sentence in sentences]<br>    <br>    print (tokenized_sentences)</pre><pre>if __name__ == &#39;__main__&#39;:<br>    main()</pre><h4>Spell Check</h4><blockquote>correct the incorrect spelled words like “wrld” to “world”</blockquote><pre>from autocorrect import spell</pre><pre>text = &quot;This is a wrld of hope&quot;<br>spells = [spell(w) for w in (nltk.word_tokenize(text))]<br>print (spells)</pre><pre><strong>[OUTPUT]</strong>: [&#39;This&#39;, &#39;is&#39;, &#39;a&#39;, &#39;world&#39;, &#39;of&#39;, &#39;hope&#39;]</pre><h4>PS:</h4><p>You can get all the above code from <a href="https://github.com/pemagrg1/Text-Pre-Processing-in-Python">GITHUB</a></p><p>Once you are done with Pre-processing, you can then move to NER, clustering, word count, sentiment analysis, etc.</p><p>let me give you a small demo on word count which helps us to get the main words from the document. So, once you are done with pre-processing you are left with a clean list of words i.e without stopwords, numbers, punctuations, etc. Now what you can do is count the number of times the words are repeated which is known as word frequency. To iterate through the pre-processed list and count the frequency.</p><pre>for word in pre_process_text:<br>    if word in word_count:<br>        word_count[word] += 1<br>    else:<br>        word_count[word] = 1</pre><pre>#Print in sorted order<br>for w in sorted(word_count, key=word_count.get, reverse=True):<br>    print (w, word_count[w])</pre><p>Then you get a list of words with its frequency which you can analyze and say that these are the main words of this document or that this document talks about “give the top frequency words”. Easy right?</p><p>All the best with your research!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ad13ea544dae" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/pre-processing-text-in-python-ad13ea544dae">Pre-Processing Text in Python</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Buzz of IoT]]></title>
            <link>https://medium.com/ekbana/the-buzz-of-iot-e4d520ab5785?source=rss----8a692552e928---4</link>
            <guid isPermaLink="false">https://medium.com/p/e4d520ab5785</guid>
            <category><![CDATA[nepal]]></category>
            <category><![CDATA[ektech]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[internet-of-things]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[EKbana]]></dc:creator>
            <pubDate>Tue, 22 Oct 2019 10:37:17 GMT</pubDate>
            <atom:updated>2019-10-22T10:37:17.476Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*lHj7EgpqYklNctkV" /></figure><p>The term “Internet of Things”, commonly abbreviated as (and substituted with) IoT has been with us since 1999 and Kevin Ashton; who coined the term; himself agrees the term has evolved to grow to a much wider context than he originally meant. However the central theme of IoT, which he clarifies in the 2009 article on <em>RFID Journal</em> to be something that “enables computers to observe, identify and understand the world” and something that would be “able to track and count everything, and greatly reduce waste, loss and cost” using sensors remains fairly the same. The idea was around since at least 1991 and gained popularity through the Auto-ID Lab at MIT(formerly known as Auto-ID Center which Kevin himself had cofounded).</p><p><strong>Internet of Things </strong>in modern terms is commonly understood as a network of physical devices that are connected to the Internet for collecting and sharing data. The interrelated devices may be mobile, computing devices, electromechanical or digital equipment, embedded systems or more commonly a source of raw data (such as sensors) or any embedded electronics that can transfer data over a network. The connections between the components enable then for real-time communication to automate decision making with minimal (and convenient) human involvement.</p><p>IoT is more than just a machine-to-machine (M2M) communication but could be thought of as convergence of M2M communications and data analytics/machine learning algorithms over the huge data thus gathered. This pipeline can then be applied to automate processes and decision making. M2M communications are generally simpler and less intelligent. Example of M2M communication could be vending machines, ATMs, smart meters or medical-equipments that send out information of inventory or dispensed cash or other similar measurements or interactions. The notable difference is that M2M networks are exclusively used to allow machines, sensors and controllers to communicate however in case of IoT, the same network is used by humans (via phones, tablets, computers). It is necessary to understand the degree of connectedness here and to distinguish the nuances of connected devices, interconnected devices and intelligently connected devices. IoT based systems are more commonly understood as “smart” systems, such as “smart-industry”, “smart-home”, “smart-parking” and so on. Owing to the growth of wireless technology and the Internet becoming more accessible every day and also more devices getting “smarter”, the IoT technology and industry is growing at an unprecedented scale. International Data Corporation(IDC) says the market for worldwide IoT solutions will go from $1.9 trillion in 2013 to $7.1 trillion in 2020. Predictions have been made that potential for growth from implementing Industrial IoT may generate around $12 trillion of global GDP by 2030.</p><p>According to predictions made by Microsoft (on IoT signals report 2019), 94% of businesses will be using IoT by the end of 2021. The predictions were based on a research survey conducted with over 3,000 decision-makers at the top enterprise companies involved in IoT from all over the world. The report says that 87% of decision-makers in manufacturing are already adopting some IoT and the percentage is 90%, 86% and 83% in retail/wholesale, transportation and government respectively. The same report also emphasizes that the continued success of IoT will also rely on other innovations such as AI, edge computing and 5G. According to the report, companies adopt commercial IoT for operations optimization (56 %), followed by employee productivity (47 %), and safety and security (44 %). Microsoft says “<em>There will be more than 20 billion connected devices by 2020, but IoT is more than just your smart speaker and thermostat. Connected devices can save energy in smart buildings, improve the flow of traffic in congested cities, increase crop yields, prevent equipment failures on a factory floor, and power medical breakthroughs.</em>”</p><p>Though IoT is something discussed a lot about, it’s precise implementation model is widely debated one. Part of the reason is that IoT systems are complex and varied and IoT solutions are mostly specific to a situation that needs to be addressed. However, in general, a typical IoT system includes the following key components.</p><ol><li>Sensors/Devices</li><li>Connectivity</li><li>Data Processing</li><li>User Interface</li></ol><p>An IoT system consists of sensors/devices which talks to the data cloud through some kind of protocol and commonly used network connectivity. The software that processes the cloud-data automatically makes decisions and sends commands to the sensors/devices and other notifications/alert to relevant nodes or human users. There could be few exceptions to such connectivity schemes in cases where the communication latency could create serious to human or other similar fatality such as in a self-driving car at emergency conditions or the communication is too expensive or the data is mostly trivial (such as a security camera footage of an empty hallway). In such a case, computing may be done on a gateway or on the sensor/device itself (edge computing).</p><p>A user can intervene or monitor the whole process using an interface provided in the IoT platform. User can also apply situation-specific data aggregation, visualization or decision-making methods to data in case of necessity using the IoT platform’s user interface.</p><p><strong>IoT Use cases</strong></p><p>IoT has a wide and unparalleled use case at a different hierarchical level. From making and individual consumer’s day to day experience more convenient by turning the thermostat on or off, to irrigating the agriculture farm automatically based on weather and sensor data, to automate manufacturing process, in fault / defect identification and early failure detection, components wear-out identification and decision making in redesigning manufacture plant, IoT could have a key role. Condition-based maintenance could replace scheduled maintenance in industries. Smart farms including grapevines to marijuana plants, sensors have offered deep insight on crop production. IoT can assist decision-makers and designers to decide new features on a car based on consumer behaviour, response and usage of existing features, failure data, accident patterns or notify doctors or relatives about increased blood pressure or heart rate or alert of fire and respond to it. Tesla was once able to complete a mandated fix of charger plugs on more than 29000 vehicles via a software IoT, which wouldn’t have been possible without getting the car connected to the internet. In another case, by rigorously collecting, aggregating and analyzing patient data, Medtronic (world’s largest medical device company) has been able to deliver the personalized care coordination to over 95,000 individuals. IoT can connect power grids over the city or country, pipelines, and irrigation canals. In large scale, IoT and IoT data can be used in urban planning, transportation and management, security, agriculture, disaster warning and management and so on. The key use and potential use cases of IoT as listed by Microsoft is listed here.</p><p>Manufacturing</p><ol><li>Industrial automation</li><li>Quality and compliance</li><li>Production planning and scheduling</li><li>Supply chain and logistics</li><li>Plant safety and security</li></ol><p>Retail/wholesale</p><ol><li>Supply chain optimization</li><li>Inventory optimization</li><li>Surveillance and security</li><li>Loss prevention</li><li>Energy optimization</li></ol><p>Healthcare</p><ol><li>Tracking patient, staff and inventory</li><li>Remote device monitoring and service</li><li>Remote health monitoring and assistance</li><li>Safety, security, and compliance</li><li>Facilities management</li><li>Government/smart cities</li><li>Public safety</li><li>Infrastructure and facilities management</li><li>Regulations and compliance management</li><li>Fleet and asset management</li><li>Incident response</li></ol><p>Beside the categorical list by Microsoft, IoT could be has wide applications in smart homes, wearables (gears, glasses, watches, clothes), tracking and delivery systems, office automation, smart grids, smart retail, warehouse management, smart farming and precision agriculture, traffic management and automation, self-driving transportation systems and many more.</p><p><strong>Opportunities, Value and Challenges</strong></p><p>As the growth of data and connectedness increases, there are many opportunities an enterprise should address. Capturing diverse forms of data from a variety of interconnected devices creates a potential of finding new sources of value and insight through data. Enterprises can have a better opportunity to understand customer behaviour and enable deeper customer relationships. Using AI-based applications in the data could help in operational and financial proficiency as extensive access to data can drive insights that uncover new business opportunities and create a competitive advantage in business and operating models.</p><p>With competitors offering more compelling value propositions through the use of innovative technologies, companies are recognizing a need to reinvent their enterprises by digitizing communications and interactions. And about two-thirds of executives surveyed by IBM in 2017 are executing plans to incorporate IoT into their operating models. AI and other cutting edge technologies come together in such reinvention policies and in creating insight from connected things. “AI and IoT could help us fully integrate data suppliers and data consumers into a data-driven business model,’’ says the CFO of Life Science and Pharmaceuticals depicting how businesses model change and new business model could emerge from proliferating data which could be both a challenge and opportunity.</p><p>The convergence of new technologies and intelligent machines is reshaping the future of work and the workforce. For example, traditional methods of analyzing structured data aren’t designed to efficiently process the vast amounts of real-time data that stream from IoT devices. AI-based analysis and responses become critical for extracting optimal value from the data and human workers can evolve to higher-value tasks associated with understanding and exploiting AI-driven insights, and reinforcing those findings.</p><p>The whole idea of IoT adoption, however, is a very delicate and risky task. While the complexity and technical challenges are the visible barriers, lack of skilled workers is still a major global problem in IoT adoption. Microsoft’s IoT Signals report states “<em>IoT’s promise will be unlocked by addressing the skills shortage, complexity and security</em>”. According to Microsoft, seven key ingredients for success in IoT are; business strategy, leadership and organization, a technology roadmap, talent, operations and core business processes, partnerships, and security. Right now though, IoT devices are not being configured to speak a single language and no universally accepted communications standard is here that will enable all devices to communicate. Although some progress has been made, lack of standardization is still known to slow down IoT adoption. Data privacy and openness of data; as to who owns it, who benefits and who can access it; in IoT also have been questioned frequently.</p><p>In technical level, the IoT needs an open architecture to maximise interoperability among heterogeneous systems and distributed resources including providers and consumers of information and services, whether they be human beings, software, smart objects or devices. Architecture standards should consist of well-defined abstract data models, interfaces and protocols, together with concrete bindings to neutral technologies (such as XML, web services etc.) to support the widest possible variety of operating systems and programming languages. IoT nodes may need to form peer networks with other nodes dynamically and autonomously locally or remotely, this should be done through a decentralized, distributed approach to the architecture, with support for semantic search, discovery and peer networking. At the analytics stage, the architecture should be able to dock with machine learning algorithms in cloud or edge computing device. Effective caching, disconnected operations with the ability for synchronization with event-driven architecture needs to be established for an effective IoT system.</p><p><strong>IoT platforms</strong></p><p>While tools such as Node-RED used with devices such as Raspberry-Pi are faster to prototype, easier to test and give decent opportunity to wire hardware devices, APIs and services, a robust platform that promises all major requirements for a complete IoT solution backed up with other cutting-edge support requires much more than just that. Giants such as IBM, Microsoft, Amazon, Google, Cisco, Oracle have developed their IoT platforms and companies such as Libelium are developing both plug-and-play type hardware and sensor platform suited for IoT and smart systems.</p><p><strong>Closing remarks</strong></p><p>The growth of IoT is evident and companies in all industries are looking for ways to leverage IoT. However, IoT could be fruitless without enough data-analytics capabilities and could even pose security threats and serious privacy concerns without a proper way to manage and secure data. Certainly, the technologies of the Internet of things such as RFID and sensor make our life better, easier and more comfortable. With IoT’s contribution in saving and optimizing energy and water usage, monitoring water and other natural resources, alerting of wildfire and other hazards, strategists are planning to leverage IoT to achieve sustainable development goals for 2030.</p><p><strong>Reference</strong></p><p><a href="http://www.libelium.com/libelium-summarizes-the-most-demanded-features-of-iot-technology-for-smart-agriculture-in-a-new-quick-report/">http://www.libelium.com/libelium-summarizes-the-most-demanded-features-of-iot-technology-for-smart-agriculture-in-a-new-quick-report/</a></p><p><a href="http://www.libelium.com/iot-helps-to-reach-sustainable-development-goals-to-transform-our-world/">http://www.libelium.com/iot-helps-to-reach-sustainable-development-goals-to-transform-our-world/</a></p><p><a href="https://en.wikibooks.org/wiki/A_Bit_History_of_Internet/Chapter_8_:_Internet-of-Things">https://en.wikibooks.org/wiki/A_Bit_History_of_Internet/Chapter_8_:_Internet-of-Things</a></p><p><a href="https://internetofthingsagenda.techtarget.com/definition/Internet-of-Things-IoT">https://internetofthingsagenda.techtarget.com/definition/Internet-of-Things-IoT</a></p><p><a href="https://www.link-labs.com/blog/what-is-m2m">https://www.link-labs.com/blog/what-is-m2m</a></p><p><a href="http://www.prnewswire.com/news-releases/low-power-wide-area-lpwa-networks-address-a-sweet-spot-in-iot-according-to-strategy-analytics-273916181.html">http://www.prnewswire.com/news-releases/low-power-wide-area-lpwa-networks-address-a-sweet-spot-in-iot-according-to-strategy-analytics-273916181.html</a></p><p><a href="http://www.cmswire.com/cms/internet-of-things/idc-the-internet-of-things-will-grow-to-71t-infographic-025455.php">http://www.cmswire.com/cms/internet-of-things/idc-the-internet-of-things-will-grow-to-71t-infographic-025455.php</a></p><p><a href="https://www.forbes.com/sites/insights-hitachi/2018/06/12/how-tesla-nest-and-medtronic-are-using-iot-derived-data-to-move-their-businesses-forward/#14b620f7487a">https://www.forbes.com/sites/insights-hitachi/2018/06/12/how-tesla-nest-and-medtronic-are-using-iot-derived-data-to-move-their-businesses-forward/</a></p><p><a href="https://data2gowireless.com/biggest-obstacles-facing-iot/">https://data2gowireless.com/biggest-obstacles-facing-iot/</a></p><p><a href="https://www.leverege.com/blogpost/iot-explained-how-does-an-iot-system-actually-work-part-2">https://www.leverege.com/blogpost/iot-explained-how-does-an-iot-system-actually-work-part-2</a><a href="https://azure.microsoft.com/en-us/iot/signals/">https://azure.microsoft.com/en-us/iot/signals/</a></p><p><a href="https://news.microsoft.com/2019/07/30/microsoft-announces-iot-signals-research-report-on-state-of-iot-adoption/">https://news.microsoft.com/2019/07/30/microsoft-announces-iot-signals-research-report-on-state-of-iot-adoption/</a></p><p><a href="https://blogs.microsoft.com/blog/2019/07/30/iot-signals-report-iots-promise-will-be-unlocked-by-addressing-skills-shortage-complexity-and-security/">https://blogs.microsoft.com/blog/2019/07/30/iot-signals-report-iots-promise-will-be-unlocked-by-addressing-skills-shortage-complexity-and-security/</a></p><p><a href="https://www.ibm.com/services/insights/c-suite-study/iot">https://www.ibm.com/services/insights/c-suite-study/iot</a></p><p><a href="https://www.networkworld.com/article/3196191/wifi-s-evolving-role-in-iot.html">https://www.networkworld.com/article/3196191/wifi-s-evolving-role-in-iot.html</a></p><p>PS:</p><p>The article also contains compiled and as is summary and images from different resources mentioned in the reference. Some of the figures and quotes are directly taken from the sources. No infringement of any form of IP rights is intended.</p><p>This article was written by Madav Humagain- ML engineer at Ekbana<br>Link to <a href="https://medium.com/@scimad/the-buzz-of-iot-a136df36ed51">Madav</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e4d520ab5785" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ekbana/the-buzz-of-iot-e4d520ab5785">The Buzz of IoT</a> was originally published in <a href="https://medium.com/ekbana">EKbana</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>