Jekyll2023-01-28T01:16:28+00:00/feed.xmlChamoda PandithageOccasionally updated.Public-key Cryptography Explained2017-12-13T00:00:00+00:002017-12-13T00:00:00+00:00/2017/12/13/public-key-cryptography-explained<p>Public-key cryptography is one of the most used cryptosystems today. It refers to any system that uses a key pair, one for encrypting data and another one for decrypting data. If data encrypted using a key, other key is used to decrypt it. This seems pretty magical at first, but in the end of blog post you will understand how this works. In this blog post I’ll start with an analogy to understand what is the purpose of using two key pairs. Then I’ll explain the mathematical concepts behind the algorithm. Then I’ll implement a toy algorithm to understand it further (But never design your own crypto algorithms). Next I’ll explain some <code class="language-plaintext highlighter-rouge">openssl</code> commands to generate RSA public and private keys which you can use in real world applications.</p>
<h1 id="lets-start-with-an-analogy">Let’s start with an analogy</h1>
<p>Suppose you are at home and need to send your passbook to the bank. You have to send this by a bad courier service, in fact they always try to inspect and spy what’s inside every package. So you buy a locker box with two identical keys. You keep one to yourself and send other one to the bank. You can’t send the key with the package because the courier service will open the locker box and inspect. You can’t send it separately because this bad courier always make copies of the keys they deliver in hopes of trying them out on future deliveries. So you walk in to the bank yourself to deliver the key to the bank. Now you go home and put the passbook inside the locker box and lock it with your identical key and send it via the bad courier. They deliver the box hopelessly without being able to see what’s inside. It was inefficient, you had to visit yourself to the bank first. But can we do better?</p>
<p>This time the bank buy a new kind of locker for this purpose. The new locker box has two keys, one for locking the box (public key), another one for unlocking (private key). The key used to lock can’t be used to unlock the box. The key used to unlock can’t be used to lock the box. Bank send over the locker box to you along with the locking key (public key). As usual the bad courier makes a copy of the key in hopes of future endeavors. Now you put the passbook inside the box and lock it with your locking key (public key). You keep the key and send the box. Courier tries to unlock the box using the key the have had copied, but no luck. It can only unlock using the unlocking key (private key) only owned by the bank.</p>
<p>Modern Internet is like the bad courier, filled with hackers inspecting unencrypted packets. We need something similar to the second paragraph to secure the Internet communication.</p>
<p>The first paragraph is an analogy of symmetric encryption. Second paragraph discuss asymmetric encryption, which is the category public encryption belongs. But how could we design such a lock digitally?</p>
<h1 id="the-underlying-mathematics">The Underlying Mathematics</h1>
<p>We can write our encryption function as \(C = E(M)\) where \(M\) is the message we want to encrypt, \(E\) is the function that does the encryption and \(C\) is the encrypted message. Decryption is \(M = D(C)\). Let’s define our functions.</p>
\[E(M) = M^e \bmod n\]
\[D(C) = C^d \bmod n\]
<p>\(e\) and \(d\) are public and private keys respectively. \(n\) is a large number which is a multiple of two large prime numbers. \(\bmod\) means the modulo operation. Most programming languages represent this by <code class="language-plaintext highlighter-rouge">%</code> symbol. Given two positive numbers \(a\) and \(n\) result of \(a \bmod n\) is the reminder when \(a\) divides by \(n\). For example \(7 \bmod 3 = 1\) because when \(7\) divides by \(3\) remainder is \(1\). We could write above equations in another way using modular arithmetic syntax.</p>
\[E(M) \equiv M^e \pmod{n}\]
\[D(C) \equiv C^d \pmod{n}\]
<p>If \(a \equiv b \pmod{n}\) then \(a\) and \(b\) has a congruence relationship. That means \(a - b\) is dividable by \(n\) and \(a \bmod n\) and \(b \bmod n\) both has the same reminder.</p>
<p>Now we can write the following congruence relation. This relationship must be true for encryption and decryption to work properly. \(D(E(M))\) must always return original \(M\) back.</p>
\[M \equiv D(E(M)) \equiv (M^e)^d \pmod{n}\]
<p>Now we need to find a relationship between \(e\) and \(m\) in such that \(D(E(M))\) holds true. To move forward we need another equation. For that we are going to start with Fermat’s little theorem from number theory. Theorem state the follows</p>
\[a^p \equiv a \pmod{p}\]
<p>If \(p\) is a prime number and \(a\) is any integer number \(a^p - a\) is an integer multiple of \(p\). We can also write this as</p>
\[a \times a^{p - 1} \equiv a \times 1 \pmod{p}\]
<p>Removing \(a\) from both sides we get</p>
\[a^{p - 1} \equiv 1 \pmod{p}\]
<p>Now let’s look at a function called Euler’s totient function. In number theory Euler’s totient function counts number of positive integers given integer \(n\) that are relatively prime to \(n\). Relative primes are numbers that don’t have divisors other than \(1\). If \(a\) and \(b\) are relatively prime then we can write that greatest common divisor is \(1\). We usually write this as \(gcd(a, b) = 1\). For example \(10\) and \(7\) are relatively prime because \(gcd(10, 7) = 1\). Now to the totient function, notice that if \(n\) is a prime all the postive integers less than \(n\) are relatively prime to \(n\). We can write this as</p>
\[\phi(n) = n - 1\]
<p>So what does it have to do with our previous equations? Now we can write something like this using Euler’s totient function.</p>
\[M^{\phi(n)} \equiv 1 \pmod{n}\]
<p>Now we are going to prepare above equations to match \(M^{ed} \equiv M \pmod{n}\) which is similar to \(M \equiv D(E(M)) \equiv (M^e)^d \pmod{n}\) with some reordering. As the first step of the preparation take the power \(k\) on both sides</p>
\[M^{k\phi(n)} \equiv 1^k \pmod{n}\]
<p>Since \(1^k\) is \(1\)</p>
\[M^{k\phi(n)} \equiv 1 \pmod{n}\]
<p>Now Let’s multiply both sides by \(M\)</p>
\[M \times M^{k\phi(n)} \equiv M \times 1 \pmod{n}\]
\[M^{k\phi(n) + 1} \equiv M \pmod{n}\]
<p>Now we also have the following equation</p>
\[M^{ed} \equiv M \pmod{n}\]
<p>Did you notice the similarity in the right side of the two equations? Now we can write</p>
\[k\phi(n) + 1 = ed\]
<p>Now this is exactly the following by the definition of modular arithmetic syntax.</p>
\[ed \equiv 1 \pmod{\phi(n)}\]
<p>We can expand \(\phi(n)\) further because \(n\) is a multiple of two prime numbers.</p>
\[n = p \times q\]
\[\phi(n) = \phi(p) \times \phi(q)\]
\[\phi(n) = (p - 1) \times (q - 1)\]
<p>So we can write</p>
\[ed \equiv 1 \pmod{(p -1 )(q - 1)}\]
<p>Now choose a random private key \(d\) such that \(1 < d < \phi(n)\) and \(gcd(d, \phi(n)) = 1\) (Which means they need to be relatively prime). Now we can find \(e\) using modular inverse algorithm.</p>
<p>Now that we have found keys \(e\) and \(d\) we can use them to encrypt and decrypt messages. In the next section I will give a numeric example which will clear up most of the details.</p>
<h1 id="numerical-example">Numerical Example</h1>
<p>Let’s choose two prime numbers \(p\) \(q\). We choose small values to make calculations easier but in practice RSA algorithm use very large prime numbers.</p>
\[p = 3\]
\[q = 11\]
\[n = p \times q = 3 \times 11 = 33\]
<p>And let’s calculate the Euler’s totient function</p>
\[\phi(n) = (p - 1)(q - 1) = 2 \times 10 = 20\]
<p>Now we choose \(d\) such that it’s relatively prime to \(\phi(n)\). Let’s say</p>
\[d = 7\]
<p>Now compute \(e\) such that</p>
\[d \times e \bmod \phi(n) = 1\]
\[7 \times e \bmod 20 = 1\]
<p>If \(e = 3\) then \(7 \times 3 \bmod 20 = 1\) satisfies the equation.</p>
<p>Now private key is \(d = (7, 33)\) and public key is \(e = (3, 33)\). Now let’s encrypt some number \(M = 2\) using public key. Of course if you need to encrypt chars you must convert them to integers first.</p>
\[C = M^e \bmod n\]
\[C = 2^3 \bmod 33\]
\[C = 8\]
<p>Now let’s decrypt \(C\) using private key.</p>
\[M = C^d \bmod n\]
\[M = 8^7 \bmod 33\]
\[M = 2\]
<p>It works :). I’ve implemented this as a toy algorithm using python. Never implement your own cryptography algorithms though, to use in production environments.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="kn">from</span> <span class="nn">__future__</span> <span class="kn">import</span> <span class="n">division</span>
<span class="kn">from</span> <span class="nn">Crypto.Util</span> <span class="kn">import</span> <span class="n">number</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">from</span> <span class="nn">fractions</span> <span class="kn">import</span> <span class="n">gcd</span>
</code></pre></div></div>
<p><code class="language-plaintext highlighter-rouge">from Crypto.Util import number</code> for generating random primes.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">class</span> <span class="nc">RSA</span><span class="p">:</span>
<span class="n">p</span> <span class="o">=</span> <span class="n">q</span> <span class="o">=</span> <span class="n">n</span> <span class="o">=</span> <span class="n">d</span> <span class="o">=</span> <span class="n">e</span> <span class="o">=</span> <span class="n">pi_n</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">def</span> <span class="nf">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="p">.</span><span class="n">generate</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">generate</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="p">.</span><span class="n">p</span> <span class="o">=</span> <span class="n">number</span><span class="p">.</span><span class="n">getPrime</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">os</span><span class="p">.</span><span class="n">urandom</span><span class="p">)</span>
<span class="bp">self</span><span class="p">.</span><span class="n">q</span> <span class="o">=</span> <span class="n">number</span><span class="p">.</span><span class="n">getPrime</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">os</span><span class="p">.</span><span class="n">urandom</span><span class="p">)</span>
<span class="bp">self</span><span class="p">.</span><span class="n">n</span> <span class="o">=</span> <span class="bp">self</span><span class="p">.</span><span class="n">p</span> <span class="o">*</span> <span class="bp">self</span><span class="p">.</span><span class="n">q</span>
<span class="bp">self</span><span class="p">.</span><span class="n">pi_n</span> <span class="o">=</span> <span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">p</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span> <span class="o">*</span> <span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">q</span> <span class="o">-</span> <span class="mi">1</span><span class="p">)</span>
<span class="bp">self</span><span class="p">.</span><span class="n">d</span> <span class="o">=</span> <span class="bp">self</span><span class="p">.</span><span class="n">choose_d</span><span class="p">()</span>
<span class="bp">self</span><span class="p">.</span><span class="n">e</span> <span class="o">=</span> <span class="bp">self</span><span class="p">.</span><span class="n">choose_e</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">choose_d</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="p">.</span><span class="n">d</span> <span class="o">=</span> <span class="bp">self</span><span class="p">.</span><span class="n">find_a_coprime</span><span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">pi_n</span><span class="p">)</span>
<span class="k">return</span> <span class="bp">self</span><span class="p">.</span><span class="n">d</span>
<span class="k">def</span> <span class="nf">find_a_coprime</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">a</span><span class="p">):</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="n">a</span><span class="p">):</span>
<span class="k">if</span> <span class="n">gcd</span><span class="p">(</span><span class="n">i</span><span class="p">,</span> <span class="n">a</span><span class="p">)</span> <span class="o">==</span> <span class="mi">1</span><span class="p">:</span>
<span class="k">return</span> <span class="n">i</span>
<span class="k">def</span> <span class="nf">choose_e</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">n</span><span class="p">):</span>
<span class="k">if</span> <span class="p">(</span><span class="n">i</span> <span class="o">*</span> <span class="bp">self</span><span class="p">.</span><span class="n">d</span><span class="p">)</span> <span class="o">%</span> <span class="bp">self</span><span class="p">.</span><span class="n">pi_n</span> <span class="o">==</span> <span class="mi">1</span><span class="p">:</span>
<span class="k">return</span> <span class="n">i</span>
<span class="k">def</span> <span class="nf">public_key</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">e</span><span class="p">,</span> <span class="bp">self</span><span class="p">.</span><span class="n">n</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">private_key</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="n">d</span><span class="p">,</span> <span class="bp">self</span><span class="p">.</span><span class="n">n</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">encrypt</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">key</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">pow</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="n">key</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span> <span class="o">%</span> <span class="n">key</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">decrypt</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">c</span><span class="p">,</span> <span class="n">key</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">pow</span><span class="p">(</span><span class="n">c</span><span class="p">,</span> <span class="n">key</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span> <span class="o">%</span> <span class="n">key</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
</code></pre></div></div>
<p>Now we can create <code class="language-plaintext highlighter-rouge">RSA</code> class</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">rsa</span> <span class="o">=</span> <span class="n">RSA</span><span class="p">()</span>
</code></pre></div></div>
<p>Following should output <code class="language-plaintext highlighter-rouge">99</code> if our algorithm works and it does :)</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">rsa</span><span class="p">.</span><span class="n">decrypt</span><span class="p">(</span><span class="n">rsa</span><span class="p">.</span><span class="n">encrypt</span><span class="p">(</span><span class="mi">99</span><span class="p">,</span> <span class="n">rsa</span><span class="p">.</span><span class="n">public_key</span><span class="p">()),</span> <span class="n">rsa</span><span class="p">.</span><span class="n">private_key</span><span class="p">())</span>
</code></pre></div></div>
<h1 id="how-is-it-secure">How is it secure</h1>
<p>Whole security of public key encryption depends on that given a public key no one should be able to generate private key from that public key. Remember \(ed \equiv 1 \pmod{\phi(n)}\)? So, to calculate d from e attacker needs to know \(\phi(n)\). But only way to calculate that is \((p - 1)(q - 1)\). Only key generator knows \(p\) and \(q\). When \(p\) \(q\) are large enough no one can calculate (Yet!) \(p\), \(q\) from \(n\). The whole cryotosystem depends on this assumption.</p>
<h1 id="practical-usage">Practical Usage</h1>
<p>Public key cryptography is used everywhere. HTTPS run on public key cryptography (Not only RSA but it plays a huge role). If you need to use public key cryptography for your own application you can use openssl. openssl is a full featured toolkit which can generate RSA keys among lot of other things. To generate a key pair type following in command line (Assuming you are on UNIX based system)</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>openssl genpkey -algorithm RSA -out private_key.pem -pkeyopt rsa_keygen_bits:2048
</code></pre></div></div>
<p>You will generate <code class="language-plaintext highlighter-rouge">private_key.pem</code> file from above command. In this key there are lot of information encoded that it can be used to extract public key.</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>openssl rsa -pubout -in private_key.pem -out public_key.pem
</code></pre></div></div>
<p><code class="language-plaintext highlighter-rouge">public_key.pem</code> only contain details to encrypt or decrypt something so it can be shared. But you should never share your <code class="language-plaintext highlighter-rouge">private_key.pem</code></p>
<p>Now let’s encrypt some data using public key.</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>echo "reqviescat in constantia, ergo, repræsentatio cvpidi avctoris religionis" > key.bin
</code></pre></div></div>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>openssl rsautl -encrypt -inkey public_key.pem -pubin -in key.bin -out key.bin.enc
</code></pre></div></div>
<p>Encrypted data is written to <code class="language-plaintext highlighter-rouge">key.bin.enc</code>. Now let’s decrypt it again</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>openssl rsautl -decrypt -inkey private_key.pem -in key.bin.enc -out key.bin
</code></pre></div></div>
<p>If you open <code class="language-plaintext highlighter-rouge">key.bin</code> file you can see that same data is there.</p>
<h1 id="additional-resources">Additional Resources</h1>
<ul>
<li>Toy RSA Algorithm - <a href="https://github.com/chamoda/rsa-algorithm">Github</a></li>
<li>Original RSA Paper - <a href="https://people.csail.mit.edu/rivest/Rsapaper.pdf">A Method for Obtaining Digital Signatures and Public-Key Cryptosystems</a></li>
<li>Modular Arithmetic - <a href="https://en.wikipedia.org/wiki/Modular_arithmetic">Wikipedia</a></li>
<li>Fermat’s Little Theorem - <a href="https://en.wikipedia.org/wiki/Fermat%27s_little_theorem">Wikipedia</a></li>
<li>Coprime integers - <a href="https://en.wikipedia.org/wiki/Coprime_integers">Wikipedia</a></li>
<li>Euler’s totient function - <a href="https://en.wikipedia.org/wiki/Euler%27s_totient_function">Wikipedia</a></li>
<li>Modular multiplicative inverse - <a href="https://en.wikipedia.org/wiki/Modular_multiplicative_inverse">Wikipedia</a></li>
</ul>Public-key cryptography is one of the most used cryptosystems today. It refers to any system that uses a key pair, one for encrypting data and another one for decrypting data. If data encrypted using a key, other key is used to decrypt it. This seems pretty magical at first, but in the end of blog post you will understand how this works. In this blog post I’ll start with an analogy to understand what is the purpose of using two key pairs. Then I’ll explain the mathematical concepts behind the algorithm. Then I’ll implement a toy algorithm to understand it further (But never design your own crypto algorithms). Next I’ll explain some openssl commands to generate RSA public and private keys which you can use in real world applications.Real time face detection using OpenCV and Python2017-11-18T00:00:00+00:002017-11-18T00:00:00+00:00/2017/11/18/realtime-face-detection-using-opencv-and-python<p>Detection is an important application of computer vision. In this post I’m going to detail out how to do real time face detection using Viola Jones Algorithm introduced in paper <a href="http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.10.6807">Rapid object detection using a boosted cascade of simple features (2001)</a>. Mind the word detection, we are not going to recognize, means which one the face belong. This is merely detection that there is a face in a given image. We are going to use <a href="https://opencv.org">OpenCV</a> (Open Source Computer Vision Library). OpenCV is written in C++ but there are interfaces for other languages so will use, preferably python.</p>
<h1 id="installation">Installation</h1>
<p>Instructions are for OSX. Steps as follows</p>
<ul>
<li>First install <a href="https://www.anaconda.com/download/">Anaconda</a>.</li>
<li>Then create python 3 virtual environment with <code class="language-plaintext highlighter-rouge">conda create -n py3 python=3.6</code>.</li>
<li>Then type <code class="language-plaintext highlighter-rouge">source activate py3</code> which will activate python 3 environment.</li>
<li>Now install opencv with <code class="language-plaintext highlighter-rouge">conda install -c conda-forge opencv</code>.</li>
<li>Check the installation with <code class="language-plaintext highlighter-rouge">echo -e "import cv2; print(cv2.__version__)" | python</code> command. It should output <code class="language-plaintext highlighter-rouge">3.3.0</code></li>
<li>Make sure your system has <code class="language-plaintext highlighter-rouge">ffmpeg</code> installed which is required for reading a video stream from video formats like <code class="language-plaintext highlighter-rouge">mp4, mkv</code>. Using brew you can install ffmpeg with one command <code class="language-plaintext highlighter-rouge">brew install ffmpeg</code></li>
</ul>
<h1 id="time-for-action">Time for action</h1>
<p>Let’s do a quick implementation.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="kn">import</span> <span class="nn">cv2</span>
<span class="n">cap</span> <span class="o">=</span> <span class="n">cv2</span><span class="p">.</span><span class="n">VideoCapture</span><span class="p">(</span><span class="s">"input.mp4"</span><span class="p">)</span>
<span class="n">face_cascade</span> <span class="o">=</span> <span class="n">cv2</span><span class="p">.</span><span class="n">CascadeClassifier</span><span class="p">(</span><span class="s">'haarcascade_frontalface_default.xml'</span><span class="p">)</span>
<span class="k">while</span><span class="p">(</span><span class="bp">True</span><span class="p">):</span>
<span class="n">ret</span><span class="p">,</span> <span class="n">frame</span> <span class="o">=</span> <span class="n">cap</span><span class="p">.</span><span class="n">read</span><span class="p">()</span>
<span class="n">gray</span> <span class="o">=</span> <span class="n">cv2</span><span class="p">.</span><span class="n">cvtColor</span><span class="p">(</span><span class="n">frame</span><span class="p">,</span> <span class="n">cv2</span><span class="p">.</span><span class="n">COLOR_BGR2GRAY</span><span class="p">)</span>
<span class="n">faces</span> <span class="o">=</span> <span class="n">face_cascade</span><span class="p">.</span><span class="n">detectMultiScale</span><span class="p">(</span><span class="n">gray</span><span class="p">,</span> <span class="mf">1.3</span><span class="p">,</span> <span class="mi">5</span><span class="p">)</span>
<span class="k">for</span> <span class="p">(</span><span class="n">x</span><span class="p">,</span><span class="n">y</span><span class="p">,</span><span class="n">w</span><span class="p">,</span><span class="n">h</span><span class="p">)</span> <span class="ow">in</span> <span class="n">faces</span><span class="p">:</span>
<span class="n">cv2</span><span class="p">.</span><span class="n">rectangle</span><span class="p">(</span><span class="n">frame</span><span class="p">,</span> <span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">),</span> <span class="p">(</span><span class="n">x</span><span class="o">+</span><span class="n">w</span><span class="p">,</span> <span class="n">y</span><span class="o">+</span><span class="n">h</span><span class="p">),</span> <span class="p">(</span><span class="mi">0</span> <span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">255</span><span class="p">),</span> <span class="mi">2</span><span class="p">)</span>
<span class="n">cv2</span><span class="p">.</span><span class="n">imshow</span><span class="p">(</span><span class="s">'frame'</span><span class="p">,</span> <span class="n">frame</span><span class="p">)</span>
<span class="k">if</span> <span class="n">cv2</span><span class="p">.</span><span class="n">waitKey</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span> <span class="o">&</span> <span class="mh">0xFF</span> <span class="o">==</span> <span class="nb">ord</span><span class="p">(</span><span class="s">'q'</span><span class="p">):</span>
<span class="k">break</span>
<span class="n">cap</span><span class="p">.</span><span class="n">release</span><span class="p">()</span>
<span class="n">cv2</span><span class="p">.</span><span class="n">destroyAllWindows</span><span class="p">()</span>
</code></pre></div></div>
<p>We are going to read a video file frame by frame, applying Viola Jones algorithm with trained parameters then apply a rectangle layer if a face found, displaying modified output frame by frame.</p>
<p><code class="language-plaintext highlighter-rouge">cap = cv2.VideoCapture("input.mp4")</code> imports the video. You can also use a web cam instead of video, just pass <code class="language-plaintext highlighter-rouge">0</code> as parameter like <code class="language-plaintext highlighter-rouge">cap = cv2.VideoCapture(0)</code>.</p>
<p>Next we are going to load the trained model. <code class="language-plaintext highlighter-rouge">haarcascade_frontalface_default.xml</code> is a model already trained using lot of faces and non faces and lot of computing power. Training good models sometime takes days not hours. Thankfully above model is trained by Intel using lot of data. The model we are using here comes with the installation of OpenCv. Generally you can find those models at your <code class="language-plaintext highlighter-rouge">opencv-installation-directory/share/OpenCV/haarcascades</code> but this could differ depends on your OS and installation method.</p>
<p>Next in the <code class="language-plaintext highlighter-rouge">while</code> loop we are reading frame by frame. Then get given frame and convert into gray scale.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">gray</span> <span class="o">=</span> <span class="n">cv2</span><span class="p">.</span><span class="n">cvtColor</span><span class="p">(</span><span class="n">frame</span><span class="p">,</span> <span class="n">cv2</span><span class="p">.</span><span class="n">COLOR_BGR2GRAY</span><span class="p">)</span>
</code></pre></div></div>
<p>A frame is a array of 3 matrices where each matrix is for the respective color blue, green, red. Did you noticed the revered order? In OpenCV default representation is BGR not RGB. In the above step we are converting it to gray scale image. <code class="language-plaintext highlighter-rouge">gray</code> is a single matrix now.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">faces</span> <span class="o">=</span> <span class="n">face_cascade</span><span class="p">.</span><span class="n">detectMultiScale</span><span class="p">(</span><span class="n">gray</span><span class="p">,</span> <span class="mf">1.3</span><span class="p">,</span> <span class="mi">5</span><span class="p">)</span>
</code></pre></div></div>
<p><code class="language-plaintext highlighter-rouge">detectMultiScale</code> function detect faces and return an array of position coordinates and sizes. Second parameter is <code class="language-plaintext highlighter-rouge">scaleFactor</code>. To reduce true negatives you must use a value near to zero. Basically the algorithm can only detect it’s trained size usually around 20x20 pixel. To detect large object area get scaled by <code class="language-plaintext highlighter-rouge">scaleFactor</code>. If <code class="language-plaintext highlighter-rouge">scaleFactor</code> is \(1.05\) scaled block size would equal to \(20 \times 1.05 = 21\). If <code class="language-plaintext highlighter-rouge">scaleFactor</code> is equal to 1.3 in above image scaled block size is \(20 \times 1.3 = 26\) so you may miss some pixels, so do some faces. Trade off is accuracy vs performance.</p>
<p>3rd parameter is <code class="language-plaintext highlighter-rouge">minNeighbors</code>. It defines how many neighbor rectangles should identified to retain it. Higher value means less false positives.</p>
<p>Next we draw red rectangles if faces detected. Notice how we passed red color in BGR format <code class="language-plaintext highlighter-rouge">(0 , 0, 255)</code>.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">for</span> <span class="p">(</span><span class="n">x</span><span class="p">,</span><span class="n">y</span><span class="p">,</span><span class="n">w</span><span class="p">,</span><span class="n">h</span><span class="p">)</span> <span class="ow">in</span> <span class="n">faces</span><span class="p">:</span>
<span class="n">cv2</span><span class="p">.</span><span class="n">rectangle</span><span class="p">(</span><span class="n">frame</span><span class="p">,</span> <span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">),</span> <span class="p">(</span><span class="n">x</span><span class="o">+</span><span class="n">w</span><span class="p">,</span> <span class="n">y</span><span class="o">+</span><span class="n">h</span><span class="p">),</span> <span class="p">(</span><span class="mi">0</span> <span class="p">,</span> <span class="mi">0</span><span class="p">,</span> <span class="mi">255</span><span class="p">),</span> <span class="mi">2</span><span class="p">)</span>
</code></pre></div></div>
<p>Next we are writing to a window frame by frame. If you press key <code class="language-plaintext highlighter-rouge">q</code> the loop will break and video will stop. I’ve created a quick video from the output here. Notice how it detects only frontal faces, because model is trained for frontal faces.</p>
<iframe width="100%" height="450" src="https://www.youtube.com/embed/eAJwY8fQvXs" frameborder="0" allowfullscreen=""></iframe>
<h1 id="how-it-works">How it works</h1>
<p>Viola Jones algorithm is a machine learning algorithm developed by Paul Viola and Michael Jones in 2001. It was designed to be very fast even to be possible in embedded systems. We can break down it to 4 parts</p>
<ul>
<li>Haar like features</li>
<li>Integral image</li>
<li>AdaBoost (Adaptive Boosting)</li>
<li>Cascading</li>
</ul>
<h1 id="haar-like-features">Haar like features</h1>
<p>Haar like features, named after the Hungarian mathematician Alfred Haar is a way of identifying features in a image in a more abstract way.</p>
<p><img src="/assets/posts/realtime-face-detection-using-opencv-and-python/haar.png" alt="Haar" /></p>
<p>To calculate a feature following equation is used.</p>
\[Value = \text{(Sum of pixels in black area)} - \text{(Sum of pixels in white area)}\]
<p>In the case of face detection following feature will give a higher value in the positioned area, that defines a feature.</p>
<p><img src="/assets/posts/realtime-face-detection-using-opencv-and-python/face.png" alt="Face" /></p>
<p>Eye area in the are generally darker than under the eye, so \(\text{(Black area - White area)}\) will give higher value. And that’s defines single feature.</p>
<h1 id="integral-image">Integral Image</h1>
<p>Intergral image is a way to calculate rectangle features quickly</p>
\[ii(x, y) = \sum_{x{'} \le x, y{'} \le y} i(x', y')\]
<p>Every position is sum of the top and left values. Note this code is written for clarity, there are more efficeint way to write this. In this case you will be feeding a normalized image to the fuction. Purpose of normaizing first is to get rid of conditions of lighting effects.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">def</span> <span class="nf">my_intergral_image</span><span class="p">(</span><span class="n">img</span><span class="p">):</span>
<span class="n">intergral_img</span> <span class="o">=</span> <span class="n">np</span><span class="p">.</span><span class="n">zeros</span><span class="p">(</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">)</span>
<span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]):</span>
<span class="k">for</span> <span class="n">y</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]):</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">x</span> <span class="o">+</span> <span class="mi">1</span><span class="p">):</span>
<span class="k">for</span> <span class="n">j</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">y</span> <span class="o">+</span> <span class="mi">1</span><span class="p">):</span>
<span class="n">intergral_img</span><span class="p">[</span><span class="n">y</span><span class="p">,</span> <span class="n">x</span><span class="p">]</span> <span class="o">+=</span> <span class="n">img</span><span class="p">[</span><span class="n">i</span><span class="p">,</span> <span class="n">j</span><span class="p">]</span>
<span class="n">zero_padded_intergral_image</span> <span class="o">=</span> <span class="n">np</span><span class="p">.</span><span class="n">zeros</span><span class="p">((</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="mi">1</span><span class="p">))</span>
<span class="n">zero_padded_intergral_image</span><span class="p">[</span><span class="mi">1</span><span class="p">:</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">+</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">:</span><span class="n">img</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="mi">1</span><span class="p">]</span> <span class="o">=</span> <span class="n">intergral_img</span>
<span class="k">return</span> <span class="n">zero_padded_intergral_image</span>
</code></pre></div></div>
<p>Whole purpose of integral image is to calulate sum of pixels inside a given rectangle fast.</p>
<p><img src="/assets/posts/realtime-face-detection-using-opencv-and-python/sum.png" alt="Intergral sum" /></p>
<p>We knows values p, q, r, s in integral image and they represnt the sum of all the left and top values. Our goal is to find pixel sum of D.</p>
\[p = A \\
q = A + B \\
r = A + C \\
s = A + B + C + D \\\]
<p>So \(D = (p + s) - (q + r)\) which will reduce computation steps to calulate sum of pixels in defined rectangle significantly.</p>
<p>Here’s the equalant python code.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">def</span> <span class="nf">sum_region</span><span class="p">(</span><span class="n">integral_img</span><span class="p">,</span> <span class="n">top_left</span><span class="p">,</span> <span class="n">bottom_right</span><span class="p">):</span>
<span class="c1"># to numpy matrix notation
</span> <span class="n">top_left</span> <span class="o">=</span> <span class="p">(</span><span class="n">top_left</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">top_left</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="n">bottom_right</span> <span class="o">=</span> <span class="p">(</span><span class="n">bottom_right</span><span class="p">[</span><span class="mi">1</span><span class="p">],</span> <span class="n">bottom_right</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
<span class="k">if</span> <span class="n">top_left</span> <span class="o">==</span> <span class="n">bottom_right</span><span class="p">:</span>
<span class="k">return</span> <span class="n">integral_img</span><span class="p">[</span><span class="n">top_left</span><span class="p">]</span>
<span class="n">top_right</span> <span class="o">=</span> <span class="p">(</span><span class="n">bottom_right</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">top_left</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
<span class="n">bottom_left</span> <span class="o">=</span> <span class="p">(</span><span class="n">top_left</span><span class="p">[</span><span class="mi">0</span><span class="p">],</span> <span class="n">bottom_right</span><span class="p">[</span><span class="mi">1</span><span class="p">])</span>
<span class="c1"># s + p - (q + r)
</span> <span class="k">return</span> <span class="n">integral_img</span><span class="p">[</span><span class="n">bottom_right</span><span class="p">]</span> <span class="o">+</span> <span class="n">integral_img</span><span class="p">[</span><span class="n">top_left</span><span class="p">]</span> <span class="o">-</span> <span class="n">integral_img</span><span class="p">[</span><span class="n">top_right</span><span class="p">]</span> <span class="o">-</span> <span class="n">integral_img</span><span class="p">[</span><span class="n">bottom_left</span><span class="p">]</span>
</code></pre></div></div>
<h1 id="adaboost-adaptive-boosting">AdaBoost (Adaptive Boosting)</h1>
<p>Now we need a way to select best features from all possible features that can correctly classify a face. AdaBoost algorithm was formulated by Yoav Freund and Robert Schapire in 1997 and won the prestigious <a href="https://en.wikipedia.org/wiki/G%C3%B6del_Prize">Gödel Prize</a> in 2003. This elegant machine learning approach can be applied to wide range of problems not only image detection.</p>
<p>I’ll start with the idea of weak and strong classifiers.</p>
<ul>
<li>Weak Classifier - Classifier that’s little bit better than random guessing.</li>
<li>Strong Classifier - A combination of weak classifiers. It represents wisdom of a weighted crowd of experts.</li>
</ul>
<p>Adaboost itself is a inhertantly incomplete algorithm, so it’s called a meta algorithm. Let’s express the idea of strong classifier mathematically.</p>
\[H(x) = sign\Big( h_1(x) + h_2(x) + ... +h_T(x)\Big)\]
<p>H(x) is a strong classifier which classify according to the sign of the sum of weak classifiers. Every weak classifier is a Haar feature combined with some other parameters I’ll detail out in next few paragraphs. Suppose there are only 3 weak classifiers so \(T = 3\). Every classifier outputs +1 or -1 so sum of weak classifier will have either + or - sign.</p>
\[H(x) = sign\Big( +1 + -1 + -1 \Big)\]
<p>Here strong classifier sign is negative so it may not be a face. I started with this analogy but we have to also weight the weak classifiers because some classifiers may have strong influence than others.</p>
<p>By adding weights strong classifier get little complicated but it’s nothing more than a simple inequality match.</p>
\[H(x) = \begin{cases}
+1, \text{if}\ \sum_{t = 1}^{T} \alpha_th_t(x) \ge \frac{1}{2} \sum_{t = 1}^{T} \alpha_t \\
-1, \text{otherwise}
\end{cases}\]
<p>Now how we decide which weak classifier to use, which \(\alpha\) weights to use? That’s why we need to train over existing labeled data. Suppose image is \(x_i\) and label is \(y_i\) which is 1 for face and 0 for non face. Given example images \((x_1, y_1), ... , (x_m, y_m)\) we will initialize weights for each example. Don’t confuse this weight with the \(\alpha\) weight discussed before. This algorithm has two kind of weights. One is \(\alpha\) weight for selected weak classifier. Other one is for each example and each step denoted by \(w_{t,i}\). Now we are going to intialize weights mentioned later for step one \(t = 1\)</p>
\[w_{1, i} = \frac{1}{m}\]
<p>Here we have normalized the weights so sum of the weights is 1. We do the normalize in every step to make sure weight distribution is always adds up to 1.</p>
\[\sum_{i = 1}^{m} w_{t, i} = 1\]
<p>The generalized normalized equation is</p>
\[w_{t,i} = \frac{w_{t,i}}{\sum_{j = 0}^{m}w_{t,i}}\]
<p>Next we loop over all the features to select the best weak classifier which minimize error rate.</p>
\[\epsilon_t = min_{f, p, \theta} \sum_{i = 1}^m w_{t, i} | h(x_i, f, p, \theta) - y_i |\]
<p>Here calculating the sum of weights of misclassified examples which is the error rate represented by epsilon \(\epsilon\). Notice \(h(x_i, f, p, \theta) - y_i\) return 1 or -1 if the misclassified, 0 is correctly classified. We are taking the absolute value out of it so weight get multiplied by 1 if misclassified. Let’s dive into the definition of weak classifier.</p>
\[h(x) = h(x_i, f, p, \theta)\]
<p>\(x_i\) is the \(i\)‘th image example. \(f\) is the Haar feature. \(p\) is the polarity which is either -1 or +1 which defines direction of the inequality. \(\theta\) is the threshold.</p>
\[h(x, f, p , \theta) = \begin{cases}
1, \text{if } pf(x) \lt p\theta \\
0, \text{otherwise}
\end{cases}\]
<p>To select the \(f\) feature we need loop over f(x) Haar classifier. To select the threshold we need to minimize the following equation.</p>
\[error_{\theta} = min\Big( (S_+) + (T_-) - (S_-), (S_-) + (T_+) - (S_+))\Big)\]
<p>Here’s the definition of symbols</p>
<ul>
<li>\(T_+\) is total sum of positive sample weights</li>
<li>\(T_-\) is total sum of negetive sample weights</li>
<li>\(S_+\) is sum of positive sample weights below threshold</li>
<li>\(S_-\) is sum of negetive sample weights below threshold</li>
</ul>
<p>After finding the minized error \(\epsilon_t\) for step \(t\) we can find the weights for the next step \(t + 1\)</p>
\[w_{t+1, i} = w_{t, i} \Big(\frac{\epsilon_t}{1 - \epsilon_t}\Big)^{1 -e_i}\]
<p>Where \(e_i = 0\) if sample image \(x_i\) is correctly classified, 0 otherwise. The goal of the equation is to make the weights of incorrectly classified samples slightly larger so in the next round will be unforgiving to weak classifiers that’s going to classify same samples incorrectly. So in each step it’s going choose an unique weak classifier with unique feature. To do so in the above equation correctly classified weights will decreased so misclassified weights will increase relative to correct weights.</p>
<p>Finally calculate \(\alpha_t\) for the selected classifier.</p>
\[\alpha_t = log(\frac{1 - \epsilon_t}{\epsilon_t})\]
<p>Notice how \(\alpha\) going to be a higher value if error rate is small so that weak classifier has more contribution to strong classifier.</p>
<p>So finally algorithm need to loop \(T\) steps to find \(T\) classifiers to get good results.</p>
<h1 id="cascading">Cascading</h1>
<p>You may have noticed how many loops we have in the algorithm so this AdaBoost along with Haar Features is computationally expensive for real time detection. So we are using attentional cascade to reduce some unnecessary computations. More efficient cascade can be constructed so that negative sub windows will rejected early. Every stage of cascade is a strong classifier, so all the features are grouped into several stages where each stage has a several number of features.</p>
<p><img src="/assets/posts/realtime-face-detection-using-opencv-and-python/cascade.png" alt="Cascade" /></p>
<p>For the cascade we need following parameters</p>
<ul>
<li>Number of stages in cascade</li>
<li>Number of features in each cascade</li>
<li>Threshold of each strong classifier</li>
</ul>
<p>Finding optimum values for above parameters is a difficult task. Viola Jones introduced a simple method to find the optimum combination.</p>
<ul>
<li>Select \(f_i\) the maximum acceptable false positive rate per stage</li>
<li>Select \(d_i\) the maximum acceptable true positive rate per stage</li>
<li>Select \(F_{target}\) Overall false positive rate</li>
</ul>
<p>Now we are looping until pre defined \(F_{target}\) is met by adding new stages. In stages we keep adding features until \(f_i\) and \(d_i\) is met. By doing this we are going to create a cascade of strong classifiers.</p>
<h1 id="additional-resources">Additional Resources</h1>
<ul>
<li>Paper (Revised) <a href="http://www.vision.caltech.edu/html-files/EE148-2005-Spring/pprs/viola04ijcv.pdf">Viola Jones 2001</a>.</li>
<li>Viola Jones Python Implementation <a href="https://github.com/Simon-Hohberg/Viola-Jones">Github</a></li>
<li>To learn more about AdaBoost read the book <a href="https://www.amazon.com/Boosting-Foundations-Algorithms-Adaptive-Computation/dp/0262526034">Boosting Foundations and Algorithms</a>. It’s written by the original authors of the algorithm.</li>
<li>Source code for the face detection in this post - <a href="https://github.com/chamoda/realtime-face-detection">Realtime face detection</a></li>
<li>Pull requests are welcome if you find anything wrong in the post - <a href="https://github.com/chamoda/chamoda.github.io">Blog</a></li>
</ul>Detection is an important application of computer vision. In this post I’m going to detail out how to do real time face detection using Viola Jones Algorithm introduced in paper Rapid object detection using a boosted cascade of simple features (2001). Mind the word detection, we are not going to recognize, means which one the face belong. This is merely detection that there is a face in a given image. We are going to use OpenCV (Open Source Computer Vision Library). OpenCV is written in C++ but there are interfaces for other languages so will use, preferably python.Introduction to Machine Learning with Linear Regression2017-10-05T00:00:00+00:002017-10-05T00:00:00+00:00/2017/10/05/introduction-to-machine-learing-with-linear-regression<p>Machine learning is all about computers learning itself from data to predict new data. There are two types of categories in machine learning</p>
<ul>
<li>Supervised Learning</li>
<li>Unsupervised Learning</li>
</ul>
<p>In supervised learning computer learns from input and output while creating a general model which can be used to predict output from new input. Unsupervised learning is about discovering pattern in data without knowing explicit labels beforehand. In this post I’m going to talk about linear regression which is a supervised learning method.</p>
<h1 id="jupyter-notebook">Jupyter Notebook</h1>
<p>Jupyter Notebook is an interactive python environment. You can install jupyter notebook with <a href="https://www.anaconda.com/download">Anaconda</a> which will also install all necessary packages. After installation just run <code class="language-plaintext highlighter-rouge">jupyter notebook</code> in command line, which will open jupyter notebook instance in browser. Click <code class="language-plaintext highlighter-rouge">New</code> and create a Python 2 notebook.</p>
<p>Enter following lines of code in the cell and press <code class="language-plaintext highlighter-rouge">Shift + Enter</code> which will execute the code. Full notebook of the code I used in this post is also available on <a href="https://github.com/chamoda/LinearRegression">github</a>.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="n">np</span>
<span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="n">plt</span>
</code></pre></div></div>
<p><a href="http://www.numpy.org/">Numpy</a> is used for matrix operations. It’s the most important library in python scientific computing echo system.</p>
<p><a href="https://matplotlib.org/">Matplotlib</a> is a plotting library. We will use it to visualize our data.</p>
<h1 id="data">Data</h1>
<p>In this post I’m not going to use any real data set. Instead I’m going to generate some random data.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">rng</span> <span class="o">=</span> <span class="n">np</span><span class="p">.</span><span class="n">random</span><span class="p">.</span><span class="n">RandomState</span><span class="p">(</span><span class="mi">42</span><span class="p">)</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">rng</span><span class="p">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">100</span><span class="p">)</span> <span class="o">*</span> <span class="mi">10</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">x</span><span class="p">[:,</span> <span class="n">np</span><span class="p">.</span><span class="n">newaxis</span><span class="p">]</span>
<span class="n">b</span> <span class="o">=</span> <span class="n">rng</span><span class="p">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">100</span><span class="p">)</span> <span class="o">*</span> <span class="mi">5</span>
<span class="n">b</span> <span class="o">=</span> <span class="n">b</span><span class="p">[:,</span> <span class="n">np</span><span class="p">.</span><span class="n">newaxis</span><span class="p">]</span>
<span class="n">y</span> <span class="o">=</span> <span class="mi">3</span> <span class="o">*</span> <span class="n">x</span> <span class="o">+</span> <span class="n">b</span>
</code></pre></div></div>
<p>Here we generate 100 rows of random data we will say \(m = 100\). \(m\) is usually used in machine learning to denote number of data, in this case pairs of \(x\) and \(y\). Now we are going to plot the dataset using Matplotlib.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">plt</span><span class="p">.</span><span class="n">scatter</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="n">plt</span><span class="p">.</span><span class="n">xlabel</span><span class="p">(</span><span class="s">'x axis'</span><span class="p">)</span>
<span class="n">plt</span><span class="p">.</span><span class="n">ylabel</span><span class="p">(</span><span class="s">'y axis'</span><span class="p">)</span>
<span class="n">plt</span><span class="p">.</span><span class="n">show</span><span class="p">()</span>
</code></pre></div></div>
<p>This is the plot generate from the code. Every blue dot is a data point. There are 100 blue dots here.</p>
<p><img src="/assets/posts/linear-regression/plot.png" alt="plot" /></p>
<p>Data is distributed in sort of linear nature because of the way we generate random data.</p>
<h1 id="linear-regression">Linear Regression</h1>
<p>So purpose of the linear regression is to find the equation of a line that do justice to all data points. We will define the equation of the line as \(h(x) = \theta_0 + \theta_1x\). In machine learning \(h(x)\) is called the hypothesis function. But this is just a fancy way of writing old school \(y = mx + c\) where \(c = \theta_0\) and \(m = \theta_1\). Now our goal is to find find \(\theta_0\) and \(\theta_1\). If we know the \(\theta_0\) and \(\theta_1\) we can draw the line. To find \(\theta_0\) and \(\theta_1\) we are going to use a function called cost function.</p>
<h1 id="cost-function">Cost Function</h1>
<p>We going to use following equation which is called least square method.</p>
\[J(\theta_0, \theta_1) = \frac{1}{2m}\sum_{i=1}^m(h(x^{(i)}) - y^{(i)})^2\]
<p>Goal is to minimize \((h(x) - y)\) so difference between the hypothesis function output and \(y\) is minimized as possible. Suppose \(\sum_{i=1}^m(h(x^{(i)}) - y^{(i)})^2\) is \(0\), that means \(J(\theta_0, \theta_1) = 0\) so data will fit perfectly to the strait line \(h(x) = \theta_0 + \theta_1x\). But for a real dataset that may be not the case. If data has spread all over the plot this value may be hight. Ok, you get now what \((h(x) - y)\) means but why squaring it? We sqaure it because there could be negetive or positive values for \(h(x^{(i)}) - y^{(i)}\) depending on the the way data is distributetd when using range of values for \(\theta_0\) and \(\theta_1\). By squaring we make sure there is no negetive values so their is no cancelling out situations. By that means \(0\) realy means that it’s fit the data distrubbution. In a real world situation if data is not perfectly linear we can’t get \(J(\theta_0, \theta_1)\) to zero but try to get a minimum value possible. We say this as ‘minimizing the cost function’. We divide by \(m\) get a relatively small value. We devided by 2 because, by a future derivative operation anther function will a be much simpler equation.</p>
<p>Now we are going to implement the cost function in python. First we implement the hypothesis function \(h(x) = \theta_0 + \theta_1x\).</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">def</span> <span class="nf">h</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">theta0</span><span class="p">,</span> <span class="n">theta1</span><span class="p">):</span>
<span class="k">return</span> <span class="n">theta0</span> <span class="o">+</span> <span class="n">theta1</span> <span class="o">*</span> <span class="n">x</span>
</code></pre></div></div>
<p>If we pass normal python integer values to <code class="language-plaintext highlighter-rouge">x</code>, <code class="language-plaintext highlighter-rouge">theta0</code>, <code class="language-plaintext highlighter-rouge">theta1</code> it will output an integer. What will happen if we change <code class="language-plaintext highlighter-rouge">x</code> to a numpy array. All values get multiplied by <code class="language-plaintext highlighter-rouge">theta1</code>, then all values will get added by <code class="language-plaintext highlighter-rouge">theta0</code>. Output will be a numpy array. This happens because of a python feature called broadcasting. It will save us from writing a for loop to calculate all the values step by step. Broadcastng is a very powerpull concept. Use of broadcasting will be used everywhere when doing machine learning in python. Next we implement cost function in python</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">def</span> <span class="nf">cost</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">theta0</span><span class="p">,</span> <span class="n">theta1</span><span class="p">):</span>
<span class="k">return</span> <span class="n">np</span><span class="p">.</span><span class="nb">sum</span><span class="p">(</span><span class="n">np</span><span class="p">.</span><span class="n">power</span><span class="p">(</span><span class="n">h</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">theta0</span><span class="p">,</span> <span class="n">theta1</span><span class="p">)</span> <span class="o">-</span> <span class="n">y</span><span class="p">,</span> <span class="mi">2</span><span class="p">))</span> <span class="o">*</span> <span class="mi">1</span> <span class="o">/</span> <span class="mi">2</span> <span class="o">*</span> <span class="n">x</span><span class="p">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
</code></pre></div></div>
<p><code class="language-plaintext highlighter-rouge">np.sum()</code> function take sum of all the values inside a numpy array and return a scaler value. <code class="language-plaintext highlighter-rouge">np.power()</code> take power, in this case square of all the values and retrun a same size array with modified values. we pass a numpy array as <code class="language-plaintext highlighter-rouge">x</code> so <code class="language-plaintext highlighter-rouge">x.shape[0]</code> is number of elements in <code class="language-plaintext highlighter-rouge">x</code> which is also equal to \(m\).</p>
<p>Now we need to find <code class="language-plaintext highlighter-rouge">theta0</code> and <code class="language-plaintext highlighter-rouge">theta1</code> which minimize the cost function. One approach is to brute force a range of <code class="language-plaintext highlighter-rouge">theta0</code> and <code class="language-plaintext highlighter-rouge">theta1</code> and pick values where return value of cost function is minimal. But can we to better?</p>
<h1 id="gradient-descent">Gradient Descent</h1>
<p>Gradient descent is an iterative algorithm to find out the minimum of a function. This is what we are going to do.</p>
\[\theta_0 := \theta_0 - \alpha \frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1)\]
\[\theta_1 := \theta_1 - \alpha \frac{\partial}{\partial \theta_1} J(\theta_0, \theta_1)\]
<p>I will explain step by step. First \(:=\) is the assignment operator. In programming languages we use \(=\) operator but in math it means equality. So \(:=\) means assignment in mathematics. \(\alpha\) is called the learning rate which I will explain later in detail. \(\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1)\) is called the derivative part.</p>
\[\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1) = \frac{\partial}{\partial \theta_0} \frac{1}{2m}\sum_{i=1}^m(h(x^{(i)}) - y^{(i)})^2\]
<p>This is what the plot looks with \(\theta_0\) against cost function like after 100,000 iterations.</p>
<p><img src="/assets/posts/linear-regression/theta0_cost.png" alt="plot" /></p>
<p>Derivative is measuring the slope, You can see the slope is a negative value as the plot going down. \(\alpha\) is the learning rate which is \(0.0001\). So</p>
\[\theta_0 := \theta_0 - \alpha \frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1)\]
<p>You can see \(\theta_0\) will increase because \(\alpha\) is positive \(\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1)\) is negative. Rate of change in lope is decreasing so \(\theta_0\) will become a stabilized value.</p>
<p>Here’s the derivative steps for \(\theta_0\).</p>
\[\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1) = \frac{1}{2m} \frac{\partial}{\partial \theta_0} \sum_{i=1}^m(h(x^{(i)}) - y^{(i)})^2\]
<p>I’ve taken out the \(\frac{1}{2m}\) out of the derivative. Since \(h(x^{(i)}) = \theta_0 + \theta_1x^{(i)}\) we can write</p>
\[\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1) = \frac{1}{2m} \frac{\partial}{\partial \theta_0} \sum_{i=1}^m(\theta_0 + \theta_1x^{(i)} - y^{(i)})^2\]
<p>Now we are ready to take the derivative.</p>
\[\frac{\partial}{\partial \theta_0} J(\theta_0, \theta_1) = \frac{1}{m} \sum_{i=1}^m(\theta_0 + \theta_1x^{(i)} - y^{(i)})\]
<p>Next derivative steps for \(\theta_1\).</p>
\[\frac{\partial}{\partial \theta_1} J(\theta_0, \theta_1) = \frac{1}{2m} \frac{\partial}{\partial \theta_1} \sum_{i=1}^m(h(x^{(i)}) - y^{(i)})^2\]
\[\frac{\partial}{\partial \theta_1} J(\theta_0, \theta_1) = \frac{1}{2m} \frac{\partial}{\partial \theta_1} \sum_{i=1}^m(\theta_0 + \theta_1x^{(i)} - y^{(i)})^2\]
<p>After taking the derivative,</p>
\[\frac{\partial}{\partial \theta_1} J(\theta_0, \theta_1) = \frac{1}{m} \sum_{i=1}^m(\theta_0 + \theta_1x^{(i)} - y^{(i)})x^{(i)}\]
<p>Now let’s look at the equivalent python code.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">theta0</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">theta1</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">alpha</span> <span class="o">=</span> <span class="mf">0.001</span>
<span class="n">iteration</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">while</span> <span class="bp">True</span><span class="p">:</span>
<span class="n">temp0</span> <span class="o">=</span> <span class="p">(</span><span class="mi">1</span> <span class="o">/</span> <span class="n">m</span><span class="p">)</span> <span class="o">*</span> <span class="n">np</span><span class="p">.</span><span class="nb">sum</span><span class="p">(</span><span class="n">h</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">theta0</span><span class="p">,</span> <span class="n">theta1</span><span class="p">)</span> <span class="o">-</span> <span class="n">y</span><span class="p">)</span>
<span class="n">temp1</span> <span class="o">=</span> <span class="p">(</span><span class="mi">1</span> <span class="o">/</span> <span class="n">m</span><span class="p">)</span> <span class="o">*</span> <span class="n">np</span><span class="p">.</span><span class="nb">sum</span><span class="p">((</span><span class="n">h</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">theta0</span><span class="p">,</span> <span class="n">theta1</span><span class="p">)</span> <span class="o">-</span> <span class="n">y</span><span class="p">)</span> <span class="o">*</span> <span class="n">x</span><span class="p">)</span>
<span class="n">theta0</span> <span class="o">=</span> <span class="n">theta0</span> <span class="o">-</span> <span class="n">alpha</span> <span class="o">*</span> <span class="n">temp0</span>
<span class="n">theta1</span> <span class="o">=</span> <span class="n">theta1</span> <span class="o">-</span> <span class="n">alpha</span> <span class="o">*</span> <span class="n">temp1</span>
<span class="n">iteration</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">if</span> <span class="n">iteration</span> <span class="o">></span> <span class="mi">100000</span><span class="p">:</span>
<span class="k">print</span><span class="p">(</span><span class="s">"theta0: "</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="n">theta0</span><span class="p">)</span> <span class="o">+</span> <span class="s">"theta1:"</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="n">theta1</span><span class="p">))</span>
<span class="k">break</span>
</code></pre></div></div>
<p>We are initializing <code class="language-plaintext highlighter-rouge">theta0</code> and <code class="language-plaintext highlighter-rouge">theta1</code> to 0. Learning rate <code class="language-plaintext highlighter-rouge">alpha</code> to be <code class="language-plaintext highlighter-rouge">0.001</code>. You must choose a smaller value unless it may not going to converge smoothly. Also note that you must use temporary values while assigning.</p>
<p>Now that we have found optimal <code class="language-plaintext highlighter-rouge">theta0</code> and <code class="language-plaintext highlighter-rouge">theta1</code> values below we draw a line in top of the first plot.</p>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>theta0: 2.567988282theta1:2.98323417806
</code></pre></div></div>
<p><img src="/assets/posts/linear-regression/final.png" alt="plot" /></p>
<p>You can see that algorithm learned to draw the most optimal line over the data distribution.</p>
<h1 id="prediction">Prediction</h1>
<p>Here comes the easy part. Now we have a model with learned parameters which can be used to predict. We can use our hypothesis function.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">y</span> <span class="o">=</span> <span class="n">h</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="n">theta1</span><span class="p">,</span> <span class="n">theta2</span><span class="p">)</span>
</code></pre></div></div>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>11.517690816184686
</code></pre></div></div>
<h1 id="scikit-learn">Scikit Learn</h1>
<p>Scikit Learn is a python library with one of the most used classical machine learning algorithms. Now you know the what’s happens under the hood of linear regression function. So you are ready to use a library to do production level linear regression machine learning.</p>
<p>Here’s how you import the library</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="kn">from</span> <span class="nn">sklearn.linear_model</span> <span class="kn">import</span> <span class="n">LinearRegression</span>
</code></pre></div></div>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">model</span> <span class="o">=</span> <span class="n">LinearRegression</span><span class="p">()</span>
<span class="n">model</span><span class="p">.</span><span class="n">fit</span><span class="p">(</span><span class="n">x</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
</code></pre></div></div>
<p><code class="language-plaintext highlighter-rouge">fit()</code> function train the data.</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">model</span><span class="p">.</span><span class="n">predict</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
</code></pre></div></div>
<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>11.51769082
</code></pre></div></div>
<p>You can see that values are similar to our own implementation.</p>
<p>Github repo of example code is available <a href="https://github.com/chamoda/linear-regression">here</a></p>Machine learning is all about computers learning itself from data to predict new data. There are two types of categories in machine learning