#techbuilders

Smart Contracts. Smarter Results. Our expert Blockchain Development Services are designed to create, test, and deploy secure smart contracts that align with your business goals.

Whether you're building a DeFi app, launching a DAO, or need NFT utilities—we’ve got you.

Building on our attempt to conquer machine learning, we will add more complicated tools to our repertoire. We will start off regression with simple linear regression. Simple linear regression helps us predict the value of one variable based on the value of another variable. The variable we are predicting is usually plotted on the Y axis and the variable driving the prediction is plotted on the X axis.

Let's take an example problem of predicting a wife's age given the husband's age. Just as in the case of Bayesian reasoning, we will start off with a previous sampling:

Husband : 36 72 37 36 51 50 47 50 37 41 Wife        : 35 67 33 35 50 46 47 42 36 41

Visualising this in a scatter plot:

Hmmm, seems like the two variables show a linear relationship. Pearson's Co-relation (r) gives us an estimate of how "linear" this relationship is. r ranges from -1 to +1. A value of -1 implies a negative linear relationship, 0 implies NO linear relationship and +1 implies a positive linear relationship. The r value for this sampling is 0.97, and it does fall in line with our observation that the 2 ages have a positive linear relationship. (I have left out the r formula and calculation but encourage everyone to learn about r and do the calculation)

Now, given a new value for X (husband's age) how do we predict the value for Y (wife's age)? Since we are anyway assuming a linear relationship, will we be able to fit a line which roughly defines this relationship? Linear regression gives us a nice way to come up with an equation for such a line which has the least errors in "prediction".

Let us first derive the equation of this "best-fitting" line and then see what least errors in prediction means. The line equation is given by Y = bX + A. The slope b is given by r * (std_dev(X) / std_dev(Y)) The intercept A is given by Mean(Y) - b * Mean(X)

Following the formula, the equation of the regression line Y = 0.81X + 6.183. Now, plotting this line along with the scatter plot, we can define "error" in prediction.

It is simply the difference between the actual Y value and the predicted value Y' which lies on the regression line. This regression line is best-fitting since it has the least sum of squared errors in prediction.

The Bangalore Random Hacks of Kindness is a community built with the principles of innovation and openness. We believe technology has a huge impact in solving humanitarian Issues. Hence, twice a year experts from development areas pitch real world problems to the hackers/tech­experts after which hackers form teams choose their problems and hack over the weekend to create a proof of concept to solve part of the problem with technology.

The June 2014 RHoK event was held on 28 ­& 29th. The event this time was hosted in partnership with a set of communities like Beaglesloft, Arduino India , Breakthrough and Saakshin (The Bachchao Project ). This collaboration brought in various expertise ranging from developmental hardware to other expertise of usability on the table, facilitating effective solutions.

Here's a gist of what went down during the weekend long overnight Hackathon :

It was attended by around 25 hackers on day 1. We witnessed around 8 pitches, some of which were posted before on the hackathon event page. After 3 reviews the hackers worked on 6 solutions at the end of day 2.

The Problems that were pitched :

Enable India:

Training a visually impaired person to learn to use a computer requires them to become familiar with a keyboard which usually takes a while. How about a better way of handling this.

Doctors without Borders :

Meeting Water and Sanitation needs of displaced people in emergencies is a top priority in order to avoid outbreaks of diarrhoea and other hygiene­ related diseases. This is however not known to all. Hence we need a digital system to share this information .

Taarifa.org :

Taarifa Platform is an open source web API, designed to close citizen feedback loops. Recently they created a water point management application for Nairobi. Taarifa needed an Extrapolation on the platform to work in new contexts and places .

Breakthrough :

We do not know which police stations are women friendly / sensitive to women who are coming to register a complaint. It would be great if we can find a solution that can help the public know about the friendly police stations and thereby encouraging other police stations to take measures to be more gender sensitive.

Pausch Initiative :

Underage labourers and street kids are one of the pressing issues in urban india.They miss out on education and rights because of poverty. Hence there is a need to create a platform to solve this issue involving citizens passively ,actively or anonymously to erradicate this .

NGO Reach :

Lack of interactive platforms to connect the NGO and volunteers makes it hard to find the desired volunteers with right skill set.

Farmalution :

In India, a set of farmers are connected to a supervisor. He helps in educating them with better agricultural techniques, what type of crops to grow in which season, insights about what fertilisers and pesticides to use, irrigation etc.,After the supervisor educates the farmers, it has been a challenging job to monitor the farmers, But there is no feedback mechanism to understand if methods are being implemented or not and are working in a manner intended.

Project Child Care :

Indian Government spends huge sums of money to provide children with education. They have introduced the midday meal scheme and also provide clothing, bicycles, bus passes, scholarships etc. But there is rampant corruption by middlemen which prevents the intended from getting the help .Hence there is a need for cost effective,efficient, simple solution to effectively track these freebies from government, so that it reaches the intended.

The Problems that were worked upon were :

Farmalution :

After the supervisor educates the farmers, it has been a challenging job to monitor the farmers, if they have implemented the technique or if they need any help etc. Hence a solution where a outbound call(VOIP) is made to a farmer in his local language, and feedback is got based on choosing the desired option. This is later analysed to get statistical information about the implementation.

Team : Satvik Jagannath , Akash Nidhi , Sharath R

Git Hub link : https://github.com/bozzmob/RHoK2014

Tolk :

The solution was done in two ways;

1. Proof of concept for talking keyboard via hardware cracking ­ by developing overlay ­with braille on top of existing keyboard, which could talk when key is touched

2. Working prototype on raspberry pi with usb keyboard & earphone with rasbian OS, Python, espeak TTS. ­They bypassed login & ran espeak login script­ so when a key was pressed on the keyboard it read out via head phone.

Team : Muthu Raj , Anil Prasad M N , Rishi Gaurav Bhatnagar , Rohit Shetty

Project Child Care :

Since the technology chosen should be simple, should have ease of use and be cost effective .QR Code was chosen to be used on school ids. These are generated with the details of the students. A basic smartphone with QR Code Reader app is used by school representatives to scan, after the student receives the benefit thus preventing misuse.

Team : Srikanth Mattihalli , Preetham Hegde

Git Hub Repository: https://github.com/srikantmatihali/childcare

NGO Reach :

A platform was created to create and publicise the events at the click of a button . Such a repository will help those interested in volunteering to help find events happening in their localities.

Team : Varun Vadigeri , Brateesh Kumar , Chandrahas M , Gagan K, Anoop Chandra

Pausch Initiative :

A platform where one can download crowd sourced tutorials was created. This can be downloaded and used in an app which serves as a tutorial for volunteers to teach illiterate people. A case of child abuse can also be reported through this app.

Team : Rohit Shetty , Anoop Chandra

Police Station Safety Audit :

There was no pre­existing data available hence a data collection and display mechanism was created during this event

Team : Ankita H

Farmalution , TolK and Project Child Care were chosen as top hacks among those presented.

Sponsorship :

We would like to thank our sponsors for showing us so much love and support.

Our sponsors are;

Venue : Beaglesloft

Food : Nithin R and Saakshin Technologies

Prizes : Baishampayan Ghose

Printed Bragging Rights : Yogesh Londhe

Banners : Breakthrough

Sandbox collective provided a much needed break after all the intense brain buzzing, by performing “Rosencrantz and Guildenstern are dead”.

BeaglesLoft has kick-started its tech building initiative with TechBuilders 1.0. as we reiterate the need for innovation and disruption in the startup capital of India. This past Saturday saw a healthy turnout of tech-enthusiasts and we decided to try something crazy: a hackathon without laptops.

There were no frameworks or APIs or programming language wars; just good ol’ pen, paper, probability and perseverance.

While The Loft Team knew there would be plenty of challenges, the attendees did not make it any easier for us with their brain buzzing questions. After a brief introduction and a refresher on Probability Mass functions, we broke out into smaller teams.

Each team was challenged to find ways to apply Bayesian Reasoning to real life problems and make Thomas Bayes (and probably our mothers, too) proud.

 The problems we worked on ranged from serious topics like out-foxing evil spammers to the downright hilarious question of how to tell if someone likes you when they walk by and smile at you. After a few hours of laughs mixed with intense table banging and brain busting, each team presented their solutions, bracing for the brickbats. It was great to see how each team tackled the same data in radically different ways.

When the event "ended," participants hung around the Loft and bonded over more ‘real-life problems!' Bayesian Reasoning permeated throughout. In a short afternoon, strangers had begun to form a tight community. Many participants volunteered to help take TechBuilders to the next level. 

  The overall response from our first meeting is a validation in part for our vision at BeaglesLoft. There is a growing need for hardcore computer science and math within the tech community in Bangalore. BeaglesLoft is incredibly excited to be a pioneer in this community, and we can't wait to grow the community and host more incredible events in the future.

In the field of computer science, Bayes' theorem is most often used in "classification”. Classification deals with the problem of assigning a label or a tag to input (and thereby classifying it). 

Let us revisit the problem of "classifying" someone as belonging to a certain city. To keep matters simple, let us limit ourselves to 3 cities: Bangalore, Chennai and Kolkatta.

A classification task, generally begins with an input data set.The input data set gives us more information about the classifier labels and their features. Let us build a sample data set for our problem:

City             Beer Drinker | Not    || BJP | Not || Curly Hair  | Not | Total  _______________________________________________________________

Bangalore   |  700      |       300    || 650  | 350   ||   300       | 700 | 1000

Chennai      |  100      |     1900    || 300  | 1700 ||   1300     | 700 |  2000

Kolkata       |  50         |      450    || 100  | 400    ||   25         | 475 |  500

The general idea is to progress as follows: 1. Treat each classifier label as a hypothesis 2. Construct prior probabilities for each hypothesis (agnostic of its features) 3. Each feature is treated as evidence and what we learn about the input data is used to update our prior beliefs 

Looking back at our data set, we see that there are 3 features for every city. Let's say you learn that someone is a craft beer drinker, has curly hair and did not vote for BJP.  Which city is he most probably from?

Now we are dealing with encountering multiple evidences which we haven't dealt with before. To be able to apply Bayes' theorem, we will have to make one crucial assumption. We will have to assume that the series of evidences we see are "independent" of each another.

This allows us to apply the following algorithm to update our beliefs: 1. Construct prior hypothesis from the input data set, irrespective of the features (let’s call it priors) For ex. P(Bangalore) = 1000 / 3500 2. Convert the evidences you have encountered into an array (let's call it evidences) 3.For evidence in evidences:          For hypothesis in priors:                    Calculate P(hypothesis | evidence) as we have done before                   Update probability for hypothesis in priors 4. Now, priors will contain our updated beliefs, after learning a set of features about the input data

Generally the hypothesis (classifier label) which has the maximum probability in the ‘priors' array wins and input data is classified as that.This algorithm gives us a very rudimentary skeleton to be able to implement a Bayesian classifier.

It seldom happens in real life, that we encounter data which is independent of each other.However, the Bayesian classifier does give surprisingly good results in many cases. This implementation of a Bayesian classifier raises a few questions: 1. Are there any strategies for picking priors, if we do not have a data set like the one we just saw?  2. Is the best strategy for classification picking the label with maximum probability? 3. Why does the evidence need to be "independent" of each another?

This concludes our series of posts on Bayesian Reasoning. We cannot wait to apply all that we have learned at our first TechBuilders event on 7th.

Before we start employing Bayesian reasoning into software systems, I wanted to dedicate one blog post for some of the philosophical implications of Bayes’ theorem. My intention is to demonstrate how elegantly Bayes' theorem expresses reasoning in a mathematical and computer-program friendly way.

1. Closing gaps in reasoning: In the cancer problem in the first post, most people guess 98% because they fail to consider 2 key factors: a. The fact that only 1 in 100 people actually develop cancer. Bayes' theorem covers this in the prior probability. b. Considering the fact that a person has tested positive under ALL hypothesis. Yes, 98 out of 100 people with cancer test positive, but, what about the fact that 2 out of 100 people without cancer also test positive?  The normalising factor of Bayes' theorem makes sure we take this into account while reasoning.

2. Dangers of having a high prior probability: Referring back to Bayes' theorem derived in the second post, it is interesting to see how the terms of Bayes' theorem affects the posterior probability. Prior probability and likelihood have a direct effect, while the normalising factor has an inverse effect. If you have a high prior probability, your belief in the initial hypothesis will not change by much, unless you see data which is true in a lot of other hypothesis. Therefore, it is highly unlikely you will come across data which will make you adjust your previous belief by much. Giving it some thought, this almost seems like a mathematical representation of the very human attribute of stubbornness.

Before we dive into Bayes' theorem, we will have to understand two basic concepts:

1. Conjoint probability: What is the probability of two events being true? Written as P(A & B). 2. Conditional probability: What is the probability of one event occurring given another has already occurred? Written as P(A|B) So what is P(A & B)? That depends on the type of events A & B. If they are independent events, meaning if A has no bearing on B, it is simply P(A) * P(B). For example, coin tosses are independent events. The output of your first coin toss has no bearing on the output of your second coin toss. Predicting if I will run today is a dependent event. If I run today, the probability that I will run tomorrow is much less. Probability that I will run on a given day, depends on what I have done the previous day. If A is the event that I will run today and B corresponds to I will run tomorrow, then: P(A & B) cannot be written as P(A) * P(B). It should be written as P(A) * P(B|A).

Generalising, P(A & B) is written as P(A) * P(B|A). For independent event P(B|A) == P(B) It is interesting to see that the direction of dependence does not matter. P(A & B) can also be written as P(B) * P(A|B) Equating the two, P(A) * P(B|A) = P(B) * P(A|B) Which gives us, P(A|B) = (P(A) * P(B|A)) / P(B)   This is where the creativity and genius of Laplace and Bayes comes in. They equated A to an original hypothesis and B to new evidence presenting itself.