Tools in the world of AI

Hello world,

This past Saturday (10th November), we at School of AI — Raleigh had our second meet-up. I had been preparing for this meet-up for a month, to collect and prepare all the information and reading material ready in time for everyone. 

By far the greatest danger of Artificial Intelligence is that people conclude too early that they understand it. - Eliezer Yudkowsky

I started off the meet-up by setting a foundation for understanding business requirements and translating them to technical requirements. Essentially going from data to information. Taking an example of an Artificial Intelligence, or machine learning, or data analytics problem. The first step is to brainstorm and to come up with ideas. Once the basic premise of the idea to be executed can be clearly envisioned, the data can be collected or already available data can be used according to the business requirement of the problem.

The next set of steps that need to be executed depending on the business requirement, would be to extract the data from the source file. This extraction process can be facilitated by using the pandas library. Once the data is extracted, we would like to perform some kind of mathematical operations on the extracted data. That is where NumPy comes into the picture. NumPy is a tool for performing complex mathematical operations on the extracted data. Once the data has been manipulated, and the results are generated, the next step is to convey these results to the stakeholders involved the project. For this purpose, Matplotlib comes very handy.

Again, using LinkedIn as a platform, and building up off our last meet-up, I approached Karen McCann — the founder of AI Now - RTP, to give a talk at our meet-up. Her presentation was three fold — past, present, and the future. Karen started talking about the history of AI, focusing on topics that were essential for the development of AI in the 1960s. She then talked about the present situation of Artificial Intelligence, drawing everyone’s attention to present hot topics like Robotics, self-driving cars and the problems being focused on with academia and business right now. It was an interesting digression that the conversation took when one of the audience members pointed out the Ethics involved making decisions in self-driving cars. 

The main goal of the meet-up was to introduce the topics of pandas, NumPy, and Matplotlib; followed by a general discussion about TensorFlow. The reason we want to use libraries like pandas is two-fold. The first reason is to optimize the solution. When we implement a solution, we would like our code to be optimal in both space and time. However, not every approach that we take would be the most optimal. For libraries such as pandas, thousands of developers have worked on it to provide the most optimal method to perform certain functionality. The second reason is that we developers are the lazier of the bunch. We would like our code to be as little as possible, so even if we are capable of writing extremely optimal code similar to the ones in the libraries, we would rather want to use the library, than type of thousands of lines of code to perform the same functionality.

With pandas, I focused on several topics, with each topic being encapsulated in a jupyter notebook. The general motivation behind the notebooks was to give an idea to the audience — what pandas is, what it can do, and how you can use pandas in your project. Starting off with reading off of CSV files, followed by using pandas to visualize data, merging columns to create new columns, cleaning up the data that has been collected, and loading data from different SQL databases. This section was necessary to give the audience a brief overview of the question “How can I make sense out of data?”

With NumPy, the mathematical part of dealing with data was taken into consideration. This primarily was done, because, with Machine Learning and Artificial Intelligence applications, or problems, a major part of the problem is Math based, including but not limited to — Linear Algebra, Probability, and Statistics. NumPy provides easy and useful functions to perform complex mathematical operations with the data that you’ve extracted using pandas.

Visualizing the results is as important to the stakeholders as it is to the developers since you want to be able to visualize the work that you’ve been doing. For example, while building machine learning models, it is very important to visualize the rate of accuracy with respect to the different hyperparameters that are being tuned.

Finally, I gave a brief introduction to the world of TensorFlow. TensorFlow is a hot topic right now, primarily because of its ease of use, flexibility, and more importantly scalability. Topics that were discussed included classification, linear regression, underwriting and overfitting, and text classification. I also provided a set of six notebooks, that would get beginners up and running with TensorFlow introducing topics like — sessions, feed dictionaries, placeholders and variables, math operations with TensorFlow, and an overview of computational graphs. You can have look at the presentation here

It was an awesome experience talking for more than 90 minutes, even more so helping out the community to learn something new. We at School of AI, truly believe that eduction is a resource meant to be provided for free, through these meet-ups all around the world we're achieving just that.

You can visit my GitHub repository for a detailed description of all the resources discussed in the meet-up plus additional exercises and quizzes to help you get accustomed to the libraries discussed.