My journey till now in AI and Demystifying the technology
Anecdote:
I worked in IBM from January 2018 till January 2020 as a mainframe system programmer and resigned because I want to pursue higher studies. No doubt the profile is so niche and secure, but I always wanted to develop my AI skills. IBM was also focusing on AI, and fortunately, my interest matches their business.
Demystifying machine learning:
After leaving IBM, I took some time to clear the formalities of getting into McMaster University in Canada, so I started my journey in September 2020 for AI. I took the course Complete Machine Learning & Data Science Bootcamp 2020 by Daniel Bourke from Udemy. It is a great course to start coding the machine learning models straight away and understand the practical applications. I took the course without any research, just saw the ratings, and enrolled myself. Luckily it was the best accidental decision that I took. Professor Daniel is a self-taught machine learning engineer who has many AI skills and has also worked for a company in the same field as a machine learning engineer and has a youtube channel where he reviews many other AI courses. I completed the course in about three months, after which I felt a gap in understanding the algorithm behind machine learning.
Now I will summarize what generally happens while creating machine learning applications — a bit of an astonishing spoiler. You do not need to know the algorithm’s understanding; their names are enough to become a machine learning engineer. WHAT!!!. Yes, it is true; ask anyone who is in the same field.
The steps:
1. Import libraries in python related to machine learning(around four to five usually if you want a decent performing application)
2. Data on which you want to apply A.I. For instance, a tabular data of medical record having various health parameters as columns and one column stating as heart disease yes/no
3. Bifurcate data into X and y, with X having all the rows without a column of heart disease, y having all rows with heart disease.
4. Now, apply any one machine learning algorithm on the X and y data, for instance, SVC; it is as easy as it looks below:
from sklearn.SVM import svc
svc_clf=svc(C=0.2)
svc_clf.fit(X,y)
5. To test, provides some real data having the same columns excluding the heart disease column to svc_clf.predict(X_test). Now, this will provide us with yes/no result
6. That is it your most basic machine learning model is ready, and most of the time, it is efficient.
Answering if there is a need for studying algorithms:
See, it is so easy. We have lots of other algorithms and lots of different problems to solve, like Regression, which guesses the value, like Covid cases projection in future months according to the availability of vaccine and other environmental factors. Most of these things were taught by Daniel in his course. However, you will eventually have a curiosity to know SVC and what it is doing under the hood, and is it beneficial to spend more time knowing their working, or are we good to go without knowing it. Now observe, there is a parameter C=0.2 that I had used in point 4; it is known as hyperparameter used to fine-tune the algorithm. These algorithms already have a default value provided by the libraries, and you do not require to provide the value if you do not know. To be a good and outstanding AI engineer, we need to know a little bit about these hyperparameters and their tuning. To answer these queries, I took another three-month course on machine learning basics by Andrew NG on Coursera suggested by Daniel himself. After completing it, 90 percent of the queries were solved, and to add the cherry on the top, I completed part 1 of “Hands-on Machine learning with Scikit-Learn, Keras and Tensorflow, “a book written by Aurelien Geron. He is one of the famous personalities in the world of AI. He is also one of the pioneers of the youtube recommendation engine.
My imminent goals:
It is almost eight months into the self-study phase, and my next goals are to complete two courses from Fast.ai and another two courses from Deeplearning.ai. Post which, I will be completing the remainder of the book from Aurelien Geron. It already seems more than enough, and I would say to some extent, yes, anyone can make wonder AI applications with just completing part 1 of the book by the later author, but I feel that I have now come so far and why to stop now, so I have decided to learn the remaining things and complete the book once and for all. I will share a project which classifies people who ring the doorbell of my home after completing the above goals.
Result of all the effort:
It is well known that we should not worry about the fruit of our labor, instead concentrate on doing things more efficiently and enjoy the process. Enough with the philosophy Ankur come on !!!. Yeah, sorry for that. It is true. Nevertheless, I have completed one competition on Kaggle where I scored in the top ten for a project Blue Bulldozer. I have realized many things that I could improve in my previous job with AI’s help. To name a few, prediction of spool getting full; error logs reporting from the Syslog.
Advice for young explorers in AI:
My advice to beginners is that there is no right or wrong path in learning machine learning but completing what you have started and improving python skills.
Links to resources that I have studied:
Machine learning Bootcamp by Daniel Bourke
Andrew Ng machine learning course
Hands-on Machine Learning Scikit Learn TensorFlow