Machine Learning Explained
Hello everyone, my name is Nisarg Kadam, I am Senior Consultant at WonderBotz, UiPath MVP 2021 and a YouTuber.
Whenever we read “Machine Learning” in any article or in any subject we immediately click on it out of curiosity. Curiosity of how machine learns? and what it can do when it learns?
For some of us who doesn’t know much about technology, machine learning is nothing but magic which predicts future. Some of us knows the technology but still thinks that machine learning is something which is difficult and hard to implement.
If you are curious to know how it all works then this article is going to be an interesting read for you, because I am going to make machine learning concept effortless for you to understand.
Let’s take an example of engineering student. Name of this student is Jack.
Assumption: Jack completes his higher education with a good score. Now it’s up-to him to chose domain of engineering. He choses software engineering as it’s his passion.
Reality: Here, Jack is a machine learning model and based on the requirement we select which algorithm to use. Similarly, Jack choses his career option as software engineering.
Assumption: Once Jack choses software engineering, he must buy all the books and get access to online training material. These books are suggested by college and the university for Jack.
Reality: Here, university is nothing but the developer who is going to build machine learning model. Based on the requirement of machine learning model developer gathers data to train this model which is nothing but books and study material for Jack.
Assumption: Jack learns throughout each semester and attends all the lectures daily. Jack is a regular student and never misses any of his lecture or practical.
Reality: In this case our machine learning model is learning from the historical data which is used for training this model. Usually training a machine learning model takes huge time based on the processing speed and amount of training dataset. This process is called training pipeline.
Assumption: Around end of semester Jack must appear for an examination. The questions for examination are relevant to the study material.
Reality: Here, examination is referred to as evaluation pipeline. When the training of machine learning model is done, the model is tested for evaluation of the training. For this evaluation we use test data which is relevant to training dataset. According to best practices the ratio of training and testing dataset must be 80:20 where if we have 1000 rows of training data then testing data has to be at least 200 rows with relevant dataset.
Assumption: Jack appears for the exam and fails as he could not meet passing score of 70%
Reality: Here, after execution of evaluation pipeline it will generate the score. If the developer is not satisfied with the score of evaluation, he will not mark machine learning model as ready to use. Developer will plan to train the model with more diverse data.
Assumption: This time jack learns from books, refers some previous solved questions papers and online study material. Jack wants to make sure to crack the examination this time.
Reality: Here, developer plans to train this machine learning model with more data. The more diverse data you have the more accurate your machine learning model becomes. Developer runs training pipeline again.
Assumption: This time jack appears for the exam again, and he manages to score 91% and he is passed with distinction.
Reality: Here, second examination is nothing but re-execution of evaluation pipeline for newly trained machine learning model. The evaluation pipeline this time gives accuracy score of 91% which is acceptable by developer, and he is now satisfied with the training of the model.
Assumption: Now, on completion of engineering Jack must appear for interviews to get selected in some software company as a developer.
Reality: After training and evaluating several times, developer starts testing with production ready data and predicts whether model works as expected.
Assumption: Jack got selected in an IT company as a software developer and now he is using his skillset to develop applications.
Reality: Now this machine learning model is deployed on production and is predicting the results based on the training and experience.
Assumption: Going further to work with new technology Jack must reboot his skills and learn new technology.
Reality: To keep this machine learning model in production we must make it smarter every week. Now a training and evaluation pipeline runs with new train and test data to make sure this model is up to date with the new information and become more matured.
Assumption: Jack is now 10 years experienced and working as an excellent performer in his company.
Reality: Here, we have now trained machine learning model several times and 10th version of model is deployed. Which is smarter, faster, and reliable based on its experience.
In this entire story of Jack, we have seen how machine learning works, as well as why do we need so much of data to train machine learning models.
I hope this story makes the core concept of machine learning even more easy, and it generates a little interest to learn more about it.
If you wish to learn more then:
- Subscribe to my YouTube channel: https://www.youtube.com/c/nisargkadam/
- Follow me on LinkedIn: https://www.linkedin.com/in/nisargkadam/