The 6 Steps to Machine Learning
Machine learning is a small field of artificial intelligence. It is computers ability to learn how to do new tasks without specifically being programmed too. They bring together statics and computer science and they slowly get better at tasks. For example, if a computer wanted to figure out the difference between a cat and dog all you would have to do is feed it a bunch of images with labels telling it if it’s a cat or dog. Then, eventually, it will be able to tell the difference between the two. There are a few steps in between the begging and end result though.
There are still 6 main steps in this process. Collecting and preparing data, choosing a model, training, evaluation, parameter tuning, and the prediction. I’m going to stick with the same example of the cats and dogs to explain these steps throughout the article.
1. Collecting and preparing data
Once you have your idea of what you want the machine to learn about you to have to do the first main step in the process of machine learning. This is collecting data. This step is crucial and needs to be done well because the quality and quantity of the data you collect have a huge impact on the final result.
Using the cats a dogs example we could find pictures of cats and dogs and categorize them so the computer knows what’s what and it starts to identify what is a cat and what is a dog or get data on different facial feature differences, tail lengths, physical shape etc.
After you finish collecting the data you need to prepare it so it can be used to train the machine. In a nutshell, preparing data is a set of things you do to the data you collect to make it more suitable for the machine learning process.
2. Choosing a model
After gathering and preparing all the data you have to chose a model. There are many different models and each one comes from a different algorithm approaches and will react differently under different data sets. Some of these models are linear models. decision tree models, a sports vector model and many more. I am going to use the linear model for the cats and dogs example because there are only 2 things the machine needs to classify.
This picture is an example of what a linear chart and nonlinear chart looks like. The line down the middle is the separating between cats and dogs and each circle represents a different data point so, for example, the average tail length or hight of the animal.
This is the part that makes up the majority of machine learning. Training. During this step, we input all the data into the machine, training it to be able to predict the difference between the two things, in this case, the difference between the dog or the cat. You can always add more inputs but in this example were are only going to use two. This step can be similar to teaching someone to drive or cook a meal etc.
The formula for a straight line is y = mx + b, where m is the slope, b is the y-intercept, x is the input, and y is the value of the line at x. We can adjust the values of m and b, which are our values for training.
There’s a lot of features in machine learning, which are all of the m’s. After collecting all of these m values they become what we call a matrix, which we call W or weights. B values are also organized into a matrix. We call these biases or b.
The process involves adding in a few random values of W and b, as well as trying to predict the output with these values. In the beggining, it won’t work too well, but we can compare the predictions we made with the output that should have been produced. Then, we adjust the values in W and b, so that the predictions we make can be more accurate.
The process repeats. Each cycle of updating the weights and biases are called one training step.
After the training process is complete we need to test the model to make sure it’s working properly. Evaluation allows us to test the model with data its never seen before. Based on how the model runs after the evaluation we will get a good idea of how it will work in a real-world situation
5. Hyperparameter tuning
Then, once we finish with the evaluation we need to take the notes from the evaluation and improve it. We can improve it by doing something called parameter tuning. There are different types of parameters. Parameters that define the model architecture are called hyperparameters. The actual process of finding the right model is called hyperparameter tuning.
After all that data collecting, training, and evaluating we finally come to the last step which is the prediction. In this stage on machine learning, we should now be able to use it to tell the difference between a cat or dog, orange or apple, or whatever you originally wanted it to be programmed to do.
In the end, machine learning is a huge topic and in the future will be applied to many, if not all industries. This includes healthcare, education, entertainment and many more. Machine learning can change how we do things and can also do similar tasks were doing right now but faster and more efficient saving lives in health care, teaching better in education and more. Machine learning and AI has been kept in the dark for a while now but it’s starting to get more popular and common. We are interacting with machine learning and AI more than we think and machine learning is being applied more and more into peoples lives every day. Machine learning is our future and I think it’s crucial that we all learn about it now.