We have discussed supervised learning in machine learning in the previous section. Is unsupervised learning just the opposite?
Well kind of that. And the story starts from here. Because in supervised learning algorithms we need a teacher who will supervise the learning. Right?
When the learning is unsupervised, we cannot expect a teacher who will supervise. Quite expected.
Let us dig deeper.
When the dataset is big enough, we cannot gain an insight from the tabular data. In that case we can ask the computer to make some sense from data.
Whether it is supervised or unsupervised, it hardly matters. The machine learning algorithms learn from the datasets, from the experience as well.
In supervised learning, we use labeled data. But in unsupervised learning we don’t have any labeled data.
That’s the big difference.
If you are a complete beginner your journey to learn TensorFlow might start from here.
For the TensorFlow beginners we have a dedicated category – TensorFlow for Beginners.
But besides that, you may need to learn several other machine learning and data science libraries.
As a result, you may check these categories as well – NumPy, Pandas, Matplotlib.
However, without learning Python, you cannot learn the usages of these libraries. Why? Because they all use Python as the Programming language.
Therefore please learn Python at the very beginning and start learning TensorFlow.
And, finally please check our Mathematics, Discrete Mathematics and Data Structures categories specially. We have tried to discuss from basic to intermediate level so that you can pick up the core ideas of TensorFlow.
When the learning is unsupervised, the algorithms first analyze the datasets and then they cluster the data which are not labeled.
In many cases, we don’t have enough knowledge about the problem.
At the same time, we might not have enough computing power to properly model the problem.
However, the unsupervised machine learning algorithms find hidden patterns when they cluster the datasets.
We may find a similarity with human cognition.
Why?
Because we come to this planet with built-in classification mechanisms. Human cognition has some advantages that we cannot expect from a machine overnight. Right?
But before classifying cats we need examples.
And we also need labels.
Humans learn with the help of a teacher. We cannot learn to swim on our own.
We cannot differentiate between cats and dogs on our own. Our parents, teachers, friends, show us examples in many forms, and label them with a ‘Name’ or a ‘Noun’.
Examples make things easier. And that is also true for machine learning when there is no teacher. In short, there is no label.
As a result, in unsupervised cases, the learning algorithms have only inputs and they extract output, or knowledge from that dataset.
As we have said before, when the learning is unsupervised there is a task that deals with the classification first. Then analysis follows.
From that analysis, the algorithms try to establish hidden relations between the attributes.
There are plenty of examples around us.
Just look around and you can get your head around the idea.
In a supermarket, they always display some products side by side.
The idea of consumerism and capitalism rests upon this great idea of the relationship between products.
Certainly sauce and chips have a close relationship.
They keep the products close to each other.
Just imagine other related products which might have close affinity.
Now as more products keep added to this dataset, the relationship becomes complex.
Moreover, representing that relationship might be challenging as well.
We will talk about it in great detail in the coming section.
So stay tuned.
Leave a Reply