5 things about Reinforcement Learning beginners must know

For beginners in Data Science who have already learned about Supervised learning and unsupervised learning have surely passed through the name Reinforcement Learning. In this blog you will get to know what Reinforcement is, and important things you should know about it.

What is Reinforcement Learning?

In simple words Reinforcement Learning is training a model by going though a again and again and getting better at predicting by reward system.

So an initial stage is there and a final stage is there. The final stage has reward. The model needs to go from initial stage to final stage. Model will start with trial and error. As it reaches the final stage it has set of movement or path. Next time it will try to find out more efficient way to reach its final stage.

After a set of iterations the most efficient way is finalised and treated as final prediction.

  • Robotics Implementation
        The most relatable and practical application of Reinforcement Learning is in Robotics. Here a robot tries to achieve a task. The task can be anything such as carrying on object from point A to point B. Here robot will first try to pick up the object, then carry it from point A to point B, finally putting the object down. Doing all these activities with constraints like keeping the object intact and be in a given environment gives higher reward with least time to do so. Hence a robot learns to complete a task.
        
  • Self learning (Tweak model by reward system)
        When we say self learning, this is the algorithm we are talking about. Here we do not need to tweak model every time before training. We only need to define a proper reward system. Then model keeps learning by itself to come up with best solution.

  • Most powerful Algorithm
        Some of the Data Scientists have tagged it as most powerful algorithm as it’s ability to self learning. If we provide enough resources and efficient reward system then a model is powerful enough to execute any task whether technical or not. An example can be to understand human emotions simply by looking at faces.

   
  • How it is different from supervised learning?
        In term of how it trains on the final result, it can be sometimes hard to differentiate it from supervised learning. The main difference is that in supervised learning we have to desired result while in Reinforcement learning we do not have any. We could have a result that’s not even imaginable. Also we have no interdependency in results in supervised learning. While in reinforcement learning solutions help to come up with more efficient solution.

  • Requirement of resources
        Question that would be coming is that if there is such algorithm, then why are we not having any great innovation or breakthrough? The thing is that model iterates many times to get a step in getting efficient. Hence to perfect even a small move is very expensive. It’s like a baby learning to walk. It would take a lot of time and as it is machine hence more processing power and storage would be required. We are still far behind a time where a robot can learn anything try to rule our world. Terminator don’t have enough RAM yet.

If you also want to start your journey in data science, you can do by downloading your FREE copy of the roadmap to Data Science from the link below.

Comments

Post a Comment

Popular posts from this blog

How docker helps Data Scientists?

Image
 Docker has been very popular service in the IT industry. Let’s find out how it helps Data Scientists. Environment Reproducibility:  Docker provides an easy way to package all the dependencies required for a specific data science project into a single container. This allows data scientists to create a standardized environment that can be easily replicated across different machines and platforms, ensuring that the code runs the same way everywhere. Portability:  Docker containers are portable and can be easily moved between machines and platforms. This means that data scientists can easily share their work with others and deploy their models into production environments with minimala effort. Version Control:  Docker images can be versioned just like code, allowing data scientists to easily roll back to a previous version if needed. This can be especially useful when working with large datasets and complex models that require frequent updates. Scalability:  Docker...
Read more

5 BIGGEST MISTAKES DONE BY BEGINNER DATA SCIENTISTS

Image
Did you know that d eleting columns in a dataset is one of the BIGGEST MISTAKES made by the beginner data scientists.  Are YOU one of those? By the end of this blog you are going to know 5 BIGGEST MISTAKES made by beginner data scientists. To become a data scientist, you must keep in my to avoid these mistakes at ALL COST 1. DELETING COLUMNS -  If you find any null value in the column, and then go for dropping the column. That's one of the BIGGEST MISTAKE. You should consider if that null value can be replaced with some value that does not change the bias of that column. Generally these values are central tendencies. That are mean, median and mode. 2. FORGET ENCODING - If you think that your model is fitting successfully without encoding, so you should skip it. That your that's a mistake. Encoding helps the model to establish greater spectrum for the column and having the column represented as numerical helps the model to have a comparable outcome. 3. STATIC PARAMETER -...
Read more

5 Things you should know about TSNe

Image
 Did you know that you can visualise data that has higher dimensions? We can do it using TSNe. TSNe stands for t - distributed stochastic neighbour embedding What is TSNe? TSNe is a statistical method that is used to visualise high dimensional data by giving each datapoint a location on two or three dimensional map. It is a non linear dimensional reduction technique. How TSNe works? First it constructs a probability distribution over pairs of high dimensional objects. Similar objects are provided high probability and dissimilar objects are provided low probability. TSNe then defines a point on low dimensional map for similar objects. When to use TSNe? Usually with big data we have large number of columns or parameters associated with the data. Representing such data on two or three dimensional map is impossible. We cannot always go for deleting columns as it would lose all of its value then. In such cases we can use TSNe to repsent visually without losing much value. How to use TSN...
Read more