Machine Learning Abstract Concept
Creating an algorithm that a computer then uses to find a model that fits the data as best as possible. And makes a very accurate prediction based on that
ML is different from traditional statistical methods because we do not give Instruction to find a Model, rather give the algorithm to the machine to learn by itself.
ML Algorithm
- Trial and Error process
- Each consecutive trial is at least as good as the previous trial
- No Instructions for training the model, the only the final goal is given.
- improve the complex computational models to solve many business problems.
Example of Above Concept from 365 DataScience (Training a robot to shot arrow — just a fictional Example ):
The optimization alg will keep adjusting the machine to Improve the Obj Fn, this process may be done 1000 times. Each trial, the Robot may learn new things on the way and improve but stop training until Objective function is minimized.
Types of Machine Learning
- Supervised Learning- Training Algorithm similar to Teacher training Students
- Labeled Data = Associating or labeling a target to a type of arrow, we know the data before training so can associate with the target
Objective Function- To measure the inaccuracy
Optimization Function — To improve the way the robot shoots
The Training Process will train the robot to hit the labeled arrows to the target such as 1- broken arrow, 2-short arrow, 3-medium arrow and 4- Long arrow
2.Unsupervised Learning
- Unlabelled Data
In this Example: You may have lots of arrow/data, so in this case, use unsupervised learning to cluster data by itself. The clustering may reveal many insights such as a new pile of weak arrows etc. In Practise when we have a lot of data, we could label them using Unsupervised learning then apply supervised Learning.
3. Reinforcement Learning
- reward system
Maximize Optimization Function rather than reducing error
Hello World Program of ML in Python — Iris DataSet
Another Example using Dataset provided in the Previous post for Basic Feature Selection process
Another Hello World of ML is Titanic Data Set at Kaggle, great to Begin Learning!
