This is a short description of machine learning process including the training and testing phases. I may add more details during my machine learning journey. If you are more interested why I am back to this topic, I wrote about it on last month.
The main source is my own bachelor thesis, which I wrote in 2012. Here you can find also nice and simple explanation of the machine learning workflow:
A Simple Explanation of the Machine Learning Workflow
In that article testing phase is described as “putting model to the production”, which makes sense.
Steps of the Machine Learning Process
- Formulate your original problem into a learning problem
- Collect all data which is related to your learning problem
- Choose and plan the usage of learning method (ML algorithm)
- Training (=learning) phase
- needs training data
- train the algorithm to create right output
- the output product is a model
- Testing phase
- needs testing data which is different than training data
- uses the model created in training phase
- makes decision on classification of test data
Training Phase
- labeled training data goes as an input to the feature extraction
- output of the feature extraction is a feature vector
- feature vector goes as an input to the learning process, where the model is build by using some machine learning algorithm
- output is a model which will be used in a testing phase
Both training & test data have to be processed in the same way from the begin to the feature extraction
Note to the training and testing phases
Testing Phase
- unlabeled testing data goes as an input to the feature extraction
- output of the featured extraction is a feature vector
- feature vector goes as an input to recognition process
- the model created by training phase is used for decision making
- output of this phase is decision
Featured image by Markus Winkler on Unsplash.