Model training is a critical phase in the development of AI models. It starts after data preparation and long before the operationalization of a model.
In the machine learning world, model training refers to the process of allowing a machine learning algorithm to learn patterns based on data. These patterns are statistically learned by observing what makes an answer correct or incorrect (supervised learning) or by discovering the inherent patterns in data without being told the correct answers (unsupervised learning).
Let’s take a machine learning algorithm like Logistic Regression which is a supervised learning algorithm. And we’ll assume that the task is to predict if an email is “spammy” or “legit”. The table below shows some sample data, with the correct answers that can be fed into an algorithm like logistic regression.
|Spam/Legit (Correct answers)|
|You owe me 10,000 USD…||Spam|
|Dear, this weight loss program…||Spam|
|Hi James, How about tonight at 7 pm?||Legit|
By itself, the logistic regression algorithm does not do anything. But, when you start feeding data to the algorithm, it starts learning which signals in the data are important. Perhaps certain words and email addresses are good indicators of spam. It also tries to optimize what it’s learning such that the predictions on the development data is accurate.
In the end, it produces a model. This model is in essence a computer program that has learned how to make predictions on new data shown to it.
It can then be used to make predictions in the real world.