Skip to main content

5 Beginner Tips For Making An AI Model Work

If you’re getting started with artificial intelligence (AI), there are plenty of tips and tricks to help. As a Data Scientist at Trayport, I spend most of my time building and reviewing AI models. Here I’ll cover a few quick tips I would recommend to beginners to get started with AI modelling.

But first, what is an AI Model?

Within the realm of artificial intelligence, models are used to mimic the learning process of a human brain. Given the sufficient amount of training examples, they can learn to replicate the decisions of a human expert (or even outperform them!).

Maria Fabianska - Trayport

Maria Fabianska

5 Tips For Beginners

1. Know Your Data
This is the first step. Before plugging the data into your model, visualize and examine it so that you really have an intuitive understanding of what is fed to the model. The same applies to the data predicted by the model: prepare some plots comparing it to the target values. It will allow you to spot mistakes easier and be more confident in whether or not your results make sense.

2. Define Your Goal
Use a loss function that reflects well your learning objective. This is of paramount importance as the loss function is used to steer the training process in the right direction.

3. Use Additional Evaluation Metrics That Are Easy To Interpret
The value of the loss function allows you to compare the performance of different models, but it is often not easy to understand in absolute terms (“Model A performs better than model B, but does it perform well in general?”). That’s why it’s worth defining additional evaluation metrics which will give you a quick and intuitive measure of the quality of the results.

4. Detect Rare Events
If you have a classification task and one class appears 1000 times more often than the other, the model will likely learn to predict the more popular class. Check if your dataset is imbalanced and if it is, employ proper techniques for the model to pay more attention to examples from an under-represented class.

5. Start Simple
Before building a complex model which uses novel optimization techniques build a strong plain vanilla baseline. It will help you see if the fancy additions are really boosting the performance.

For AI beginners, it’s easy to over-complicate your first model and it’s even easier for negative results to discourage continuation. But building an AI model doesn’t need to be complicated, you just need to start small and bear in mind the tips I have provided.

Good luck with your model.

Maria Fabianska
Data Scientist - Maria is a Data Scientist at Trayport who loves discovering patterns is large data sets and is passionate about sharing knowledge.