What is a Neural Network?
In the last decade, neural networks have seen a resurgence in popularity. Modern neural networks are toolkits of building blocks that allow model builders to design models that exactly represent the problem they wish to solve. Neural network libraries provide tools (such as auto-differentiation) to speed up the process of fitting that model to data.
Why are Neural Networks important?
Neural networks thrive in high-signal, low-noise environments – in other words, there is a lot of relevant information to your target variable and not a lot of extraneous data or random volatility. This type of problem has complicated relationships that are tough for normal machine learning models to tease out. Neural network models also complement traditional machine learning models like XGboost and make for good ensembles when both approaches are combined.
Neural Networks + DataRobot
DataRobot’s model blueprints include several “pre-baked” neural network models that are applicable to business problems easily solved with automated machine learning. These models range from very simple neural networks to state-of-the-art models that excel at capturing non-linear signals. DataRobot also employs a neural network model known as “fasttext” that results in state-of-the-art text mining, making it perfect for gleaning insights from anything from doctor’s notes to product reviews.