How the Trendskout AutoML Tool finds the best model
1. Algorithms
There are plenty of algorithms available for Artificial Intelligence, Machine Learning and Deep Learning. Popular examples include neural networks, gradient descent trees and support-vector machines or SVMs. It is not always straightforward to select the right algorithm for a particular business case. Every type of algorithm – take neural networks as an example – consists of different subtypes, increasing the challenge to make the appropriate choice of algorithm.
Trendskout evaluates each and every algorithm available and selects the most suitable approach for your selected application and data set.
2. Data
Selecting the right algorithm does not suffice to be successful. Algorithms need to have the right data to train themselves, make predictions and identify links. The shape of the data is of paramount importance: data processing will be essential before the algorithm can start analysing and interpreting those data. The options for data processing are virtually unlimited and include data merging, transformation, generation of derivatives and variations, and denormalisation.
Choices about what operations should be carried out on which data, can get complicated and depend on the selected application and algorithm.
3. Parameters
With any algorithm for AI, Deep Learning or Machine Learning, there are several options. They all depend on the type of algorithm (such as a neural network) and often also on the subtype. The possibilities quickly mount up even with the subtypes: a neural network, for example, has infinite possibilities to shape itself. Just think of the number of possible neurons and layers and the types of connections between them, or the way they exchange information. Other algorithm types also have a lot of parameters.
The right parameter configuration is thus just as crucial for the accuracy and performance of your application for AI, Machine Learning or Deep Learning.
