AutoML
Model selection, approach to automate processes such as hyperparameter adjustment.
Automatic machine learning (AutoML) is a set of approaches and tools that generally require deep expertise and time to develop a machine learning model -data pretreatment, attribute engineering, model architecture selection, hyperparameter adjustment and model evaluation- greatly automated. Traditionally, a data scientist had to decide on how many layers will be used, what algorithm (milk tree, nervous network) is appropriate for a problem, what will be the learning rate, and by hand to set dozens of more, through trial illusion; this process was connected to both time-consuming and greatly experience. AutoML systems scan this search space systematically and automatically with methods such as grid search, random search, Bayesçi optimization or evolutional algorithms.
The tools such as Google Cloud AutoML, H2O.ai and open-source Auto-sklearn can offer hundreds of model configurations in the background, only by specifying the data set and goal. The biggest value of this approach is the democraticization effect: companies without the expertise of deep machine learning, and developers can also improve predictive models of quality on their data thanks to AutoML. However, AutoML is often not kept completely in research-level problems requiring very special or creative architecture design, the location of the fine settings a experienced specialist is hand-held, so much standard is positioned as a practical accelerator for well-defined problems.
The tools such as Google Cloud AutoML, H2O.ai and open-source Auto-sklearn can offer hundreds of model configurations in the background, only by specifying the data set and goal. The biggest value of this approach is the democraticization effect: companies without the expertise of deep machine learning, and developers can also improve predictive models of quality on their data thanks to AutoML. However, AutoML is often not kept completely in research-level problems requiring very special or creative architecture design, the location of the fine settings a experienced specialist is hand-held, so much standard is positioned as a practical accelerator for well-defined problems.