Automated Machine Learning

Much of the work undertaken in a data science programme can be automated to improve accuracy and reliability as well reducing the amount of time and money necessary to complete an operation.

Exploratory data analysis

Visualising data before modelling can offer insights which can drive what questions to ask of machine learning, helping to ensure that effort is focused where it will deliver the most value. Automating tasks such as plotting all variables against the target feature can save lots of time.

Feature transformations

The expanse of choices available, and decisions that need to be made, when shaping data for machine learning can be intimidating: imputing missing values, encoding categorical variables, deriving information from dates. Many of these feature transformations are canonical and can be applied reliably through automation.

Algorithm selection

With hundreds of algorithms and modelling approaches at your disposal, selecting those that work best for your problem and dataset can consume a significant or overwhelming proportion of your data science resources.

Hyper-parameter tuning

The right algorithm is the wrong algorithm if it has not been set up with the optimal values. Pushing and pulling the levers to the right settings and testing the results is precisely the kind of operation where automation reaps dividends.

Model diagnostics

Evaluating the success and efficacy of machine learning models, using ROC curves, feature importance, learning curves and other diagnostics can be generated automatically.

Request a demo