Selection of machine learning algorithms
Systems and methods of selecting machine learning models/algorithms for a candidate dataset are disclosed. A computer system may access historical data of a set of algorithms applied to a set of benchmark datasets; select a first algorithm of the set of algorithms; apply the first algorithm to an input dataset to create a model of the input dataset; evaluate and store results of the applying; and add the first algorithm to a set of tried algorithms. The computer system may select a next algorithm of the algorithm set via submodular optimization based on the historical data and the set of tried algorithms; apply the next algorithm to the input dataset; capture a next result based on the applying; add the next result to update the set of tried algorithms; and repeat the submodular optimization. The procedure may continue until a termination condition is reached.
1. A computing system comprising:
a prediction server to:
obtain an input dataset;
select a first algorithm from a set of algorithms;
apply the first algorithm to the input dataset to create a first model of the input dataset;
evaluate and control storage of results of applying the first algorithm to the input dataset, by:
controlling performance of a holdout procedure using the set of predictions; or
controlling performance of a two-fold cross-validation procedure using the set of predictions;
add the first algorithm into a set of tried algorithms;
for each algorithm in the set of algorithms other than the first algorithm and until a termination condition is met:
generate a submodular function, the submodular function configured to determine a likelihood that an optimum algorithm has already been tried;
select, via the submodular function, a next algorithm of the set of algorithms to apply to the input dataset based on accessed historical performance data in a benchmark database, and the set of tried algorithms,
evaluate and control storage of results of applying the next algorithm to the input dataset,
add the next algorithm to the set of tried algorithms; and
identify an optimum algorithm from the set of algorithms, wherein the optimum algorithm is an individual algorithm of the set of algorithms that is closer to fulfilling a predetermined criterion than other algorithms of the set of algorithms;
generate a report indicating the optimum algorithm; and
an application server to implement a user interface, the application server to obtain the report, and transmit the report to a user system.
2. The computing system of claim 1 , further comprising:
a model server to obtain a second set of algorithms from a modeling algorithms database, obtain a set of benchmark datasets from a benchmark dataset database, apply individual algorithms of the second set of algorithms to individual benchmark datasets of the set of benchmark datasets to obtain benchmark predictions for the individual algorithms, evaluate the benchmark predictions to obtain benchmark results for the individual algorithms, and store the benchmark results in a benchmark database.
3. The computing system of claim 2 , wherein the prediction server is further configured to generate a submodular function based on the stored first results and select the next algorithm using the submodular function.