The csv file can only contain numeric values. ELF has a build-in routine to read csv datasets for regression and classification problems. Following files must exist in the data directory (CSV/DataFiles):
- trainTargetColumn=value (marks the target column, beginning from 0)
- train.txt (name defined in settings.txt)
- test.txt (optional, name defined in settings.txt)
The column of the target is specified in the settings.txt file with the option trainTargetColumn. The target column of the test set can be omitted when it is the last. For example the IRIS dataset has a csv format, here is a preconfigured directory for the ELF.