MclassSplitDataset

This function splits the source dataset context into the first and second destination dataset contexts, according to a percentage. This is typically done to split all of your data (source dataset context) into the training dataset context (first destination) and the development dataset context (second destination) for MclassTrain(). All datasets (source and destinations) must be for the same type of classifier; that is, you must have allocated all of them with either M_DATASET_IMAGES or M_DATASET_FEATURES.

You can consider the split a type of copy of dataset entries from the source to the first or second destination (all source dataset entries are copied). This function does not alter the source dataset context. Any entries currently in the destination dataset contexts are overwritten. Source entries are only copied to one destination.

MclassSplitDataset() ensures that the proper portion of entries for each class in the source dataset context are in both the first and second destination dataset contexts.

To establish the first and second destination dataset contexts, source entries are randomly selected (split). By default, a different random selection of entries is chosen each time you call this function. To select the same random entries, use M_SPLIT_CONTEXT_FIXED_SEED.

If the source dataset context contains augmented entries, they can end up in both the first and second destination dataset contexts. In this case, you must not use one of these destinations as your development dataset context or testing dataset context, since only the training dataset context can contain augmented entries. To inquire whether the source dataset context has augmented entries, use M_NUMBER_OF_AUGMENTED_ENTRIES.

An augmented entry is never split from the source entry with which it was augmented. This can cause the percentage of entries in the first and second destination dataset contexts to differ from the specified percentage.