English (English) - Change language

Testing your classifier

A classifier's quality depends on the size of the training batch. For classifier training, it is recommended to use all available images in a batch.

However, this makes it impossible to evaluate the quality of a classifier outside of its training documents. The Benchmark tool was created for classifier testing. It lets you split up the documents into training and test documents using a percentage ratio.

This tool is used only for unbiased evaluations of the classifier quality. To create the classifier itself, all available images should be used during training.

To split the batch into test and training documents, go to the Batch menu and open the Split Batch window. Select the appropriate ratio for splitting the source batch (60%-40% by default). Specify the minimum number of documents of each class in the resulting new batches. By default, each batch should have at least 1 document of each class. Specify names for the batches and click Split.

To use batches created with Split for training and testing, select the Batches window in the Batch menu. Select Benchmark. In the drop-down list, specify the batch that will be used for training. Use check marks to specify the batches that will be used for testing a trained classifier.

Clicking Run will launch the training, after which the classification will then immediately start.

Click Statistics in the Batches window for an analysis of the test batch classification results.

25.09.2020 9:24:45


Please leave your feedback about this article