Passive Learning from Execution Traces
Download and Description
The passive data sets for the challenge can be found individually on the download page for the Black-Box problems and in this tarball.
In this mode of the the competition the participants will be given valid execution traces as input/output sequences. The goal if to passively learn a model for these valid traces in an unsupervised manner (we do not provide invalid traces). There will be two objectives:
- Traditional evaluation of passive learning based on classification accuracy on an unseen test set.
- Answering logical queries on the language from which the traces are sampled, similar to those in the other competition modes.
Training Data
The format of the training data is a standard format used in earlier state machine learning competitions (Abbadingo, Gowachin, Stamina, Pautomac):
total_number_of_traces number_of_input_symbols length_of_the_trace sequence_of_input/output ...
The following is a concrete example of such a training data:
4 6 1 E/Y 4 E/Y A/Z F/V A/Y 2 E/Y A/Z 2 F/V A/U
The first line (4 6 ) means there are 4 traces in total and there are 6 input symbols. The second line (1 E/Y) is the first trace of length 1 (the minimum length). The third line (4 E/Y A/Z F/V A/Y) is the second trace of length 4 and so on.
This data has been generated from a probability distribution on a Mealy machine model and therefore there will be repetitions of the same traces.
Test Data
The format of the test data is the following:
total_number_of_traces length_of_the_trace sequence_of_input/output ...
The following is a concrete example of such a training data:
3 2 E/Y A/Z 3 E/Y E/W A/Z 1 B/Z
Submission format:
The submission has to be done in a text file having 0 or 1 in each line. 1 indicates a valid trace and 0 indicates an invalid trace. There has to be exactly as many lines in the submission file as the number of test execution is the test data. For example, for the test data above the submission file will look as the following:
1 1 0