The current model generator uses the same random number generator to determine the number of elements (random.nextInt() calls) and also the injection of errors (nextRandom() calls). As a consequence, if a single call is inserted or a pair of calls is reordered (as in https://github.com/FTSRG/trainbenchmark/commit/12a9dab9cc4cd9ec01ffa36917ff5985ce342787), we get significantly different models. This makes adding new queries difficult, so we should use separate RNGs where possible.
The current model generator uses the same random number generator to determine the number of elements (
random.nextInt()calls) and also the injection of errors (nextRandom()calls). As a consequence, if a single call is inserted or a pair of calls is reordered (as in https://github.com/FTSRG/trainbenchmark/commit/12a9dab9cc4cd9ec01ffa36917ff5985ce342787), we get significantly different models. This makes adding new queries difficult, so we should use separate RNGs where possible.Related to #101