There is randomness used in the user guide, for example in Meet the Data Objects:
>>> behavior = RegularTimeSeries(
... sampling_rate=100.0, # in Hz
... hand_vel=np.random.randn(1000, 2),
... eye_pos=np.random.randn(1000, 2),
... pupil_size=np.random.randn(1000),
... )
Occasionally, the output of a random process is shown, e.g.
>>> sliced.pupil_size
array([-0.07094018, 1.1442879 , 1.26022563, 1.57259098, ..., ])
Should we consider adding a seeding step for code that uses randomness (either with a global np.random.seed(10) or by instantiating an RNG with rng = np.random.RandomState(10)) ? My reasoning is that this can be useful particularly for more beginner users who may feel more confident if they see the same outputs when they run the code locally as what is shown in the guide. (Keeping in mind that seeding potentially different numpy versions on different setups does not guarantee reproducibility.)
There is randomness used in the user guide, for example in Meet the Data Objects:
Occasionally, the output of a random process is shown, e.g.
Should we consider adding a seeding step for code that uses randomness (either with a global
np.random.seed(10)or by instantiating an RNG withrng = np.random.RandomState(10)) ? My reasoning is that this can be useful particularly for more beginner users who may feel more confident if they see the same outputs when they run the code locally as what is shown in the guide. (Keeping in mind that seeding potentially different numpy versions on different setups does not guarantee reproducibility.)